AUTOMATION HISTORY

Language is messy, subtle, and full of meaning that shifts with context. Teaching machines to truly understand it is one of the hardest problems in artificial intelligence. That challenge is what natural language understanding (NLU) sets out to solve. From voice assistants that follow instructions to support systems that interpret user intent, NLU sits at the core of many real-world AI applications. Most systems today are trained using labeled data and supervised techniques. But there's growing interest in something more adaptive: deep reinforcement learning (DRL). Instead of learning from fixed examples, DRL allows a model to improve through trial, error, and feedback, much like a person learning through experience. This article looks at where DRL fits into the modern NLU landscape. We'll explore how it's being used to fine-tune responses, guide conversation flow, and align models with human values. Overview of Deep Reinforcement Learning What is Natural Language Understanding (NLU)? Challenges in NLU and How to Address Them Where DRL Adds Value in NLU Modern Architectures in NLU from BERT to Claude The Niche Role of DRL in Modern NLU Reinforcement Learning from Human Feedback (RLHF) Ecosystem and Tools for DRL in NLP Hands-On Demo: Simulating DRL Feedback in NLU Case Studies of DRL in NLU Wrapping Up Reinforcement learning is a subfield of machine learning. It’s inspired by behavioral psychology, in which agents learn to maximize cumulative rewards by performing behaviors in a given environment. Traditionally, reinforcement learning techniques have been used to solve simple problems with discrete state and action spaces. But the development of deep learning has opened the door to applying these techniques to more complicated, high-dimensional environments, like computer vision, natural language processing (NLP), and robotics. DRL uses deep neural networks to approximate complex functions that translate observations into actions, allowing agents to learn from raw sensory data. Deep neural networks, which represent knowledge in numerous layers of abstraction, may catch detailed patterns and relationships in data, allowing for more effective decision-making. Imagine you’re playing a video game where you’re controlling a character, and your goal is to get the highest score possible. Now, when you first start playing, you might not know the best way to play, right? You might try different things like jumping, running, or shooting, and you see what works and what doesn’t. We can think of DRL as a technique that enables computers or robots to learn how to play video games as time goes on. DRL involves a computer learning from its environment, learning from its experiences and mistakes. The computer, like the player, tries different actions and receives feedback based on its performance. If it performs well, it gets rewards, while if it fails, it gets a penalty. The computer’s job is to figure out the best possible actions to take in different situations to maximize rewards. Instead of learning from trial and error, DRL uses deep neural networks, which are like super-smart brains that can understand vast amounts of data and patterns. These neural networks help the computer make better decisions in the future, and over time, it can become even better at playing the game – sometimes even better than humans. Image Source NLU is a subfield of artificial intelligence (AI), and its aim is to help computers understand, interpret, and respond to human language in meaningful ways. It involves creating algorithms and models that can process and analyze text to extract meaningful information, determine the intent behind it, and provide appropriate replies. NLU is a basic part of many AI applications, such as chatbots, virtual assistants, and personalized recommendation systems, which require the ability to interpret and respond to human language. Its key components include: Text processing: NLU systems must be able to process and interpret text, which includes tokenization (cutting it down into words or phrases), part-of-speech tagging, and named entity recognition. Sentiment analysis: Identifying the sentiment communicated in a piece of text (positive, negative, or neutral) is a common task in NLU. Intent recognition: Identifying the goal or objective of a user’s input, such as buying a flight or requesting weather forecasts. Language generation: (technically part of Natural Language Generation, or NLG): While NLU focuses on understanding text, NLG is about producing coherent, contextually appropriate text. Many AI systems combine both, first interpreting the input through NLU, then generating an appropriate response using NLG. Entity extraction: Identifying and categorizing essential details in the text, such as dates, locations, and people. NLU aims to help machines interpret, understand, and respond to human language in ways that make sense. While it has made great progress, there are still challenges that limit how well it works in practice. Below are some of these challenges and how Deep Reinforcement Learning (DRL) can play a supportive role. DRL is not a replacement for large-scale pretraining or instruction tuning, but it can complement them by helping models adapt through interaction and feedback. Naturally, words can have more than one meaning, and a single sentence or phrase might be understood in different ways. This makes it hard for NLU systems to always pinpoint what the speaker or writer intends. DRL can help reduce ambiguity by allowing models to learn from feedback. If a certain interpretation gets positive results, the model can prioritize it. If not, it can try a different approach. While this does not remove ambiguity entirely, it can improve a model’s ability to make better choices over time, especially when combined with a strong pretrained foundation. Understanding language often depends on context such as cultural references, sarcasm, or the tone behind certain words. These are straightforward for people but challenging for machines to recognize. By learning from interaction signals such as whether a user is satisfied with a response, DRL can help a model adapt to context more effectively. However, the core ability to understand context still comes from large-scale pretraining. DRL mainly fine-tunes and adjusts this behavior during use. Human language comes in many forms including different dialects, slang, colloquialisms, and regional expressions. This variety can challenge NLU systems that have not seen enough examples of these patterns during training. With DRL, models can adapt to new language styles when exposed to them repeatedly in real-world use. This makes them more flexible and responsive, although their base understanding still relies on the diversity of the data used during pretraining. As text data continues to grow, NLU systems must be able to process large volumes quickly and efficiently, especially in real-time applications such as chatbots and virtual assistants. DRL can contribute by helping models optimize certain processing steps through trial and feedback. While it will not replace architectural or infrastructure improvements, it can help fine-tune performance for specific high-traffic tasks. Training advanced NLU models is resource-intensive, which can be a challenge for mobile devices, edge computing, or other resource-limited environments. DRL can make the learning process more efficient by reusing past experiences through techniques such as off-policy learning and reward modeling. Combined with smaller, distilled model architectures, this can make it easier to deploy capable NLU systems even with limited computing power. DRL is not a primary training method for most NLU models. Its main value comes when interaction, feedback, or rewards can be used to improve how a system behaves after it has already been pretrained. When applied selectively, DRL can help refine and personalize model performance in ways that matter for specific use cases. Here are some areas where DRL has shown potential: Dialogue systems DRL can help chatbots and virtual assistants manage conversations more smoothly. It can be used to refine turn-taking, handle vague questions in a better way, or adjust responses to improve user satisfaction during longer conversations. Text summarization Most summarization models rely on supervised learning. DRL can be added as a fine-tuning step to focus on factors such as relevance or fluency, especially when custom reward signals are linked to specific goals or user preferences. Response generation and language modeling DRL can guide language generation toward outputs that are more useful, aligned with user intent, or better suited to certain tone and safety requirements. Reward-based optimization in parsing or classification In certain cases, DRL has been used to improve outputs based on downstream objectives such as increasing label confidence or enhancing the quality of supporting explanations, alongside accuracy. Interactive machine translation DRL can help translation systems adapt over time by learning from reinforcement signals like human corrections or post-editing feedback, leading to gradual improvements in quality. In short, DRL works best as a targeted enhancement. It is not used to build general-purpose NLU systems from scratch, but it can make existing systems more adaptable, aligned, and responsive when feedback loops are part of the application. Early NLU systems used Recurrent Neural Networks (RNNs) and Convolutional Neural Networks (CNNs), but most modern systems use transformers. These models use a mechanism called self-attention to capture long-range dependencies. Self-attention allows each word to “attend” to every other word in the input, assigning weights that determine relevance for understanding the current word. Long-range dependencies occur when the meaning of one word depends on another far away in the text (like linking “he” to “the president” from earlier sentences). This helps maintain context over large spans of text. Here’s how the main types of transformer models are used today: Examples: BERT, RoBERTa, ALBERT, DeBERTa These models process text input and create rich contextual representations without generating new text. They are excellent for classification, entity extraction, and tasks that require understanding rather than producing language. The encoder reads the whole input and encodes it into a vector representation, which is then used by a task-specific head for predictions. They're often fine-tuned for specific tasks and perform especially well in structured language understanding. Examples: T5, FLAN-T5 These models have two components: an encoder that reads and encodes the input text, and a decoder that generates an output sequence based on that encoded representation. They are ideal for sequence-to-sequence tasks such as summarization, translation, and instruction following. The encoder captures the meaning of the input, while the decoder produces coherent output in the target form. They’re flexible and particularly useful in multi-task learning setups Examples: GPT-4, Claude 3, Gemini These models generate text one token at a time, predicting the next token based on all previous tokens in the sequence. They excel in open-ended text generation, creative writing, and reasoning tasks. Because they are trained to predict the next word given any context, they can perform many tasks simply by being prompted, without additional training. They’re typically aligned with human preferences using techniques like Reinforcement Learning from Human Feedback (RHLF). These models are now widely used in real-world applications, such as chatbots, enterprise tools, and multilingual digital assistants, and many can handle new tasks with just a prompt, requiring no additional training. DRL is not a general-purpose solution for most NLU challenges, such as handling ambiguity or understanding context. These problems are typically addressed using large-scale pretraining and supervised or instruction-based fine-tuning. That said, DRL still plays a valuable role in specific areas where feedback and long-term optimization are useful. It is commonly applied in: Improving dialogue strategy: DRL helps conversational agents manage turn-taking, adjust tone, and adapt to user preferences across multiple interactions. Aligning model behavior using RLHF: Reinforcement learning from human feedback (RLHF – more on this below) uses DRL to train models that respond in ways people find more helpful, safe, or contextually appropriate. Reward modeling for alignment and safety: DRL enables the training of reward models that guide language systems toward ethical, culturally aware, or domain-specific behavior. Looking ahead, DRL is likely to grow in importance for applications that involve real-time interaction, long-horizon reasoning, or agent-driven workflows. For now, it serves as a targeted enhancement alongside more widely used training methods. Let’s talk a bit more about RLHF, as it’s pretty important here. It’s also currently the primary way DRL is applied in large-scale language models such as GPT‑4, Claude, and Gemini. It works in three main steps: Reward model training – Human annotators rank model outputs for the same prompt. These rankings are used to train a reward model that scores outputs based on how helpful, safe, or relevant they are. Policy optimization – Using algorithms such as PPO (Proximal Policy Optimization), the base language model is fine-tuned to maximize the reward model’s score. Iteration and safety – RLHF loops are often combined with safety-focused reward modeling, constitutional AI (following explicit guidelines for safe behavior), refusal strategies for harmful requests, and red‑teaming to probe weaknesses. Data‑efficient variants are increasingly common, such as offline RL, replay buffers, and leveraging implicit feedback like click‑through logs. In practice, RLHF has significantly improved the ability of models to follow instructions, avoid harmful outputs, and align with human values. If you're looking to explore DRL in NLU, you don't have to start from scratch. There’s a solid ecosystem of tools that make it easier to test ideas, build prototypes, and fine-tune models using rewards and feedback. Here are a few go-to libraries: trl by Hugging Face: A lightweight framework built specifically for applying reinforcement learning to transformer models. It's widely used for RLHF, reward modeling, and steering model outputs based on human preferences. Stable-Baselines3: A simple, well-documented library for classic DRL algorithms like PPO, A2C, and DQN. It’s great for testing DRL setups in smaller or custom environments. RLlib (part of Ray): Designed for scaling up. If you're working on distributed training or combining DRL with larger pipelines, RLlib helps manage the complexity. These libraries pair well with open-source large language models like LLaMA, Mistral, Gemma, and Command R+. Together, they give you everything you need to experiment with DRL-backed language systems, whether you're tuning responses in a chatbot or building a reward model for alignment. You don’t need a full reinforcement learning pipeline to understand reward signals. This notebook demonstrates how you can simulate preference-based feedback using GPT-3.5. Users interact with the model, provide binary feedback (good or bad), and the system logs each interaction with a corresponding reward. It mirrors the principles behind techniques like RLHF. First, you’ll need to install the required packages and set up your API key. What this does: Installs and loads required libraries Reads your OpenAI key from an environment variable or prompts for it interactively Now, try sending a prompt and seeing what response you get: What this does: Uses OpenAI’s GPT-3.5 to generate a response Handles errors if the API call fails You can now track user responses and simulated reward signals like this: This code initializes a list to store logs of each interaction. Now you can capture the prompt, display the response, and accept feedback. What this does: Shows GPT-3.5’s response to the user’s prompt Displays feedback buttons Logs reward and shows feedback history You can also visualize reward trends: This plots the user’s reward signals over time to simulate policy shaping. You can also allow users to type and submit prompts. This sets up a simple form to collect prompts and connects the generate button to the main interaction logic. This gives the output: This demo captures the fundamentals of preference-based learning using GPT-3.5. It doesn’t update model weights but shows how feedback can be structured as a reward signal. This is the foundation of reinforcement learning in modern LLM pipelines. Note: This demo only logs feedback. In true RLHF, a second phase fine-tunes the model weights based on it. A real-world example of this is InstructGPT. This is a version of OpenAI’s GPT models that’s trained to follow instructions written by people. Instead of just predicting the next word, it tries to really figure out and then do what you’ve asked, the way you asked it. Despite being over 100× smaller than GPT-3, InstructGPT was preferred by humans in 85% of blind comparisons. And one of the key reasons was that is uses RLHF. This made it safer, more truthful, and better at following complex instructions, showing how reward signals like the one simulated here can greatly improve real-world model performance. While DRL is not the default approach for most NLU tasks, it has shown promising results in targeted use cases, especially where learning from interaction or adapting over time adds value. Below are a few examples that illustrate how DRL can enhance language understanding in practice: A global e-commerce platform partnered with Welocalize to launch a DRL-powered multilingual NLU system capable of interpreting customer intent across 30+ languages and domains. This system used reinforcement learning to adapt to cultural nuances and refine predictions through user interaction. Over 13 million high-quality utterances delivered for culturally adaptive, accurate customer support and product recommendations. Researchers introduced a framework called RLLR (Reinforcement Learning with Label-Sensitive Reward) to improve NLU tasks like sentiment classification, topic labeling, and intent detection. By incorporating label-sensitive reward signals and optimizing via Proximal Policy Optimization (PPO), the model aligned its predictions with both rationale quality and true label accuracy. These examples show how DRL, when paired with specific feedback signals or interactive goals, can be a useful layer on top of traditional NLU systems. Though still niche, the approach continues to evolve through research and industry experimentation. The integration of DRL with NLU has shown promising results in niche but growing areas. Adaptive learning through various interactions and feedback allows DRL to enhance NLU models’ ability to handle ambiguity, context, and linguistic differences. As research progresses, the link between DRL and NLU is expected to drive advancements in AI-powered language applications, making them more efficient, scalable, and context-aware. I hope this was helpful! Python developer Studying Mechanical Engineering Machine Learning enthusiast If this article was helpful, share it. Learn to code for free. freeCodeCamp's open source curriculum has helped more than 40,000 people get jobs as developers. Get started freeCodeCamp is a donor-supported tax-exempt 501(c)(3) charity organization (United States Federal Tax Identification Number: 82-0779546) Our mission: to help people learn to code for free. We accomplish this by creating thousands of videos, articles, and interactive coding lessons - all freely available to the public. Donations to freeCodeCamp go toward our education initiatives, and help pay for servers, services, and staff. You can make a tax-deductible donation here.

Advertisement You have full access to this open access article 278 Accesses Explore all metrics We are providing an unedited version of this manuscript to give early access to its findings. Before final publication, the manuscript will undergo further editing. Please note there may be errors present which affect the content, and all legal disclaimers apply. Reinforcement learning (RL), as an emerging interdisciplinary field formed by the integration of artificial intelligence and control science, is currently demonstrating a cross-disciplinary development trend led by artificial intelligence and has become a research hotspot in the field of optimal control. This paper systematically reviews the development context of RL, focusing on the intrinsic connection between single-agent reinforcement learning (SARL) and multi-agent reinforcement learning (MARL). Firstly, starting from the formation and development of RL, it elaborates on the similarities and differences between RL and other learning paradigms in machine learning, and briefly introduces the main branches of current RL. Then, with the basic knowledge and core ideas of SARL as the basic framework, and expanding to multi-agent system (MAS) collaborative control, it explores the coherence characteristics of the two in theoretical frameworks and algorithm design. On this basis, this paper reconfigures SARL algorithms into dynamic programming, value function decomposition and policy gradient (PG) type, and abstracts MARL algorithms into four paradigms: behavior analysis, centralized learning, communication learning and collaborative learning, thus establishing an algorithm mapping relationship from single-agent to multi-agent scenarios. This innovative framework provides a new perspective for understanding the evolutionary correlation of the two methods, and also discusses the challenges and solution ideas of MARL in solving large-scale MAS problems. This paper aims to provide a reference for researchers in this field, and to promote the development of cooperative control and optimization methods for MAS as well as the advancement of related application research. Avoid common mistakes on your manuscript. No datasets were generated or analysed during the current study. Agrawal R, Imieliński T, Swami A (1993) Mining association rules between sets of items in large databases. In: Proceedings of the 1993 ACM SIGMOD international conference on Management of data, pp 207–216 Ahad A, Tahir M, Sheikh MAS, et al (2021) Optimal route selection in 5G-based smart health-care network: a reinforcement learning approach. In: 2021 26th IEEE Asia-Pacific conference on communications (APCC), pp 248–253 Alelaiwi A (2020) Resource allocation management in patient-to-physician communications based on deep reinforcement learning in smart healthcare services. In: 2020 IEEE International conference on multimedia & expo workshops (ICMEW), pp 1–5 Anderson JR (1983) Machine learning: an artificial intelligence approach, vol 2. Morgan Kaufmann, Burlington Google Scholar Arthur D, Vassilvitskii S (2007) K-means++: the advantages of careful seeding. In: Proceedings of the eighteenth annual ACM-SIAM symposium on discrete algorithms, vol 1, pp 1027–1035 Asha Rani GS, Lal Priya PS, Jayan J et al (2024) Data-driven energy management of an electric vehicle charging station using deep reinforcement learning. IEEE Access 12:65956–65966 Google Scholar Bai H, Shen R, Lin Y et al (2024) Lamarckian platform: pushing the boundaries of evolutionary reinforcement learning toward asynchronous commercial games. IEEE Trans Games 16(1):51–63 Google Scholar Bellman R (1957a) A Markovian decision process. J Math Mech 6(5):679–684 Google Scholar Bellman RE (1957b) Dynamic programming. Princeton University Press, Princeton Google Scholar Bertsekas DP (2011) Approximate policy iteration: a survey and some new methods. J Control Theory Appl 9(3):310–335 Google Scholar Bertsekas D (2012) Dynamic programming and optimal control, vol 4. Athena scientific, Nashua Google Scholar Breiman L (2001) Random forests. Mach Learn 45(1):5–32 Google Scholar Cao H, Xiong H, Zeng W et al (2024) Safe reinforcement learning-based motion planning for functional mobile robots suffering uncontrollable mobile robots. IEEE Trans Intell Transp Syst 25(5):4346–4363 Google Scholar Castañeda AO (2016) Deep reinforcement learning variants of multi-agent learning algorithms. PhD thesis, University of Edinburgh, Edinburgh Chapelle O, Scholkopf B, Zien A (2009) Semi-supervised learning (chapelle, O. et al., eds.; 2006) [book reviews]. IEEE Trans Neural Netw 20(3):542–542 Google Scholar Chen H, Luo H, Huang B et al (2023) Transfer learning-motivated intelligent fault diagnosis designs: a survey, insights, and perspectives. IEEE Trans Neural Netw Learning Syst 35(3):2969–2983 Google Scholar Chen W, Yang S, Li W et al (2024) Learning multi-intersection traffic signal control via coevolutionary multi-agent reinforcement learning. IEEE Trans Intell Transp Syst 25(11):15947–15963 Google Scholar Chen J, Zhu B, Zhang M et al (2025) Multi-agent deep reinforcement learning cooperative control model for autonomous vehicle merging into platoon in highway. World Electr Veh J 16(4):225 Google Scholar Diprasetya MR, Pullani AN, Schwung A (2024) Sim-to-real transfer for robotics using model-free curriculum reinforcement learning. In: 2024 IEEE international conference on industrial technology (ICIT), pp 1–6 Drudi C, Fechner M, Mollura M et al (2024) Reinforcement learning for heart failure treatment optimization in the intensive care unit. In: 2024 46th annual international conference of the IEEE Engineering in Medicine and Biology Society (EMBC), pp 1–4 Foerster JN, Assael YM, de Freitas N et al (2016) Learning to communicate with deep multi-agent reinforcement learning. In: Proceedings of the 30th international conference on neural information processing systems, Barcelona, pp 2145–2153 Foerster J, Farquhar G, Afouras T et al (2017) Counterfactual multi-agent policy gradients. arXiv:1705.08926 Fukushima K, Miyake S, Ito T (1983) Neocognitron: a neural network model for a mechanism of visual pattern recognition. IEEE Trans Syst Man Cybern 13(5):826–834 Google Scholar Gauss CF (1809) Theoria Motus Corporum Coelestium in Sectionibus Conicis Solem Ambientium. Perthes et Besser, Hamburg Google Scholar Geranmayeh P, Grass E (2025) Optimization of beamforming and transmit power using DGN and comparison with traditional techniques. IEEE Access 13:94275–94285 Google Scholar Geurts P, Irrthum A, Wehenkel L (2009) Supervised learning with decision tree-based methods in computational and systems biology. Mol BioSyst 5(12):1593–1605 Google Scholar Gong J, Yu N, Han F et al (2024) Energy scheduling optimization for microgrids based on partially observable Markov game. IEEE Trans Artif Intell 5(11):5371–5380 Google Scholar Guan H (2020) Analysis on deep reinforcement learning in industrial robotic arm. In: 2020 international conference on intelligent computing and human–computer interaction (ICHCI), pp 426–430 Guan H, Gao Y, Zhao M et al (2022) Ab-mapper: attention and bicnet based multi-agent path planning for dynamic environment. In: 2022 IEEE/RSJ international conference on intelligent robots and systems (IROS), IEEE, pp 13799–13806 Gui J, Chen T, Zhang J et al (2024) A survey on self-supervised learning: algorithms, applications, and future trends. IEEE Trans Pattern Anal Mach Intell 46(12):9052–9071 Google Scholar Han J, Pei J, Yin Y (2000) Mining frequent patterns without candidate generation. ACM SIGMOD Rec 29(2):1–12 Google Scholar Hausknecht MJ, Stone P (2015) Deep recurrent Q-learning for partially observable MDPs. In: AAAI fall symposia, p 141 Hosmer DW, Lemeshow S (2000) Applied logistic regression. Wiley, Hoboken Google Scholar Hou Y, Zhao J, Zhang R et al (2024) UAV swarm cooperative target search: a multi-agent reinforcement learning approach. IEEE Trans Intell Veh 9(1):568–578 Google Scholar Hu B, Gong Y, Liang X et al (2024) Safe deep reinforcement learning-based real-time multi-energy management in combined heat and power microgrids. IEEE Access 12:193581–193593 Google Scholar Huang Z, van der Aalst W, Lu X et al (2011) Reinforcement learning based resource allocation in business process management. Data Knowl Eng 70(1):127–145 Google Scholar Iqbal S, Sha F (2018) Actor-attention-critic for multi-agent reinforcement learning. arXiv:1810.02912 Iskandar A, Rostum HM, Kovács B (2023) Using deep reinforcement learning to solve a navigation problem for a swarm robotics system. In: 2023 24th international Carpathian control conference (ICCC), pp 185–189 Isla-Cernadas D, Fernández-Delgado M, Cernadas E et al (2025) Closed-form gaussian spread estimation for small and large support vector classification. IEEE Trans Neural Netw Learn Syst 36(3):4336–4344 Google Scholar Jiang Q, Qin M, Zhang H et al (2024) Blindlight: high robustness reinforcement learning method to solve partially blinded traffic signal control problem. IEEE Trans Intell Transp Syst 25(11):16625–16641 Google Scholar Jiang Y, Di K, Qian R et al (2025) Optimizing risk-aware task migration algorithm among multiplex UAV groups through hybrid attention multi-agent reinforcement learning. Tsinghua Sci Technol 30(1):318–330 Google Scholar Jordan MI, Kearns MJ, Solla SA (1998) Advances in neural information processing systems. In: proceedings of the 1997 conference, vol 10. MIT Press Kaloev M, Krastev G (2023) Comprehensive review of benefits from the use of neuron connection pruning techniques during the training process of artificial neural networks in reinforcement learning: experimental simulations in atari games. In: 2023 7th International symposium on multidisciplinary studies and innovative technologies (ISMSIT), pp 1–6 Karimi S, Asadi S, Payberah AH (2024) Bazigooshi: a hybrid model of reinforcement learning for generalization in gameplay. IEEE Trans Games 16(3):722–734 Google Scholar Kell AJM, Forshaw M, Stephen McGough A (2020) Exploring market power using deep reinforcement learning for intelligent bidding strategies. In: 2020 IEEE international conference on big data (big data), pp 4402–4411 Kim S (2023) Learning and game based spectrum allocation model for internet of medical things (IOMT) platform. IEEE Access 11:48059–48068 Google Scholar Kim D, Moon S, Hostallero D et al (2019) Learning to schedule communication in multi-agent reinforcement learning. arXiv:1902.01554 Kiran BR, Sobh I, Talpaert V et al (2022) Deep reinforcement learning for autonomous driving: a survey. IEEE Trans Intell Transpl Syst 23(6):4909–4926 Google Scholar Kobren A, Monath N, Krishnamurthy A et al (2017) A hierarchical algorithm for extreme clustering. In: Proceedings of the 23rd ACM SIGKDD international conference on knowledge discovery and data mining, pp 255–264 Lagoudakis MG, Parr R (2003) Least-squares policy iteration. J Mach Learn Res 4(Dec):1107–1149 Google Scholar Lance GN, Williams WT (1967) A general theory of classificatory sorting strategies: 1. Hierarchical systems. Comput J 9(4):373–380 Google Scholar Li M, Zhou ZH (2007) Improve computer-aided diagnosis with machine learning techniques using undiagnosed samples. IEEE Trans Syst Man Cybern-Part A 37(6):1088–1098 Google Scholar Li K, Hu Q, Liu Q et al (2024a) A predefined-time consensus algorithm of multi-agent system for distributed constrained optimization. IEEE Trans Netw Sci Eng 11(1):957–968 Google Scholar Li X, Yang Y, Yuan Z et al (2024b) Progression cognition reinforcement learning with prioritized experience for multi-vehicle pursuit. IEEE Trans Intell Transp Syst 25(8):10035–10048 Google Scholar Lillicrap TP, Hunt JJ, Pritzel A et al (2015) Continuous control with deep reinforcement learning. arXiv:1509.02971 Lin M, Zhao B, Liu D (2025) Optimal learning output tracking control: a model-free policy optimization method with convergence analysis. IEEE Trans Neural Netw Learn Syst 36(3):5574–5585 Google Scholar Littman ML (1994) Markov games as a framework for multi-agent reinforcement learning. In: Machine learning proceedings 1994. Elsevier, pp 157–163 Liu Y, Wang Q (2022) Game confrontation of 5v5 multi-agent based on Mappo reinforcement learning algorithm. In: 2022 37th youth academic annual conference of chinese association of automation (YAC), pp 1395–1398 Liu D, Wei Q (2013) Policy iteration adaptive dynamic programming algorithm for discrete-time nonlinear systems. IEEE Trans Neural Netw Learning Syst 25(3):621–634 Google Scholar Liu D, Wei Q (2014) Policy iteration adaptive dynamic programming algorithm for discrete-time nonlinear systems. IEEE Trans Neural Netw Learn Syst 25(3):621–634 Google Scholar Liu D, Wei Q, Wang D et al (2017) Adaptive dynamic programming with applications in optimal control. Springer, Cham Google Scholar Liu D, Xue S, Zhao B et al (2021a) Adaptive dynamic programming for control: a survey and recent advances. IEEE Trans Syst Man Cybern Syst 51(1):142–160 Google Scholar Liu X, Zhang F, Hou Z et al (2021b) Self-supervised learning: generative or contrastive. IEEE Trans Knowl Data Eng 35(1):857–876 Google Scholar Liu D, Dou L, Zhang R et al (2023) Multi-agent reinforcement learning-based coordinated dynamic task allocation for heterogenous UAVs. IEEE Trans Veh Technol 72(4):4372–4383 Google Scholar Liu B, Han W, Wang E et al (2024) An efficient message dissemination scheme for cooperative drivings via cooperative hierarchical attention reinforcement learning. IEEE Trans Mob Comput 23(5):5527–5542 Google Scholar Lloyd SP (1982) Least squares quantization in PCM. IEEE Trans Inf Theory 28(2):129–137 Google Scholar Lu R, Jiang Z, Yang T et al (2024) A novel hybrid-action-based deep reinforcement learning for industrial energy management. IEEE Trans Ind Inf 20(10):12461–12475 Google Scholar Ma Z, Liu X, Huang Y (2024) Unsupervised reinforcement learning for multi-task autonomous driving: expanding skills and cultivating curiosity. IEEE Trans Intell Transpl Syst 25(10):14209–14219 Google Scholar Malathy V, Al-Jawahry HM, GKM et al (2024) A reinforcement learning method in cooperative multi-agent system for production control system. In: 2024 international conference on data science and network security (ICDSNS), pp 1–4 Marden JR, Arslan G, Shamma JS (2009) Cooperative control and potential games. IEEE Trans Syst Man Cybern B 39(6):1393–1407 Google Scholar Markov AA (1906) Extension of the law of large numbers to dependent quantities. Izv Fiz-Matem Obsch Kazan Univ 15(1):135–156 Google Scholar Ministry of Industry and Information Technology of the People’s Republic of China (2024) Notice on issuing the national comprehensive standardization system construction guidelines for the artificial intelligence industry (2024 edition). https://www.miit.gov.cn/zwgk/zcwj/wjfb/tz/art/2024/art_e8ebf5600ec24d3db644150873712c5f.html Minsky M (2007) Steps toward artificial intelligence. Proc IRE 49(1):8–30 Google Scholar Mnih V, Kavukcuoglu K, Silver D et al (2015) Human-level control through deep reinforcement learning. Nature 518(7540):529–533 Google Scholar Mnih V, Kavukcuoglu K, Silver D et al (2013) Playing Atari with deep reinforcement learning. arXiv:1312.5602 Mnih V, Badia AP, Mirza M et al (2016) Asynchronous methods for deep reinforcement learning. In: International conference on machine learning, PmLR, pp 1928–1937 Mocanu E, Mocanu DC, Nguyen PH et al (2019) On-line building energy optimization using deep reinforcement learning. IEEE Trans Smart Grid 10(4):3698–3708 Google Scholar Neftci EO, Averbeck BB (2019) Reinforcement learning in artificial and biological systems. Nat Mach Intell 1(3):133–143 Google Scholar Noaro G, Zhu T, Cappon G et al (2023) A personalized and adaptive insulin bolus calculator based on double deep q-learning to improve type 1 diabetes management. IEEE J Biomed Health Inform 27(5):2536–2544 Google Scholar Park JS, Chen MS, Yu PS (1995) An effective hash-based algorithm for mining association rules. ACM SIGMOD Rec 24(2):175–186 Google Scholar Pearson K (1901) Liii. On lines and planes of closest fit to systems of points in space. Lond Edinburgh Dublin Philos Mag J Sci 2(11):559–572 Google Scholar Peng P, Yuan Q, Wen Y et al (2017) Multiagent bidirectionally-coordinated nets for learning to play Starcraft combat games. arXiv:1703.10069 Prudencio RF, Maximo MR, Colombini EL (2023) A survey on offline reinforcement learning: taxonomy, review, and open problems. IEEE Trans Neural Netw Learn Syst 35(8):10237–57 Google Scholar Qi L, Sun Y, Luan W (2024) Large-scale traffic signal control based on multi-agent q-learning and pressure. IEEE Access 12:1092–1101 Google Scholar Qin C, Qiao X, Wang J et al (2024) Barrier-critic adaptive robust control of nonzero-sum differential games for uncertain nonlinear systems with state constraints. IEEE Trans Syst Man Cybern Syst 54(1):50–63 Google Scholar Qin C, Hou S, Pang M et al (2025a) Reinforcement learning-based secure tracking control for nonlinear interconnected systems: an event-triggered solution approach. Eng Appl Artif Intell 161:112243 Google Scholar Qin C, Jiang K, Wang Y et al (2025b) Event-triggered H∞ control for unknown constrained nonlinear systems with application to robot arm. Appl Math Model 144:116089 Google Scholar Qin C, Ran X, Zhang D (2025c) Unsupervised image stitching based on generative adversarial networks and feature frequency awareness algorithm. Appl Soft Comput 183:113466 Google Scholar Qu G, Wu H, Li R et al (2021) DMRO: a deep meta reinforcement learning-based task offloading framework for edge-cloud computing. IEEE Trans Netw Serv Manage 18(3):3448–3459 Google Scholar Rashid T, Samvelyan M, Schroeder C et al (2018) Qmix: monotonic value function factorisation for deep multi-agent reinforcement learning. In: International conference on machine learning, PMLR, pp 4295–4304 Rashid T, Farquhar G, Peng B et al (2020) Weighted qmix: expanding monotonic value function factorisation for deep multi-agent reinforcement learning. In: Advances in neural information processing systems, pp 10199–10210 Rattal S, Badri A, Moughit M et al (2025) Ai-driven optimization of low-energy IoT protocols for scalable and efficient smart healthcare systems. IEEE Access 13:48401–48415 Google Scholar Rawat RS, Rana DS (2023) Implementation of reinforcement learning and imaging for better decision-making in the medical sector. In: 2023 IEEE 8th international conference for convergence in technology (I2CT), pp 1–4 Ren Y, Zhang H, Du L et al (2024) Stealthy black-box attack with dynamic threshold against Marl-based traffic signal control system. IEEE Trans Ind Inf 20(10):12021–12031 Google Scholar Rosenblatt F (1958) The perceptron: a probabilistic model for information storage and organization in the brain. Psychol Rev 65(6):386–408 Google Scholar Rummery GA, Niranjan M (1994) On-line Q-learning using connectionist systems, vol 37. University of Cambridge, Department of Engineering, Cambridge Google Scholar Saeed AK, Holguin F, Yasin AS et al (2024) Multi-agent and multi-target reinforcement learning for satellite sensor tasking. In: 2024 IEEE aerospace conference, pp 1–13 Schubert E, Sander J, Ester M et al (2017) DBSCAN revisited, revisited: why and how you should (still) use DBSCAN. ACM Trans Database Syst (TODS) 42(3):1–21 Google Scholar Schütz HJ, Kolisch R (2012) Approximate dynamic programming for capacity allocation in the service industry. Eur J Oper Res 218(1):239–250 Google Scholar Schwung A, Schwung D, Abdul Hameed MS (2019a) Cooperative robot control in flexible manufacturing cells: centralized vs. distributed approaches. In: 2019 IEEE 17th international conference on industrial informatics (INDIN), pp 233–238 Schwung D, Schwung A, Ding SX (2019b) Actor-critic reinforcement learning for energy optimization in hybrid production environments. Int J Comput 18(4):360–371 Google Scholar Schwung D, Schwung A, Ding SX (2022) Distributed self-optimization of modular production units: a state-based potential game approach. IEEE Trans Cybern 52(4):2174–2185 Google Scholar Shang C, Fu L, Xiao H et al (2024) Joint optimization of power generation and voyage scheduling in ship power system based on operating scene clustering and multitask deep reinforcement learning. IEEE Trans Transp Electr 10(4):8432–8442 Google Scholar Shen X, Zhang X, Wang Y (2021) Kernel temporal difference based reinforcement learning for brain machine interfaces. In: 2021 43rd annual international conference of the IEEE engineering in medicine & biology society (EMBC), pp 6721–6724 Shi B, Yuan H, Shi R (2018) Pricing cloud resource based on multi-agent reinforcement learning in the competing environment. In: 2018 IEEE Intl conf on parallel & distributed processing with applications, ubiquitous computing & communications, big data & Cloud computing, social computing & networking, sustainable computing & communications (ISPA/IUCC/BDCloud/SocialCom/SustainCom), pp 462–468 Silver D, Lever G, Heess N et al (2014) Deterministic policy gradient algorithms. In: International conference on machine learning, PMLR, pp 387–395 Singh A, Jain T, Sukhbaatar S (2019) Learning when to communicate at scale in multiagent cooperative and competitive tasks. In: 7th international conference on learning representations, ICLR 2019, New Orleans, LA, USA Skinner BF (1956) A case history in scientific method. Am Psychol 11(5):221–233 Google Scholar Skinner BF (1958) Reinforcement today. Am Psychol 13(3):94–99 Google Scholar Son K, Kim D, Kang WJ et al (2019) QTRAN: learning to factorize with transformation for cooperative multi-agent reinforcement learning. In: International conference on machine learning, PMLR, pp 5887–5896 Son K, Kim D, Kang WJ et al (2020) Qtran++: improved value transformation for cooperative multi-agent reinforcement learning. arXiv:2006.12010 Sui F, Yue W, Zhang Z et al (2023) Trial-and-error learning for mems structural design enabled by deep reinforcement learning. In: 2023 IEEE 36th international conference on micro electro mechanical systems (MEMS), pp 503–506 Sukhbaatar S, Fergus R (2016) Learning multiagent communication with backpropagation. In: Proceedings of the 29th conference on neural information processing systems, NIPS, Barcelona, pp 2252–2260 Sun C, Mu C (2020) Some key scientific problems in multi-agent deep reinforcement learning. Acta Automatica Sinica 46(7):1301–1312 Google Scholar Sun Z, Zhou Y, Tang S et al (2024) Noise suppression zeroing neural network for online solving the time-varying inverse kinematics problem of four-wheel mobile manipulators with external disturbances. Artif Intell Rev 57(8):211 Google Scholar Sunehag P, Lever G, Gruslys A et al (2017) Value-decomposition networks for cooperative multi-agent learning. arXiv:1706.05296 Sutton RS (1988) Learning to predict by the methods of temporal differences. Mach Learn 3:9–44 Google Scholar Sutton R, Barto A (1998) Reinforcement learning: an introduction. IEEE Trans Neural Netw 9(5):1054–1054 Google Scholar Sutton RS, McAllester D, Singh S et al (1999) Policy gradient methods for reinforcement learning with function approximation. Adv Neural Inf Process Syst 12:1–7 Google Scholar Tamar A, Wu Y, Thomas G et al (2016) Value iteration networks. Advances in neural information processing systems 29 Tampuu A, Matiisen T, Kodelja D et al (2017) Multiagent cooperation and competition with deep reinforcement learning. PLoS ONE 12(4):e0172395 Google Scholar Tan M (1993) Multi-agent reinforcement learning: independent vs. cooperative agents. In: Proceedings of the tenth international conference on machine learning, pp 330–337 Uc-Cetina V, Navarro-Guerrero N, Martin-Gonzalez A et al (2023) Survey on reinforcement learning for language processing. Artif Intell Rev 56(2):1543–1575 Google Scholar Urbanowicz RJ, Moore JH (2009) Learning classifier systems: a complete introduction, review, and roadmap. J Artif Evol Appl 2009:1–25 Google Scholar Van der Maaten L, Hinton G (2008) Visualizing data using t-SNW. J Mach Learn Res 9(11):2579 Google Scholar Van Hasselt H, Guez A, Silver D (2016) Deep reinforcement learning with double q-learning. In: Proceedings of the AAAI conference on artificial intelligence Waltz DL, Fu KS (1965) A heuristic approach to reinforcement learning control systems. IEEE Trans Autom Control 10(4):390–398 Google Scholar Wang Z, Schaul T, Hessel M et al (2016) Dueling network architectures for deep reinforcement learning. arXiv:1511.06521 Wang J, Ren Z, Liu T et al (2020) Qplex: duplex dueling multi-agent q-learning. arXiv:2008.01062 Wang H, Tao J, Peng T et al (2022a) Dynamic inventory replenishment strategy for aerospace manufacturing supply chain: combining reinforcement learning and multi-agent simulation. Int J Prod Res 60(13):4117–4136 Google Scholar Wang L, Zhang Y, Hu Y et al (2022b) Individual reward assisted multi-agent reinforcement learning. In: Proceedings of the 39th international conference on machine learning, ACM Press, Baltimore, pp 23417–23432 Wang D, Gao N, Liu D et al (2024a) Recent progress in reinforcement learning and adaptive dynamic programming for advanced control applications. IEEE/CAA J Autom Sin 11(1):18–36 Google Scholar Wang H, Liu Z, Hu G et al (2024b) Offline meta-reinforcement learning for active pantograph control in high-speed railways. IEEE Trans Ind Inform Wang S, Yue Q, Xu Z et al (2025) A collaborative multi-agent reinforcement learning approach for non-stationary environments with unknown change points. Mathematics 13(11):1738 Google Scholar Watkins CJ, Dayan P (1992) Q-learning. Mach Learn 8:279–292 Google Scholar Wei S, Wang S, Sun S et al (2022) Stock ranking prediction based on an adversarial game neural network. IEEE Access 10:65028–65036 Google Scholar Wen G, Yang T, Zhou J et al (2023) Reinforcement learning and adaptive/approximate dynamic programming: a survey from theory to applications in multi-agent systems. Kongzhi yu Juece/Control Decision 38(5):1200–1230 Google Scholar Wu J, Li D, Yu Y et al (2024) An attention mechanism and adaptive accuracy triple-dependent MADDPG formation control method for hybrid UAVs. IEEE Trans Intell Transpl Syst 25(9):11648–11663 Google Scholar Xia X, Fu X, Zhong S et al (2023) A multi-agent convolution deep reinforcement learning network for aeroengine fleet maintenance strategy optimization. J Manuf Syst 68:410–425 Google Scholar Xing X, Zhou Z, Li Y et al (2024) Multi-UAV adaptive cooperative formation trajectory planning based on an improved MATD3 algorithm of deep reinforcement learning. IEEE Trans Veh Technol 73(9):12484–12499 Google Scholar Xing-Xing L, Yang-He F, Yang M et al (2020) Deep multi-agent reinforcement learning: a survey. Acta Automatica Sinica 46(12):2537–2557 Google Scholar Xu D, Yu Z, Liao X et al (2024) A graph deep reinforcement learning traffic signal control for multiple intersections considering missing data. IEEE Trans Veh Technol 73(12):18307–18319 Google Scholar Xue Y, Chen W (2023) Multi-agent deep reinforcement learning for UAVs navigation in unknown complex environment. IEEE Trans Intell Veh 9(1):2290–2303 Google Scholar Xue S, Luo B, Liu D (2021) Event-triggered adaptive dynamic programming for unmatched uncertain nonlinear continuous-time systems. IEEE Trans Neural Netw Learn Syst 32(7):2939–2951 Google Scholar Xue S, Luo B, Liu D et al (2022) Event-triggered ADP for tracking control of partially unknown constrained uncertain systems. IEEE Trans Cybern 52(9):9001–9012 Google Scholar Xue S, Zhang W, Luo B et al (2025a) Integral reinforcement learning-based dynamic event-triggered nonzero-sum games of USVs. IEEE Trans Cybern 55(4):1706–1716 Google Scholar Xue S, Zhao N, Zhang W et al (2025b) A hybrid adaptive dynamic programming for optimal tracking control of USVs. IEEE Trans Neural Netw Learn Syst 36(6):9961–9969 Google Scholar Xu H, Zuo L, Sun F et al (2022) Low-latency patient monitoring service for cloud computing based healthcare system by applying reinforcement learning. In: 2022 IEEE 8th international conference on computer and communications (ICCC), pp 1373–1377 Yan Z, Kreidieh AR, Vinitsky E et al (2023) Unified automatic control of vehicular systems with reinforcement learning. IEEE Trans Autom Sci Eng 20(2):789–804 Google Scholar Yang Y, Hao J, Chen G et al (2018) Multi-agent soft q-learning. arXiv:1804.04175 Yang Y, Hao J, Chen G et al (2020a) Q-value path decomposition for deep multi-agent reinforcement learning. arXiv:2002.03950 Yang Y, Hao J, Liao B et al (2020b) Qatten: a general framework for cooperative multi-agent reinforcement learning. arXiv:2002.03939 Yang X, Song Z, King I et al (2022a) A survey on deep semi-supervised learning. IEEE Trans Knowl Data Eng 35(9):8934–8954 Google Scholar Yang Q, Wang S, Zhang Q et al (2023) Hundreds guide millions: adaptive offline reinforcement learning with expert guidance. IEEE Trans Neural Netw Learn Syst 35(11):16288–16300 Google Scholar Yu C, Wang X, Xu X et al (2019) Distributed multiagent coordinated learning for autonomous driving in highways based on dynamic coordination graphs. IEEE Trans Intell Transp Syst 21(2):735–748 Google Scholar Yu C, Wang X, Xu X et al (2020) Distributed multiagent coordinated learning for autonomous driving in highways based on dynamic coordination graphs. IEEE Trans Intell Transp Syst 21(2):735–748 Google Scholar Yuan L, Wang J, Zhang F et al (2022) Multi-agent incentive communication via decentralized teammate modeling. In: Proceedings of the 36th AAAI conference on artificial intelligence, AAAI Press, Ottawa, pp 9466–9474 Yu C, Velu A, Vinitsky E et al (2022) The surprising effectiveness of PPO in cooperative multi-agent games. In: Advances in Neural Information Processing Systems, pp 24611–24624 Zhang X, Wang Y (2023) A kernel reinforcement learning decoding framework integrating neural and feedback signals for brain control*. In: 2023 45th annual international conference of the IEEE engineering in medicine & biology society (EMBC), pp 1–4 Zhang ML, Zhou ZH (2011) Cotrade: confident co-training with data editing. IEEE Trans Syst Man Cybern B 41(6):1612–1626 Google Scholar Zhang K, Yang Z, Başar T (2021) Multi-agent reinforcement learning: a selective overview of theories and algorithms. Handbook of reinforcement learning and control pp 321–384 Zhang D, Wang Y, Jiang K et al (2023) Safe optimal robust control of nonlinear systems with asymmetric input constraints using reinforcement learning. Appl Intell 54(1):1–13 Google Scholar Zhang D, Wang Y, Meng L et al (2024) Adaptive critic design for safety-optimal FTC of unknown nonlinear systems with asymmetric constrained-input. ISA Trans 155:309–318 Google Scholar Zhang D, Yu C, Li Z et al (2025) A lightweight network enhanced by attention-guided cross-scale interaction for underwater object detection. Appl Soft Comput 184:113811 Google Scholar Zhang Y, Zhao B, Liu D (2025) Distributed optimal containment control of wheeled mobile robots via adaptive dynamic programming. IEEE Trans Syst Man Cybern Syst 55(9):5876–5886 Google Scholar Zhao F, Hua Y, Zheng H et al (2023a) Cooperative target pursuit by multiple fixed-wing UAVs based on deep reinforcement learning and artificial potential field. In: 2023 42nd Chinese control conference (CCC), pp 5693–5698 Zhao M, Wang D, Qiao J et al (2023b) Advanced value iteration for discrete–time intelligent critic control: a survey. Artif Intell Rev 56(10):12315–12346 Google Scholar Zhao B, Zhang S, Liu D (2025) Self-triggered approximate optimal neuro-control for nonlinear systems through adaptive dynamic programming. IEEE Trans Neural Netw Learn Syst 36(3):4713–4723 Google Scholar Zheng H, Ryzhov IO, Xie W et al (2021) Personalized multimorbidity management for patients with type 2 diabetes using reinforcement learning of electronic health records. Drugs 81:471–482 Google Scholar Zhu H, Vyetrenko S, Dwarakanath K et al (2023a) Once burned, twice shy? The effect of stock market bubbles on traders that learn by experience. In: 2023 Winter simulation conference (WSC), pp 291–302 Zhu Z, Lin K, Jain AK et al (2023b) Transfer learning in deep reinforcement learning: a survey. IEEE Trans Pattern Anal Mach Intell 45(11):13344–13362 Google Scholar Download references This work was supported in part by Nanyang Normal University Foundation of China under Grant (2024PY011), in part by the Key Scientific and Technological Project of the Henan Province under Grant (232102311012, 232102311004), in part by the Key Research Projects Funding Program for Higher Education Institutions of Henan Province under Grant (24A320003), and in part by the Key Medical Science and Technology Research Project of Henan Province under Grant (SBJ202103098, LHGJ20220662). School of Artificial Intelligence, Henan University, North Section of Mingli Road, Zhengzhou, 450046, Henan, China Dehua Zhang, Qingsong Yuan, Lei Meng, Ruixue Xia & Chunbin Qin School of Intelligent Manufacturing and Electrical Engineering (Collaborative Innovation Center of Intelligent Explosion-proof Equipment, Henan Province), Nanyang Normal University, Wolong Road, Nanyang, 473061, Henan, China Wei Liu Search author on:PubMed Google Scholar Search author on:PubMed Google Scholar Search author on:PubMed Google Scholar Search author on:PubMed Google Scholar Search author on:PubMed Google Scholar Search author on:PubMed Google Scholar D.Z. and Q.Y. were responsible for the conception and design of the study and wrote the main manuscript text. L.M. and R.X. participated in the literature review section and provided theoretical support for the manuscript. W.L. prepared the figures and tables and helped write the summary. C.Q. provided the final review and revision of the manuscript. All authors participated in the discussions and reviewed and approved the manuscript content. Correspondence to Wei Liu or Chunbin Qin. The authors declare no conflict of interest. Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/. Reprints and permissions Zhang, D., Yuan, Q., Meng, L. et al. Reinforcement learning for single-agent to multi-agent systems: from basic theory to industrial application progress, a survey. Artif Intell Rev (2025). https://doi.org/10.1007/s10462-025-11439-9 Download citation Received: 29 April 2025 Accepted: 30 October 2025 Published: 27 November 2025 DOI: https://doi.org/10.1007/s10462-025-11439-9 Anyone you share the following link with will be able to read this content: Sorry, a shareable link is not currently available for this article. Provided by the Springer Nature SharedIt content-sharing initiative Avoid common mistakes on your manuscript. Advertisement 178.128.244.209 Not affiliated © 2025 Springer Nature

Written by 36Kr English Published on 3 Dec 2025 11 mins read Whether in academic research or collaborations with companies such as Ant Group, Yi Wu encourages his team to keep an entrepreneurial mindset: move quickly, iterate often, and avoid fearing failure. An assistant professor at Tsinghua University’s Institute for Interdisciplinary Information Sciences and head of the AReaL project, Wu studies reinforcement learning algorithms and applications of artificial intelligence. In May, his team and Ant Research jointly open-sourced AReaL-lite, described by the researchers as the first asynchronous reinforcement learning training framework designed to improve training efficiency and reduce GPU waste. The claim has not been independently verified. As a young tech leader, Wu emphasizes learning through trial and error. He resists the idea that a lack of resources is an acceptable reason for stalled progress, saying that building something new often requires creating the resources along the way. That philosophy surfaced at the Inclusion Conference on the Bund in September, where Wu argued that teams should release products as soon as they work at a basic level so they can learn from market feedback. The goal, he said, is not to wait for a perfect launch but to identify problems early and refine the product. His approach is rooted in earlier entrepreneurial experience. In 2023, his team founded Prosocial Intelligence, an agentic AI company that later evolved into AReaL. Wu is informally grouped with Jianyu Chen, Yang Gao, and Huazhe Xu as part of the “Berkeley Four,” a nickname referencing their shared academic background in AI research. All four studied in the US. Wu was the first to return to China, and he encouraged the others to follow. At Tsinghua University, Wu often reminds students that innovation requires venturing into unfamiliar territory. He argues that AI breakthroughs benefit more from long-term focus than from trying to chase every potential direction. He also holds a specific view of AI’s future, where intelligent agents interpret loosely defined human intentions, complete long-horizon tasks, and eventually move from digital spaces into the physical world. At this year’s World Artificial Intelligence Conference (WAIC), he described a scenario where a person could verbally ask a robot to tidy a room, and the robot would spend hours finishing the task. Reinforcement learning, his area of research, is central to that goal. He notes that the technique enables AI systems to learn through interaction and exploration, in contrast to supervised learning, which depends on continuous human instruction and struggles with long, open-ended tasks. Despite his rigorous academic work, Wu’s social media presence is lighter. On Xiaohongshu, he posts research updates, responds to questions about careers in AI, and occasionally ranks his favorite bubble tea flavors. Wu spoke with 36Kr about his take on AI’s future, entrepreneurship, and building efficient teams. The following transcript has been edited and consolidated for brevity and clarity. Wu Yi (WY): I think enabling AI to complete long-horizon tasks is an irreversible trend. Meanwhile, the commands humans give AI will become increasingly simple and vague. It’s hard to predict the exact product form, but one thing is certain: we’re moving from users actively driving AI to AI proactively anticipating what users want and completing it. This pattern already appeared during the mobile internet era. With search engines, users had to look for information themselves. Then came platforms like Zhihu, and later ByteDance’s products, whose algorithms pushed desired content directly to users. So I think, eventually, people will forget what a “search box” is as AI increasingly caters to human laziness. Ultimately, a whole new kind of product will emerge, and it will mark a generational opportunity. WY: A smart embodied agent can infer user intentions from fuzzy instructions, complete tasks accurately, and even anticipate unspoken needs. For example, if you tell your home robot that you can’t find your power bank, it may reason and act on its own, searching based on your habits and its memory of where you last used it. WY: They can cooperate to handle more complex tasks. Take robots in soccer for instance. Just like human players, when robots encounter familiar situations, a quick scan of the environment signals what formation to take. If you have several intelligent agents, the next step is defining how they communicate. In the digital world, one approach is a master agent that coordinates many smaller ones. You can use different models or even a single model structured like a planner directing many executors. That’s the idea behind a multi-agent system. I often cite Claude Code and Gemini as an example: Claude Code excels at programming but has short context and high cost, while Gemini handles large amounts of content but lacks reasoning power. Let Gemini first read an entire codebase and extract key parts, then let Claude Code write the actual code. It’s like pairing a smart but frail thinker with a strong but dull worker. The combination makes a highly efficient multi-agent system. In embodied scenarios, such as cleaning a space, robots “communicate” to plan who sweeps and who mops, working together to finish the job. WY: Transitioning from the digital to the physical world requires multimodal data, moving training environments from computers into reality. In the digital world, tools are mostly bits. They execute reliably, but in the physical world, using tools like grabbing a bag or opening a door still involves high error rates. Embodied intelligence, therefore, develops more slowly and with greater complexity. That said, if we look far enough ahead, once the physical world has been sufficiently digitized, the core technical challenges for all types of agents will converge. If one day a machine can reliably operate almost any physical tool, building an embodied agent that can function autonomously for an entire day will be technically no different from a digital agent. WY: At Prosocial, we made plenty of mistakes in early hiring. Many employees treated it like a regular job, not a startup, and didn’t grasp what entrepreneurship really means. Objectively, the team wasn’t fully ready to adopt a mindset catered to running a startup in the AI era. Still, everyone was learning. It was inevitable to stumble. Now I really dislike hearing that something can’t be done because we don’t have the resources. Startups rarely have abundant resources, and people create them while pursuing their goals. Entrepreneurial teams need that spark of innovation and the right level of conviction. Innovation isn’t about placing bets. Startups must believe deeply in what they are doing. We don’t have enough resources to hedge across multiple tracks hoping one will win. That only breeds mediocrity. Entrepreneurial spirit means believing something is right even if you fail to achieve it yourself. Someday, someone will. WY: In August 2018, I finished my internship at ByteDance in Beijing. Though I earned my PhD at UC Berkeley, my experience at ByteDance had a big influence on me. Since 2016, I’d interned intermittently in various ByteDance teams and was among the first members of its AI lab, witnessing the end of China’s mobile internet boom. By August 2018, I knew I wanted to return to China. Partly, I saw enormous opportunity in China’s development. Partly, I felt a clear ceiling for Chinese professionals in the US. Unless you become fully American, you face that question: if you want to make a real impact, do you want to be Chinese or American? I realized I didn’t want to compromise by becoming American. Many people say that they aren’t ready yet, and that they will wait until they are. Some Chinese scholars in the US say they will develop there for a few more years, then return. But my view is: if you’re sure you’ll do something someday, the best time was yesterday. The second best is today. So I decided to come back. A month later, I turned down ByteDance’s return offer. In October 2018, I joined Tsinghua University as a faculty member. Then I shared my thoughts with my fellow Berkeley classmates, telling them to seize the opportunity, and indeed, some were persuaded. Looking back, the timing really was ideal. We early returnees enjoyed the dividends of that wave. WY: Exactly. I’ve been “learning by reinforcement” all along, hitting every pitfall as quickly as possible. Honestly, learning through trial and error teaches you more deeply and generalizes better than supervised fine-tuning. Building products works the same way. I often say: once you make something, release it immediately. In the AI era, even great products need exposure. Get them out there, gather feedback, and iterate fast. Even negative feedback shows you where the pitfalls are. Of course, with high-quality supervised fine-tuning data, reinforcement learning becomes more efficient. Negative rewards are costly, so I share my experiences to help others learn faster. WY: I have a method to help me make decisions quickly: I flip a coin. Before it even lands, I usually already know my answer. I’m always the one who flips first. WY: Yes. I’ve thought about it: if I built a startup from zero to one and later, as it scaled from one to 100, I was no longer the most visible leader, would I be fine with that? The answer I arrived at is yes. At that inflection point, I’d likely bring in a professional manager and move on to the next project. Managing hundreds of people isn’t what I enjoy most. That said, I’m reflecting on whether such idealism might limit me. Maybe when that moment comes, I’ll choose differently. But if you ask me now, I’d still choose to keep creating zero-to-one projects. WY: Because it lets AI learn from real interaction. Supervised learning or fine-tuning means humans constantly tell AI what to do, but possibilities are infinite, and humans can’t give instructions for ten hours straight. Human instructions also differ from how AI “thinks.” When AIs simply memorize, they don’t truly understand, and thus generalize poorly. Reinforcement learning encourages active interaction with environments, even teaching AI to ask questions when uncertain. It cultivates self-driven exploration, something only the technique can achieve. WY: I now think the most important factor is prompting, specifically creating large numbers of high-quality prompts. Here’s an analogy: a teacher tutoring a student in math. Prompts are the teacher’s problems, search and exploration are the student’s problem-solving process, and the reward model is the teacher’s feedback. Choosing the right difficulty is crucial. Make it too advanced and the student gives up, but if it’s too easy, they learn nothing. The same applies to reinforcement learning: data volume alone doesn’t help. Appropriateness does. WY: The relationship goes two ways. One is locomotion, where reinforcement learning is already mature and doesn’t require pre-training. The other involves long-horizon reasoning and planning, usually combined with large pre-trained models. That area only became popular after ChatGPT. These two aspects form a spectrum from high-frequency control for short tasks to abstract reasoning for complex ones. Traditional reinforcement learning for control doesn’t need pre-training, think quadruped robots that can run and jump. Tiny neural networks trained in simulation can directly control real robots without pre-training. In such tasks, reinforcement learning trains the network to output control signals for each joint, enabling motion over durations as short as seconds, not hours. By contrast, models like ChatGPT and DeepSeek’s R1 use reinforcement learning after pre-training to enhance reasoning. Large models that employ reinforcement learning can think for minutes or hours, use common sense, break complex problems into subtasks, and call tools. But so far, this success remains in the digital realm, not the physical. In between lies the vision-language-action (VLA) model, which is often discussed in embodied intelligence research. WY: VLA applies pre-training to the physical world. Researchers gather massive data to pre-train models that not only complete short tasks like running or jumping but also generalize to minute-long activities like folding towels and pouring water. To reach longer-range tasks like cooking or cleaning, robots must perform for hours, combining fine control with abstract reasoning and human interaction, just like digital agents, except it’s in the physical world. So I see embodied agents as systems that merge locomotion or VLA as the “small brain” controlling motion, with language models enhanced using reinforcement learning as the “big brain.” Unlike digital agents, physical agents still get less attention. Most people focus on hardware aspects such as gripping accuracy, object sorting, and so on. But altering the physical world is always harder. Given my focus, I’m working on stabilizing long-duration reasoning before combining it with physical control. WY: Our current plan is a layered structure. As I said at WAIC, the higher you go, the more human knowledge you need. Likewise, the lower, the less. Lower layers handle instinctive reactions like grabbing a cup based on tactile feedback or simple physics. Upper layers need prior knowledge. So there’s a natural division between digital and physical agents. I don’t think VLA will be the final paradigm, because it isn’t large enough in scale to become a fully capable agent. We’ll perfect the digital agent format first while others explore the embodied side, then merge them when the timing is right. WY: In the internet era, building a product required four or five people which typically includes a frontend developer, a backend developer, and a product manager. In the AI era, one person might be sufficient to handle all that. Previously, small companies outsourced many tasks. Now, AI can streamline not just internal work but also outsourcing itself. If a team can run heavily on AI, its capabilities will naturally scale outward, because if AI can serve us, it can serve others, too. That’s a new product opportunity. Our AReaL team has only six members, with some external support. Counting everyone, we could still make it leaner. I want the team to stay minimalist, and that’s why it has always been small. WY: First, a modern agent-focused team must use many agents every day itself. Second, I combine algorithm and infrastructure teams into a single full-stack unit. Traditional structures separate algorithms from systems and add data collection teams, creating a segregation dynamic whereby algorithm teams are the “clients,” while engineers become the contractors doing the “dirty work.” That division kills innovation. Once you’re used to being the former, you avoid the grunt work. And if you’re in the latter, you lose creative space. OpenAI didn’t magically invent new algorithms, it simply perfected the details. So to excel in infrastructure and data, you need to dig deep. With that groundwork, algorithms can shine. That’s why algorithms and infrastructure must be co-designed and co-evolved. A small, highly capable team can collectively fulfill this. Large organizations, say with 200 people, can’t avoid silos. Limited communication bandwidth leads to rigid roles and inefficiency. So a compact, full-stack setup and high innovation go hand in hand. Forget the 200-person org chart. In the AI era, it’s all about going from zero to one, so take bold, radical approaches and build anew. KrASIA Connection features translated and adapted content that was originally published by 36Kr. This article was written by Fu Chong for 36Kr. Loading... Subscribe to our newsletters KrASIA A digital media company reporting on China's tech and business pulse.

Navigating in a fluid flow while being carried by it, using only information accessible from on-board sensors, is a problem commonly faced by small planktonic organisms. It is also directly relevant to autonomous robots deployed in the oceans. In the last ten years, the fluid mechanics community has widely adopted reinforcement learning, often in the form of its simplest implementations, to address this challenge. But it is unclear how good are the strategies learned by these algorithms. In this paper, we perform a quantitative assessment of reinforcement learning methods applied to navigation in partially observable flows. We first introduce a well-posed problem of directional navigation for which a quasioptimal policy is known analytically. We then report on the poor performance and robustness of commonly used algorithms (Q-Learning, Advantage Actor Critic) in flows regularly encountered in the literature: Taylor-Green vortices, Arnold-Beltrami-Childress flow, and two-dimensional turbulence. We show that they are vastly surpassed by PPO (Proximal Policy Optimization), a more advanced algorithm that has established dominance across a wide range of benchmarks in the reinforcement learning community. In particular, our custom implementation of PPO matches the theoretical quasi-optimal performance in turbulent flow and does so in a robust manner. Reaching this result required the use of several additional techniques, such as vectorized environments and generalized advantage estimation, as well as hyperparameter optimization. This study demonstrates the importance of algorithm selection, implementation details, and fine-tuning for discovering truly smart autonomous navigation strategies in complex flows. Connectez-vous pour contacter le contributeur https://hal.science/hal-05343166 Soumis le : lundi 3 novembre 2025-10:15:31 Dernière modification le : vendredi 7 novembre 2025-12:43:28 Contact Ressources Informations Questions juridiques Portails CCSD

Add the following snippet to your HTML:<iframe frameborder='0' height='385' scrolling='no' src='https://www.hackster.io/HiwonderRobot/what-is-reinforcement-learning-practical-steps-included-0954ef/embed' width='350'></iframe> Build real-world reinforcement learning skills with an open-source robot arm—fully hackable, community-driven, and ready to experiment. Read up about this project on Build real-world reinforcement learning skills with an open-source robot arm—fully hackable, community-driven, and ready to experiment. Reinforcement learning (RL) is one of the most fascinating areas of artificial intelligence—where an agent learns to make decisions by interacting with an environment, receiving feedback through rewards or penalties, and optimizing its behavior over time. From game-playing AIs like AlphaGo to robots learning to walk, RL bridges the gap between perception and action in AI. But moving from theory to real-world application is often challenging: high hardware costs, complex system integration, and difficulties in reproducing experiments can stall progress. This is where accessible, open-source hardware platforms become essential. Enter the Hiwonder SO-ARM101—a fully open-source robotic arm platform born from the Hugging Face LeRobot project. It offers a hands-on, reproducible way to explore embodied AI, imitation learning, and yes, reinforcement learning in the physical world. The SO-ARM101 isn’t just another robotic arm. It’s built on LeRobot, an open-source robotics project from Hugging Face, and follows a fully open philosophy—from hardware designs and firmware to software and example algorithms. This approach lowers the barrier to entry, allowing researchers, students, and makers to focus on experimenting with AI, not struggling with hardware integration. The kit includes 2 robotic arms in a leader-follower setup, making it particularly suited for imitation learning workflows. You can physically guide the leader arm to demonstrate a task—such as picking up an object or stacking blocks—while the follower arm records joint trajectories and camera data. After multiple demonstrations, the system learns a policy that allows the follower to perform the task autonomously. It’s an intuitive and effective way to get started with learning from demonstration (LfD), which often serves as a foundation for more advanced RL methods. To support stable and repeatable real-world experiments, the SO-ARM101 incorporates several key upgrades over the baseline LeRobot design: While the leader-follower setup naturally supports imitation learning, the SO-ARM101 is also a capable platform for exploring reinforcement learning. Consider experiments like: The platform comes with step-by-step guides and reproducible examples, regularly updated to align with the latest LeRobot releases. Even without a background in robotics or RL, you can follow along to set up the system, collect demonstration data, train models, and deploy learned behaviors. It’s not only a research tool—it’s also an educational platform that makes embodied AI tangible and approachable. Reinforcement learning is more than equations and algorithms—it’s about agents that act, learn, and adapt in the real world. Open-source platforms like the SO-ARM101 help turn theoretical concepts into running experiments. By lowering cost and complexity, they enable more people to participate in embodied AI research, iterate on ideas, and contribute back to the community. If you’ve been curious about reinforcement learning beyond simulations, or if you’re looking for a reliable hardware platform to test AI policies in physical environments, this community-driven, fully open robotic arm could be the right place to start. Download Hiwonder LeRobot tutorials! Hackster.io, an Avnet Community © 2025

AgiBot successfully deployed Real-World Reinforcement Learning (RW-RL) in an active manufacturing line with Longcheer Technology. This marks the first industrial application of reinforcement learning in robotics, bridging AI research with real-world production. Traditional robots rely on rigid programming, requiring costly reconfiguration and custom fixtures. AgiBot's RW-RL system enables robots to learn and adapt on the factory floor, acquiring new skills in minutes instead of weeks while maintaining industrial-grade stability. Unlike lab-based RL, AgiBot's system was tested in near-production conditions, proving its industrial readiness. Robots demonstrated resilience against disruptions (temperature shifts, vibrations, misalignment) while maintaining precision assembly. When production models changed, robots retrained in minutes without manual reprogramming. AgiBot plans to expand RW-RL into consumer electronics and automotive manufacturing, focusing on plug-and-play robotic solutions. Their LinkCraft platform (converting human motion into robot actions) and G2 robot (powered by NVIDIA's Jetson Thor T5000) enable real-time AI processing. If scalable, this could usher in the adaptive factory era, where robots continuously learn and optimize without halting operations. Traditional robots rely on rigid programming, requiring costly reconfiguration and custom fixtures. AgiBot's RW-RL system enables robots to learn and adapt on the factory floor, acquiring new skills in minutes instead of weeks while maintaining industrial-grade stability. Unlike lab-based RL, AgiBot's system was tested in near-production conditions, proving its industrial readiness. Robots demonstrated resilience against disruptions (temperature shifts, vibrations, misalignment) while maintaining precision assembly. When production models changed, robots retrained in minutes without manual reprogramming. AgiBot plans to expand RW-RL into consumer electronics and automotive manufacturing, focusing on plug-and-play robotic solutions. Their LinkCraft platform (converting human motion into robot actions) and G2 robot (powered by NVIDIA's Jetson Thor T5000) enable real-time AI processing. If scalable, this could usher in the adaptive factory era, where robots continuously learn and optimize without halting operations. Unlike lab-based RL, AgiBot's system was tested in near-production conditions, proving its industrial readiness. Robots demonstrated resilience against disruptions (temperature shifts, vibrations, misalignment) while maintaining precision assembly. When production models changed, robots retrained in minutes without manual reprogramming. AgiBot plans to expand RW-RL into consumer electronics and automotive manufacturing, focusing on plug-and-play robotic solutions. Their LinkCraft platform (converting human motion into robot actions) and G2 robot (powered by NVIDIA's Jetson Thor T5000) enable real-time AI processing. If scalable, this could usher in the adaptive factory era, where robots continuously learn and optimize without halting operations. AgiBot plans to expand RW-RL into consumer electronics and automotive manufacturing, focusing on plug-and-play robotic solutions. Their LinkCraft platform (converting human motion into robot actions) and G2 robot (powered by NVIDIA's Jetson Thor T5000) enable real-time AI processing. If scalable, this could usher in the adaptive factory era, where robots continuously learn and optimize without halting operations. AgiBot, a robotics firm specializing in embodied intelligence, has achieved a major milestone by successfully deploying its Real-World Reinforcement Learning (RW-RL) system in an active manufacturing line with Longcheer Technology. This marks the first industrial-scale application of reinforcement learning in robotics, bridging advanced AI research with real-world production—a breakthrough that could redefine flexible manufacturing. According to BrightU.AI's Enoch, RL is a type of machine learning where an agent learns to behave in an environment by performing actions and receiving rewards or penalties. The agent's goal is to maximize the cumulative reward over time, learning from its environment through trial and error. This learning process is akin to how humans and animals learn from their surroundings, making RL a powerful tool for solving complex problems in various fields, including robotics, gaming, resource management and more. Traditional industrial robots rely on rigid programming, requiring extensive tuning, costly reconfiguration and custom fixtures for each task. Even advanced "vision + force-control" systems struggle with parameter sensitivity and maintenance complexity. AgiBot's RW-RL system tackles these limitations by allowing robots to learn and adapt directly on the factory floor—acquiring new skills in minutes rather than weeks while maintaining industrial-grade stability. Dr. Jianlan Luo, AgiBot's Chief Scientist, stated that their "system achieves stable, repeatable learning on real machines" closing the gap between academic research and industrial deployment. Key advantages of RW-RL AgiBot highlights three core benefits of its reinforcement learning system: Rapid Deployment – Training time slashed from weeks to minutes. High Adaptability – Robots autonomously compensate for variations like part misalignment while maintaining 100 percent task completion. Flexible Reconfiguration – Production line changes require minimal hardware adjustments, eliminating costly downtime. Unlike lab-based demonstrations, AgiBot's system was validated under near-production conditions, proving its readiness for industrial use. Reinforcement learning—where robots optimize performance through trial and error—has long been confined to research papers and controlled experiments. AgiBot's breakthrough integrates perception, decision-making and motion control into a unified loop, enabling robots to self-correct in real-time. The Longcheer pilot demonstrated RW-RL's resilience against environmental disruptions—including vibration, temperature shifts and part misalignment—while maintaining precision assembly. When production models changed, the robot retrained in minutes without manual reprogramming, showcasing unprecedented flexibility. The future of adaptive factories AgiBot and Longcheer plan to expand RW-RL into consumer electronics and automotive manufacturing, focusing on modular, plug-and-play robotic solutions that integrate seamlessly with existing systems. The company's LinkCraft platform—which converts human motion videos into robot actions—complements this advancement, reducing programming barriers. Meanwhile, AgiBot's G2 robot, powered by NVIDIA's Jetson Thor T5000, suggests that real-time AI processing is enabling this leap forward. While Google's Intrinsic and NVIDIA's Isaac Lab have pioneered reinforcement learning frameworks, AgiBot appears to be the first to deploy RL in live production. If scalable, this could herald the adaptive factory era, where robots continuously learn, optimize and evolve—without halting operations. As factories face increasing demands for customization and rapid model changes, AgiBot's breakthrough may finally make self-learning robotics a commercial reality. Watch the video below about Chinese startup AgiBot beginning mass production of general-purpose humanoid robots. This video is from the SecureLife channel on Brighteon.com. Sources include: TheRobotReport.com BrightU.ai PRNewswire.com Ubergizmo.com Brighteon.com This marks the first industrial-scale application of reinforcement learning in robotics, bridging advanced AI research with real-world production—a breakthrough that could redefine flexible manufacturing. According to BrightU.AI's Enoch, RL is a type of machine learning where an agent learns to behave in an environment by performing actions and receiving rewards or penalties. The agent's goal is to maximize the cumulative reward over time, learning from its environment through trial and error. This learning process is akin to how humans and animals learn from their surroundings, making RL a powerful tool for solving complex problems in various fields, including robotics, gaming, resource management and more. Traditional industrial robots rely on rigid programming, requiring extensive tuning, costly reconfiguration and custom fixtures for each task. Even advanced "vision + force-control" systems struggle with parameter sensitivity and maintenance complexity. AgiBot's RW-RL system tackles these limitations by allowing robots to learn and adapt directly on the factory floor—acquiring new skills in minutes rather than weeks while maintaining industrial-grade stability. Dr. Jianlan Luo, AgiBot's Chief Scientist, stated that their "system achieves stable, repeatable learning on real machines" closing the gap between academic research and industrial deployment. Key advantages of RW-RL AgiBot highlights three core benefits of its reinforcement learning system: Rapid Deployment – Training time slashed from weeks to minutes. High Adaptability – Robots autonomously compensate for variations like part misalignment while maintaining 100 percent task completion. Flexible Reconfiguration – Production line changes require minimal hardware adjustments, eliminating costly downtime. Unlike lab-based demonstrations, AgiBot's system was validated under near-production conditions, proving its readiness for industrial use. Reinforcement learning—where robots optimize performance through trial and error—has long been confined to research papers and controlled experiments. AgiBot's breakthrough integrates perception, decision-making and motion control into a unified loop, enabling robots to self-correct in real-time. The Longcheer pilot demonstrated RW-RL's resilience against environmental disruptions—including vibration, temperature shifts and part misalignment—while maintaining precision assembly. When production models changed, the robot retrained in minutes without manual reprogramming, showcasing unprecedented flexibility. The future of adaptive factories AgiBot and Longcheer plan to expand RW-RL into consumer electronics and automotive manufacturing, focusing on modular, plug-and-play robotic solutions that integrate seamlessly with existing systems. The company's LinkCraft platform—which converts human motion videos into robot actions—complements this advancement, reducing programming barriers. Meanwhile, AgiBot's G2 robot, powered by NVIDIA's Jetson Thor T5000, suggests that real-time AI processing is enabling this leap forward. While Google's Intrinsic and NVIDIA's Isaac Lab have pioneered reinforcement learning frameworks, AgiBot appears to be the first to deploy RL in live production. If scalable, this could herald the adaptive factory era, where robots continuously learn, optimize and evolve—without halting operations. As factories face increasing demands for customization and rapid model changes, AgiBot's breakthrough may finally make self-learning robotics a commercial reality. Watch the video below about Chinese startup AgiBot beginning mass production of general-purpose humanoid robots. This video is from the SecureLife channel on Brighteon.com. Sources include: TheRobotReport.com BrightU.ai PRNewswire.com Ubergizmo.com Brighteon.com This marks the first industrial-scale application of reinforcement learning in robotics, bridging advanced AI research with real-world production—a breakthrough that could redefine flexible manufacturing. According to BrightU.AI's Enoch, RL is a type of machine learning where an agent learns to behave in an environment by performing actions and receiving rewards or penalties. The agent's goal is to maximize the cumulative reward over time, learning from its environment through trial and error. This learning process is akin to how humans and animals learn from their surroundings, making RL a powerful tool for solving complex problems in various fields, including robotics, gaming, resource management and more. Traditional industrial robots rely on rigid programming, requiring extensive tuning, costly reconfiguration and custom fixtures for each task. Even advanced "vision + force-control" systems struggle with parameter sensitivity and maintenance complexity. AgiBot's RW-RL system tackles these limitations by allowing robots to learn and adapt directly on the factory floor—acquiring new skills in minutes rather than weeks while maintaining industrial-grade stability. Dr. Jianlan Luo, AgiBot's Chief Scientist, stated that their "system achieves stable, repeatable learning on real machines" closing the gap between academic research and industrial deployment. Key advantages of RW-RL AgiBot highlights three core benefits of its reinforcement learning system: Rapid Deployment – Training time slashed from weeks to minutes. High Adaptability – Robots autonomously compensate for variations like part misalignment while maintaining 100 percent task completion. Flexible Reconfiguration – Production line changes require minimal hardware adjustments, eliminating costly downtime. Unlike lab-based demonstrations, AgiBot's system was validated under near-production conditions, proving its readiness for industrial use. Reinforcement learning—where robots optimize performance through trial and error—has long been confined to research papers and controlled experiments. AgiBot's breakthrough integrates perception, decision-making and motion control into a unified loop, enabling robots to self-correct in real-time. The Longcheer pilot demonstrated RW-RL's resilience against environmental disruptions—including vibration, temperature shifts and part misalignment—while maintaining precision assembly. When production models changed, the robot retrained in minutes without manual reprogramming, showcasing unprecedented flexibility. The future of adaptive factories AgiBot and Longcheer plan to expand RW-RL into consumer electronics and automotive manufacturing, focusing on modular, plug-and-play robotic solutions that integrate seamlessly with existing systems. The company's LinkCraft platform—which converts human motion videos into robot actions—complements this advancement, reducing programming barriers. Meanwhile, AgiBot's G2 robot, powered by NVIDIA's Jetson Thor T5000, suggests that real-time AI processing is enabling this leap forward. While Google's Intrinsic and NVIDIA's Isaac Lab have pioneered reinforcement learning frameworks, AgiBot appears to be the first to deploy RL in live production. If scalable, this could herald the adaptive factory era, where robots continuously learn, optimize and evolve—without halting operations. As factories face increasing demands for customization and rapid model changes, AgiBot's breakthrough may finally make self-learning robotics a commercial reality. Watch the video below about Chinese startup AgiBot beginning mass production of general-purpose humanoid robots. This video is from the SecureLife channel on Brighteon.com. Sources include: TheRobotReport.com BrightU.ai PRNewswire.com Ubergizmo.com Brighteon.com According to BrightU.AI's Enoch, RL is a type of machine learning where an agent learns to behave in an environment by performing actions and receiving rewards or penalties. The agent's goal is to maximize the cumulative reward over time, learning from its environment through trial and error. This learning process is akin to how humans and animals learn from their surroundings, making RL a powerful tool for solving complex problems in various fields, including robotics, gaming, resource management and more. Traditional industrial robots rely on rigid programming, requiring extensive tuning, costly reconfiguration and custom fixtures for each task. Even advanced "vision + force-control" systems struggle with parameter sensitivity and maintenance complexity. AgiBot's RW-RL system tackles these limitations by allowing robots to learn and adapt directly on the factory floor—acquiring new skills in minutes rather than weeks while maintaining industrial-grade stability. Dr. Jianlan Luo, AgiBot's Chief Scientist, stated that their "system achieves stable, repeatable learning on real machines" closing the gap between academic research and industrial deployment. Key advantages of RW-RL AgiBot highlights three core benefits of its reinforcement learning system: Rapid Deployment – Training time slashed from weeks to minutes. High Adaptability – Robots autonomously compensate for variations like part misalignment while maintaining 100 percent task completion. Flexible Reconfiguration – Production line changes require minimal hardware adjustments, eliminating costly downtime. Unlike lab-based demonstrations, AgiBot's system was validated under near-production conditions, proving its readiness for industrial use. Reinforcement learning—where robots optimize performance through trial and error—has long been confined to research papers and controlled experiments. AgiBot's breakthrough integrates perception, decision-making and motion control into a unified loop, enabling robots to self-correct in real-time. The Longcheer pilot demonstrated RW-RL's resilience against environmental disruptions—including vibration, temperature shifts and part misalignment—while maintaining precision assembly. When production models changed, the robot retrained in minutes without manual reprogramming, showcasing unprecedented flexibility. The future of adaptive factories AgiBot and Longcheer plan to expand RW-RL into consumer electronics and automotive manufacturing, focusing on modular, plug-and-play robotic solutions that integrate seamlessly with existing systems. The company's LinkCraft platform—which converts human motion videos into robot actions—complements this advancement, reducing programming barriers. Meanwhile, AgiBot's G2 robot, powered by NVIDIA's Jetson Thor T5000, suggests that real-time AI processing is enabling this leap forward. While Google's Intrinsic and NVIDIA's Isaac Lab have pioneered reinforcement learning frameworks, AgiBot appears to be the first to deploy RL in live production. If scalable, this could herald the adaptive factory era, where robots continuously learn, optimize and evolve—without halting operations. As factories face increasing demands for customization and rapid model changes, AgiBot's breakthrough may finally make self-learning robotics a commercial reality. Watch the video below about Chinese startup AgiBot beginning mass production of general-purpose humanoid robots. This video is from the SecureLife channel on Brighteon.com. Sources include: TheRobotReport.com BrightU.ai PRNewswire.com Ubergizmo.com Brighteon.com According to BrightU.AI's Enoch, RL is a type of machine learning where an agent learns to behave in an environment by performing actions and receiving rewards or penalties. The agent's goal is to maximize the cumulative reward over time, learning from its environment through trial and error. This learning process is akin to how humans and animals learn from their surroundings, making RL a powerful tool for solving complex problems in various fields, including robotics, gaming, resource management and more. Traditional industrial robots rely on rigid programming, requiring extensive tuning, costly reconfiguration and custom fixtures for each task. Even advanced "vision + force-control" systems struggle with parameter sensitivity and maintenance complexity. AgiBot's RW-RL system tackles these limitations by allowing robots to learn and adapt directly on the factory floor—acquiring new skills in minutes rather than weeks while maintaining industrial-grade stability. Dr. Jianlan Luo, AgiBot's Chief Scientist, stated that their "system achieves stable, repeatable learning on real machines" closing the gap between academic research and industrial deployment. Key advantages of RW-RL AgiBot highlights three core benefits of its reinforcement learning system: Rapid Deployment – Training time slashed from weeks to minutes. High Adaptability – Robots autonomously compensate for variations like part misalignment while maintaining 100 percent task completion. Flexible Reconfiguration – Production line changes require minimal hardware adjustments, eliminating costly downtime. Unlike lab-based demonstrations, AgiBot's system was validated under near-production conditions, proving its readiness for industrial use. Reinforcement learning—where robots optimize performance through trial and error—has long been confined to research papers and controlled experiments. AgiBot's breakthrough integrates perception, decision-making and motion control into a unified loop, enabling robots to self-correct in real-time. The Longcheer pilot demonstrated RW-RL's resilience against environmental disruptions—including vibration, temperature shifts and part misalignment—while maintaining precision assembly. When production models changed, the robot retrained in minutes without manual reprogramming, showcasing unprecedented flexibility. The future of adaptive factories AgiBot and Longcheer plan to expand RW-RL into consumer electronics and automotive manufacturing, focusing on modular, plug-and-play robotic solutions that integrate seamlessly with existing systems. The company's LinkCraft platform—which converts human motion videos into robot actions—complements this advancement, reducing programming barriers. Meanwhile, AgiBot's G2 robot, powered by NVIDIA's Jetson Thor T5000, suggests that real-time AI processing is enabling this leap forward. While Google's Intrinsic and NVIDIA's Isaac Lab have pioneered reinforcement learning frameworks, AgiBot appears to be the first to deploy RL in live production. If scalable, this could herald the adaptive factory era, where robots continuously learn, optimize and evolve—without halting operations. As factories face increasing demands for customization and rapid model changes, AgiBot's breakthrough may finally make self-learning robotics a commercial reality. Watch the video below about Chinese startup AgiBot beginning mass production of general-purpose humanoid robots. This video is from the SecureLife channel on Brighteon.com. Sources include: TheRobotReport.com BrightU.ai PRNewswire.com Ubergizmo.com Brighteon.com Traditional industrial robots rely on rigid programming, requiring extensive tuning, costly reconfiguration and custom fixtures for each task. Even advanced "vision + force-control" systems struggle with parameter sensitivity and maintenance complexity. AgiBot's RW-RL system tackles these limitations by allowing robots to learn and adapt directly on the factory floor—acquiring new skills in minutes rather than weeks while maintaining industrial-grade stability. Dr. Jianlan Luo, AgiBot's Chief Scientist, stated that their "system achieves stable, repeatable learning on real machines" closing the gap between academic research and industrial deployment. Key advantages of RW-RL AgiBot highlights three core benefits of its reinforcement learning system: Rapid Deployment – Training time slashed from weeks to minutes. High Adaptability – Robots autonomously compensate for variations like part misalignment while maintaining 100 percent task completion. Flexible Reconfiguration – Production line changes require minimal hardware adjustments, eliminating costly downtime. Unlike lab-based demonstrations, AgiBot's system was validated under near-production conditions, proving its readiness for industrial use. Reinforcement learning—where robots optimize performance through trial and error—has long been confined to research papers and controlled experiments. AgiBot's breakthrough integrates perception, decision-making and motion control into a unified loop, enabling robots to self-correct in real-time. The Longcheer pilot demonstrated RW-RL's resilience against environmental disruptions—including vibration, temperature shifts and part misalignment—while maintaining precision assembly. When production models changed, the robot retrained in minutes without manual reprogramming, showcasing unprecedented flexibility. The future of adaptive factories AgiBot and Longcheer plan to expand RW-RL into consumer electronics and automotive manufacturing, focusing on modular, plug-and-play robotic solutions that integrate seamlessly with existing systems. The company's LinkCraft platform—which converts human motion videos into robot actions—complements this advancement, reducing programming barriers. Meanwhile, AgiBot's G2 robot, powered by NVIDIA's Jetson Thor T5000, suggests that real-time AI processing is enabling this leap forward. While Google's Intrinsic and NVIDIA's Isaac Lab have pioneered reinforcement learning frameworks, AgiBot appears to be the first to deploy RL in live production. If scalable, this could herald the adaptive factory era, where robots continuously learn, optimize and evolve—without halting operations. As factories face increasing demands for customization and rapid model changes, AgiBot's breakthrough may finally make self-learning robotics a commercial reality. Watch the video below about Chinese startup AgiBot beginning mass production of general-purpose humanoid robots. This video is from the SecureLife channel on Brighteon.com. Sources include: TheRobotReport.com BrightU.ai PRNewswire.com Ubergizmo.com Brighteon.com Traditional industrial robots rely on rigid programming, requiring extensive tuning, costly reconfiguration and custom fixtures for each task. Even advanced "vision + force-control" systems struggle with parameter sensitivity and maintenance complexity. AgiBot's RW-RL system tackles these limitations by allowing robots to learn and adapt directly on the factory floor—acquiring new skills in minutes rather than weeks while maintaining industrial-grade stability. Dr. Jianlan Luo, AgiBot's Chief Scientist, stated that their "system achieves stable, repeatable learning on real machines" closing the gap between academic research and industrial deployment. Key advantages of RW-RL AgiBot highlights three core benefits of its reinforcement learning system: Rapid Deployment – Training time slashed from weeks to minutes. High Adaptability – Robots autonomously compensate for variations like part misalignment while maintaining 100 percent task completion. Flexible Reconfiguration – Production line changes require minimal hardware adjustments, eliminating costly downtime. Unlike lab-based demonstrations, AgiBot's system was validated under near-production conditions, proving its readiness for industrial use. Reinforcement learning—where robots optimize performance through trial and error—has long been confined to research papers and controlled experiments. AgiBot's breakthrough integrates perception, decision-making and motion control into a unified loop, enabling robots to self-correct in real-time. The Longcheer pilot demonstrated RW-RL's resilience against environmental disruptions—including vibration, temperature shifts and part misalignment—while maintaining precision assembly. When production models changed, the robot retrained in minutes without manual reprogramming, showcasing unprecedented flexibility. The future of adaptive factories AgiBot and Longcheer plan to expand RW-RL into consumer electronics and automotive manufacturing, focusing on modular, plug-and-play robotic solutions that integrate seamlessly with existing systems. The company's LinkCraft platform—which converts human motion videos into robot actions—complements this advancement, reducing programming barriers. Meanwhile, AgiBot's G2 robot, powered by NVIDIA's Jetson Thor T5000, suggests that real-time AI processing is enabling this leap forward. While Google's Intrinsic and NVIDIA's Isaac Lab have pioneered reinforcement learning frameworks, AgiBot appears to be the first to deploy RL in live production. If scalable, this could herald the adaptive factory era, where robots continuously learn, optimize and evolve—without halting operations. As factories face increasing demands for customization and rapid model changes, AgiBot's breakthrough may finally make self-learning robotics a commercial reality. Watch the video below about Chinese startup AgiBot beginning mass production of general-purpose humanoid robots. This video is from the SecureLife channel on Brighteon.com. Sources include: TheRobotReport.com BrightU.ai PRNewswire.com Ubergizmo.com Brighteon.com Dr. Jianlan Luo, AgiBot's Chief Scientist, stated that their "system achieves stable, repeatable learning on real machines" closing the gap between academic research and industrial deployment. Key advantages of RW-RL AgiBot highlights three core benefits of its reinforcement learning system: Rapid Deployment – Training time slashed from weeks to minutes. High Adaptability – Robots autonomously compensate for variations like part misalignment while maintaining 100 percent task completion. Flexible Reconfiguration – Production line changes require minimal hardware adjustments, eliminating costly downtime. Unlike lab-based demonstrations, AgiBot's system was validated under near-production conditions, proving its readiness for industrial use. Reinforcement learning—where robots optimize performance through trial and error—has long been confined to research papers and controlled experiments. AgiBot's breakthrough integrates perception, decision-making and motion control into a unified loop, enabling robots to self-correct in real-time. The Longcheer pilot demonstrated RW-RL's resilience against environmental disruptions—including vibration, temperature shifts and part misalignment—while maintaining precision assembly. When production models changed, the robot retrained in minutes without manual reprogramming, showcasing unprecedented flexibility. The future of adaptive factories AgiBot and Longcheer plan to expand RW-RL into consumer electronics and automotive manufacturing, focusing on modular, plug-and-play robotic solutions that integrate seamlessly with existing systems. The company's LinkCraft platform—which converts human motion videos into robot actions—complements this advancement, reducing programming barriers. Meanwhile, AgiBot's G2 robot, powered by NVIDIA's Jetson Thor T5000, suggests that real-time AI processing is enabling this leap forward. While Google's Intrinsic and NVIDIA's Isaac Lab have pioneered reinforcement learning frameworks, AgiBot appears to be the first to deploy RL in live production. If scalable, this could herald the adaptive factory era, where robots continuously learn, optimize and evolve—without halting operations. As factories face increasing demands for customization and rapid model changes, AgiBot's breakthrough may finally make self-learning robotics a commercial reality. Watch the video below about Chinese startup AgiBot beginning mass production of general-purpose humanoid robots. This video is from the SecureLife channel on Brighteon.com. Sources include: TheRobotReport.com BrightU.ai PRNewswire.com Ubergizmo.com Brighteon.com Dr. Jianlan Luo, AgiBot's Chief Scientist, stated that their "system achieves stable, repeatable learning on real machines" closing the gap between academic research and industrial deployment. Key advantages of RW-RL AgiBot highlights three core benefits of its reinforcement learning system: Rapid Deployment – Training time slashed from weeks to minutes. High Adaptability – Robots autonomously compensate for variations like part misalignment while maintaining 100 percent task completion. Flexible Reconfiguration – Production line changes require minimal hardware adjustments, eliminating costly downtime. Unlike lab-based demonstrations, AgiBot's system was validated under near-production conditions, proving its readiness for industrial use. Reinforcement learning—where robots optimize performance through trial and error—has long been confined to research papers and controlled experiments. AgiBot's breakthrough integrates perception, decision-making and motion control into a unified loop, enabling robots to self-correct in real-time. The Longcheer pilot demonstrated RW-RL's resilience against environmental disruptions—including vibration, temperature shifts and part misalignment—while maintaining precision assembly. When production models changed, the robot retrained in minutes without manual reprogramming, showcasing unprecedented flexibility. The future of adaptive factories AgiBot and Longcheer plan to expand RW-RL into consumer electronics and automotive manufacturing, focusing on modular, plug-and-play robotic solutions that integrate seamlessly with existing systems. The company's LinkCraft platform—which converts human motion videos into robot actions—complements this advancement, reducing programming barriers. Meanwhile, AgiBot's G2 robot, powered by NVIDIA's Jetson Thor T5000, suggests that real-time AI processing is enabling this leap forward. While Google's Intrinsic and NVIDIA's Isaac Lab have pioneered reinforcement learning frameworks, AgiBot appears to be the first to deploy RL in live production. If scalable, this could herald the adaptive factory era, where robots continuously learn, optimize and evolve—without halting operations. As factories face increasing demands for customization and rapid model changes, AgiBot's breakthrough may finally make self-learning robotics a commercial reality. Watch the video below about Chinese startup AgiBot beginning mass production of general-purpose humanoid robots. This video is from the SecureLife channel on Brighteon.com. Sources include: TheRobotReport.com BrightU.ai PRNewswire.com Ubergizmo.com Brighteon.com Key advantages of RW-RL AgiBot highlights three core benefits of its reinforcement learning system: Rapid Deployment – Training time slashed from weeks to minutes. High Adaptability – Robots autonomously compensate for variations like part misalignment while maintaining 100 percent task completion. Flexible Reconfiguration – Production line changes require minimal hardware adjustments, eliminating costly downtime. Unlike lab-based demonstrations, AgiBot's system was validated under near-production conditions, proving its readiness for industrial use. Reinforcement learning—where robots optimize performance through trial and error—has long been confined to research papers and controlled experiments. AgiBot's breakthrough integrates perception, decision-making and motion control into a unified loop, enabling robots to self-correct in real-time. The Longcheer pilot demonstrated RW-RL's resilience against environmental disruptions—including vibration, temperature shifts and part misalignment—while maintaining precision assembly. When production models changed, the robot retrained in minutes without manual reprogramming, showcasing unprecedented flexibility. The future of adaptive factories AgiBot and Longcheer plan to expand RW-RL into consumer electronics and automotive manufacturing, focusing on modular, plug-and-play robotic solutions that integrate seamlessly with existing systems. The company's LinkCraft platform—which converts human motion videos into robot actions—complements this advancement, reducing programming barriers. Meanwhile, AgiBot's G2 robot, powered by NVIDIA's Jetson Thor T5000, suggests that real-time AI processing is enabling this leap forward. While Google's Intrinsic and NVIDIA's Isaac Lab have pioneered reinforcement learning frameworks, AgiBot appears to be the first to deploy RL in live production. If scalable, this could herald the adaptive factory era, where robots continuously learn, optimize and evolve—without halting operations. As factories face increasing demands for customization and rapid model changes, AgiBot's breakthrough may finally make self-learning robotics a commercial reality. Watch the video below about Chinese startup AgiBot beginning mass production of general-purpose humanoid robots. This video is from the SecureLife channel on Brighteon.com. Sources include: TheRobotReport.com BrightU.ai PRNewswire.com Ubergizmo.com Brighteon.com AgiBot highlights three core benefits of its reinforcement learning system: Rapid Deployment – Training time slashed from weeks to minutes. High Adaptability – Robots autonomously compensate for variations like part misalignment while maintaining 100 percent task completion. Flexible Reconfiguration – Production line changes require minimal hardware adjustments, eliminating costly downtime. Unlike lab-based demonstrations, AgiBot's system was validated under near-production conditions, proving its readiness for industrial use. Reinforcement learning—where robots optimize performance through trial and error—has long been confined to research papers and controlled experiments. AgiBot's breakthrough integrates perception, decision-making and motion control into a unified loop, enabling robots to self-correct in real-time. The Longcheer pilot demonstrated RW-RL's resilience against environmental disruptions—including vibration, temperature shifts and part misalignment—while maintaining precision assembly. When production models changed, the robot retrained in minutes without manual reprogramming, showcasing unprecedented flexibility. The future of adaptive factories AgiBot and Longcheer plan to expand RW-RL into consumer electronics and automotive manufacturing, focusing on modular, plug-and-play robotic solutions that integrate seamlessly with existing systems. The company's LinkCraft platform—which converts human motion videos into robot actions—complements this advancement, reducing programming barriers. Meanwhile, AgiBot's G2 robot, powered by NVIDIA's Jetson Thor T5000, suggests that real-time AI processing is enabling this leap forward. While Google's Intrinsic and NVIDIA's Isaac Lab have pioneered reinforcement learning frameworks, AgiBot appears to be the first to deploy RL in live production. If scalable, this could herald the adaptive factory era, where robots continuously learn, optimize and evolve—without halting operations. As factories face increasing demands for customization and rapid model changes, AgiBot's breakthrough may finally make self-learning robotics a commercial reality. Watch the video below about Chinese startup AgiBot beginning mass production of general-purpose humanoid robots. This video is from the SecureLife channel on Brighteon.com. Sources include: TheRobotReport.com BrightU.ai PRNewswire.com Ubergizmo.com Brighteon.com Rapid Deployment – Training time slashed from weeks to minutes. High Adaptability – Robots autonomously compensate for variations like part misalignment while maintaining 100 percent task completion. Flexible Reconfiguration – Production line changes require minimal hardware adjustments, eliminating costly downtime. Unlike lab-based demonstrations, AgiBot's system was validated under near-production conditions, proving its readiness for industrial use. Reinforcement learning—where robots optimize performance through trial and error—has long been confined to research papers and controlled experiments. AgiBot's breakthrough integrates perception, decision-making and motion control into a unified loop, enabling robots to self-correct in real-time. The Longcheer pilot demonstrated RW-RL's resilience against environmental disruptions—including vibration, temperature shifts and part misalignment—while maintaining precision assembly. When production models changed, the robot retrained in minutes without manual reprogramming, showcasing unprecedented flexibility. The future of adaptive factories AgiBot and Longcheer plan to expand RW-RL into consumer electronics and automotive manufacturing, focusing on modular, plug-and-play robotic solutions that integrate seamlessly with existing systems. The company's LinkCraft platform—which converts human motion videos into robot actions—complements this advancement, reducing programming barriers. Meanwhile, AgiBot's G2 robot, powered by NVIDIA's Jetson Thor T5000, suggests that real-time AI processing is enabling this leap forward. While Google's Intrinsic and NVIDIA's Isaac Lab have pioneered reinforcement learning frameworks, AgiBot appears to be the first to deploy RL in live production. If scalable, this could herald the adaptive factory era, where robots continuously learn, optimize and evolve—without halting operations. As factories face increasing demands for customization and rapid model changes, AgiBot's breakthrough may finally make self-learning robotics a commercial reality. Watch the video below about Chinese startup AgiBot beginning mass production of general-purpose humanoid robots. This video is from the SecureLife channel on Brighteon.com. Sources include: TheRobotReport.com BrightU.ai PRNewswire.com Ubergizmo.com Brighteon.com Rapid Deployment – Training time slashed from weeks to minutes. High Adaptability – Robots autonomously compensate for variations like part misalignment while maintaining 100 percent task completion. Flexible Reconfiguration – Production line changes require minimal hardware adjustments, eliminating costly downtime. High Adaptability – Robots autonomously compensate for variations like part misalignment while maintaining 100 percent task completion. Flexible Reconfiguration – Production line changes require minimal hardware adjustments, eliminating costly downtime. Flexible Reconfiguration – Production line changes require minimal hardware adjustments, eliminating costly downtime. Unlike lab-based demonstrations, AgiBot's system was validated under near-production conditions, proving its readiness for industrial use. Reinforcement learning—where robots optimize performance through trial and error—has long been confined to research papers and controlled experiments. AgiBot's breakthrough integrates perception, decision-making and motion control into a unified loop, enabling robots to self-correct in real-time. The Longcheer pilot demonstrated RW-RL's resilience against environmental disruptions—including vibration, temperature shifts and part misalignment—while maintaining precision assembly. When production models changed, the robot retrained in minutes without manual reprogramming, showcasing unprecedented flexibility. The future of adaptive factories AgiBot and Longcheer plan to expand RW-RL into consumer electronics and automotive manufacturing, focusing on modular, plug-and-play robotic solutions that integrate seamlessly with existing systems. The company's LinkCraft platform—which converts human motion videos into robot actions—complements this advancement, reducing programming barriers. Meanwhile, AgiBot's G2 robot, powered by NVIDIA's Jetson Thor T5000, suggests that real-time AI processing is enabling this leap forward. While Google's Intrinsic and NVIDIA's Isaac Lab have pioneered reinforcement learning frameworks, AgiBot appears to be the first to deploy RL in live production. If scalable, this could herald the adaptive factory era, where robots continuously learn, optimize and evolve—without halting operations. As factories face increasing demands for customization and rapid model changes, AgiBot's breakthrough may finally make self-learning robotics a commercial reality. Watch the video below about Chinese startup AgiBot beginning mass production of general-purpose humanoid robots. This video is from the SecureLife channel on Brighteon.com. Sources include: TheRobotReport.com BrightU.ai PRNewswire.com Ubergizmo.com Brighteon.com Reinforcement learning—where robots optimize performance through trial and error—has long been confined to research papers and controlled experiments. AgiBot's breakthrough integrates perception, decision-making and motion control into a unified loop, enabling robots to self-correct in real-time. The Longcheer pilot demonstrated RW-RL's resilience against environmental disruptions—including vibration, temperature shifts and part misalignment—while maintaining precision assembly. When production models changed, the robot retrained in minutes without manual reprogramming, showcasing unprecedented flexibility. The future of adaptive factories AgiBot and Longcheer plan to expand RW-RL into consumer electronics and automotive manufacturing, focusing on modular, plug-and-play robotic solutions that integrate seamlessly with existing systems. The company's LinkCraft platform—which converts human motion videos into robot actions—complements this advancement, reducing programming barriers. Meanwhile, AgiBot's G2 robot, powered by NVIDIA's Jetson Thor T5000, suggests that real-time AI processing is enabling this leap forward. While Google's Intrinsic and NVIDIA's Isaac Lab have pioneered reinforcement learning frameworks, AgiBot appears to be the first to deploy RL in live production. If scalable, this could herald the adaptive factory era, where robots continuously learn, optimize and evolve—without halting operations. As factories face increasing demands for customization and rapid model changes, AgiBot's breakthrough may finally make self-learning robotics a commercial reality. Watch the video below about Chinese startup AgiBot beginning mass production of general-purpose humanoid robots. This video is from the SecureLife channel on Brighteon.com. Sources include: TheRobotReport.com BrightU.ai PRNewswire.com Ubergizmo.com Brighteon.com Reinforcement learning—where robots optimize performance through trial and error—has long been confined to research papers and controlled experiments. AgiBot's breakthrough integrates perception, decision-making and motion control into a unified loop, enabling robots to self-correct in real-time. The Longcheer pilot demonstrated RW-RL's resilience against environmental disruptions—including vibration, temperature shifts and part misalignment—while maintaining precision assembly. When production models changed, the robot retrained in minutes without manual reprogramming, showcasing unprecedented flexibility. The future of adaptive factories AgiBot and Longcheer plan to expand RW-RL into consumer electronics and automotive manufacturing, focusing on modular, plug-and-play robotic solutions that integrate seamlessly with existing systems. The company's LinkCraft platform—which converts human motion videos into robot actions—complements this advancement, reducing programming barriers. Meanwhile, AgiBot's G2 robot, powered by NVIDIA's Jetson Thor T5000, suggests that real-time AI processing is enabling this leap forward. While Google's Intrinsic and NVIDIA's Isaac Lab have pioneered reinforcement learning frameworks, AgiBot appears to be the first to deploy RL in live production. If scalable, this could herald the adaptive factory era, where robots continuously learn, optimize and evolve—without halting operations. As factories face increasing demands for customization and rapid model changes, AgiBot's breakthrough may finally make self-learning robotics a commercial reality. Watch the video below about Chinese startup AgiBot beginning mass production of general-purpose humanoid robots. This video is from the SecureLife channel on Brighteon.com. Sources include: TheRobotReport.com BrightU.ai PRNewswire.com Ubergizmo.com Brighteon.com The Longcheer pilot demonstrated RW-RL's resilience against environmental disruptions—including vibration, temperature shifts and part misalignment—while maintaining precision assembly. When production models changed, the robot retrained in minutes without manual reprogramming, showcasing unprecedented flexibility. The future of adaptive factories AgiBot and Longcheer plan to expand RW-RL into consumer electronics and automotive manufacturing, focusing on modular, plug-and-play robotic solutions that integrate seamlessly with existing systems. The company's LinkCraft platform—which converts human motion videos into robot actions—complements this advancement, reducing programming barriers. Meanwhile, AgiBot's G2 robot, powered by NVIDIA's Jetson Thor T5000, suggests that real-time AI processing is enabling this leap forward. While Google's Intrinsic and NVIDIA's Isaac Lab have pioneered reinforcement learning frameworks, AgiBot appears to be the first to deploy RL in live production. If scalable, this could herald the adaptive factory era, where robots continuously learn, optimize and evolve—without halting operations. As factories face increasing demands for customization and rapid model changes, AgiBot's breakthrough may finally make self-learning robotics a commercial reality. Watch the video below about Chinese startup AgiBot beginning mass production of general-purpose humanoid robots. This video is from the SecureLife channel on Brighteon.com. Sources include: TheRobotReport.com BrightU.ai PRNewswire.com Ubergizmo.com Brighteon.com The Longcheer pilot demonstrated RW-RL's resilience against environmental disruptions—including vibration, temperature shifts and part misalignment—while maintaining precision assembly. When production models changed, the robot retrained in minutes without manual reprogramming, showcasing unprecedented flexibility. The future of adaptive factories AgiBot and Longcheer plan to expand RW-RL into consumer electronics and automotive manufacturing, focusing on modular, plug-and-play robotic solutions that integrate seamlessly with existing systems. The company's LinkCraft platform—which converts human motion videos into robot actions—complements this advancement, reducing programming barriers. Meanwhile, AgiBot's G2 robot, powered by NVIDIA's Jetson Thor T5000, suggests that real-time AI processing is enabling this leap forward. While Google's Intrinsic and NVIDIA's Isaac Lab have pioneered reinforcement learning frameworks, AgiBot appears to be the first to deploy RL in live production. If scalable, this could herald the adaptive factory era, where robots continuously learn, optimize and evolve—without halting operations. As factories face increasing demands for customization and rapid model changes, AgiBot's breakthrough may finally make self-learning robotics a commercial reality. Watch the video below about Chinese startup AgiBot beginning mass production of general-purpose humanoid robots. This video is from the SecureLife channel on Brighteon.com. Sources include: TheRobotReport.com BrightU.ai PRNewswire.com Ubergizmo.com Brighteon.com The future of adaptive factories AgiBot and Longcheer plan to expand RW-RL into consumer electronics and automotive manufacturing, focusing on modular, plug-and-play robotic solutions that integrate seamlessly with existing systems. The company's LinkCraft platform—which converts human motion videos into robot actions—complements this advancement, reducing programming barriers. Meanwhile, AgiBot's G2 robot, powered by NVIDIA's Jetson Thor T5000, suggests that real-time AI processing is enabling this leap forward. While Google's Intrinsic and NVIDIA's Isaac Lab have pioneered reinforcement learning frameworks, AgiBot appears to be the first to deploy RL in live production. If scalable, this could herald the adaptive factory era, where robots continuously learn, optimize and evolve—without halting operations. As factories face increasing demands for customization and rapid model changes, AgiBot's breakthrough may finally make self-learning robotics a commercial reality. Watch the video below about Chinese startup AgiBot beginning mass production of general-purpose humanoid robots. This video is from the SecureLife channel on Brighteon.com. Sources include: TheRobotReport.com BrightU.ai PRNewswire.com Ubergizmo.com Brighteon.com AgiBot and Longcheer plan to expand RW-RL into consumer electronics and automotive manufacturing, focusing on modular, plug-and-play robotic solutions that integrate seamlessly with existing systems. The company's LinkCraft platform—which converts human motion videos into robot actions—complements this advancement, reducing programming barriers. Meanwhile, AgiBot's G2 robot, powered by NVIDIA's Jetson Thor T5000, suggests that real-time AI processing is enabling this leap forward. While Google's Intrinsic and NVIDIA's Isaac Lab have pioneered reinforcement learning frameworks, AgiBot appears to be the first to deploy RL in live production. If scalable, this could herald the adaptive factory era, where robots continuously learn, optimize and evolve—without halting operations. As factories face increasing demands for customization and rapid model changes, AgiBot's breakthrough may finally make self-learning robotics a commercial reality. Watch the video below about Chinese startup AgiBot beginning mass production of general-purpose humanoid robots. This video is from the SecureLife channel on Brighteon.com. Sources include: TheRobotReport.com BrightU.ai PRNewswire.com Ubergizmo.com Brighteon.com The company's LinkCraft platform—which converts human motion videos into robot actions—complements this advancement, reducing programming barriers. Meanwhile, AgiBot's G2 robot, powered by NVIDIA's Jetson Thor T5000, suggests that real-time AI processing is enabling this leap forward. While Google's Intrinsic and NVIDIA's Isaac Lab have pioneered reinforcement learning frameworks, AgiBot appears to be the first to deploy RL in live production. If scalable, this could herald the adaptive factory era, where robots continuously learn, optimize and evolve—without halting operations. As factories face increasing demands for customization and rapid model changes, AgiBot's breakthrough may finally make self-learning robotics a commercial reality. Watch the video below about Chinese startup AgiBot beginning mass production of general-purpose humanoid robots. This video is from the SecureLife channel on Brighteon.com. Sources include: TheRobotReport.com BrightU.ai PRNewswire.com Ubergizmo.com Brighteon.com The company's LinkCraft platform—which converts human motion videos into robot actions—complements this advancement, reducing programming barriers. Meanwhile, AgiBot's G2 robot, powered by NVIDIA's Jetson Thor T5000, suggests that real-time AI processing is enabling this leap forward. While Google's Intrinsic and NVIDIA's Isaac Lab have pioneered reinforcement learning frameworks, AgiBot appears to be the first to deploy RL in live production. If scalable, this could herald the adaptive factory era, where robots continuously learn, optimize and evolve—without halting operations. As factories face increasing demands for customization and rapid model changes, AgiBot's breakthrough may finally make self-learning robotics a commercial reality. Watch the video below about Chinese startup AgiBot beginning mass production of general-purpose humanoid robots. This video is from the SecureLife channel on Brighteon.com. Sources include: TheRobotReport.com BrightU.ai PRNewswire.com Ubergizmo.com Brighteon.com While Google's Intrinsic and NVIDIA's Isaac Lab have pioneered reinforcement learning frameworks, AgiBot appears to be the first to deploy RL in live production. If scalable, this could herald the adaptive factory era, where robots continuously learn, optimize and evolve—without halting operations. As factories face increasing demands for customization and rapid model changes, AgiBot's breakthrough may finally make self-learning robotics a commercial reality. Watch the video below about Chinese startup AgiBot beginning mass production of general-purpose humanoid robots. This video is from the SecureLife channel on Brighteon.com. Sources include: TheRobotReport.com BrightU.ai PRNewswire.com Ubergizmo.com Brighteon.com While Google's Intrinsic and NVIDIA's Isaac Lab have pioneered reinforcement learning frameworks, AgiBot appears to be the first to deploy RL in live production. If scalable, this could herald the adaptive factory era, where robots continuously learn, optimize and evolve—without halting operations. As factories face increasing demands for customization and rapid model changes, AgiBot's breakthrough may finally make self-learning robotics a commercial reality. Watch the video below about Chinese startup AgiBot beginning mass production of general-purpose humanoid robots. This video is from the SecureLife channel on Brighteon.com. Sources include: TheRobotReport.com BrightU.ai PRNewswire.com Ubergizmo.com Brighteon.com As factories face increasing demands for customization and rapid model changes, AgiBot's breakthrough may finally make self-learning robotics a commercial reality. Watch the video below about Chinese startup AgiBot beginning mass production of general-purpose humanoid robots. This video is from the SecureLife channel on Brighteon.com. Sources include: TheRobotReport.com BrightU.ai PRNewswire.com Ubergizmo.com Brighteon.com As factories face increasing demands for customization and rapid model changes, AgiBot's breakthrough may finally make self-learning robotics a commercial reality. Watch the video below about Chinese startup AgiBot beginning mass production of general-purpose humanoid robots. This video is from the SecureLife channel on Brighteon.com. Sources include: TheRobotReport.com BrightU.ai PRNewswire.com Ubergizmo.com Brighteon.com Watch the video below about Chinese startup AgiBot beginning mass production of general-purpose humanoid robots. This video is from the SecureLife channel on Brighteon.com. Sources include: TheRobotReport.com BrightU.ai PRNewswire.com Ubergizmo.com Brighteon.com Watch the video below about Chinese startup AgiBot beginning mass production of general-purpose humanoid robots. This video is from the SecureLife channel on Brighteon.com. Sources include: TheRobotReport.com BrightU.ai PRNewswire.com Ubergizmo.com Brighteon.com This video is from the SecureLife channel on Brighteon.com. Sources include: TheRobotReport.com BrightU.ai PRNewswire.com Ubergizmo.com Brighteon.com This video is from the SecureLife channel on Brighteon.com. Sources include: TheRobotReport.com BrightU.ai PRNewswire.com Ubergizmo.com Brighteon.com This video is from the SecureLife channel on Brighteon.com. Sources include: TheRobotReport.com BrightU.ai PRNewswire.com Ubergizmo.com Brighteon.com This video is from the SecureLife channel on Brighteon.com. Sources include: TheRobotReport.com BrightU.ai PRNewswire.com Ubergizmo.com Brighteon.com Sources include: TheRobotReport.com BrightU.ai PRNewswire.com Ubergizmo.com Brighteon.com Sources include: TheRobotReport.com BrightU.ai PRNewswire.com Ubergizmo.com Brighteon.com TheRobotReport.com BrightU.ai PRNewswire.com Ubergizmo.com Brighteon.com TheRobotReport.com BrightU.ai PRNewswire.com Ubergizmo.com Brighteon.com BrightU.ai PRNewswire.com Ubergizmo.com Brighteon.com BrightU.ai PRNewswire.com Ubergizmo.com Brighteon.com PRNewswire.com Ubergizmo.com Brighteon.com PRNewswire.com Ubergizmo.com Brighteon.com Ubergizmo.com Brighteon.com Ubergizmo.com Brighteon.com Brighteon.com Brighteon.com This site is part of the Natural News Network © 2022 All Rights Reserved. Privacy | Terms All content posted on this site is commentary or opinion and is protected under Free Speech. Truth Publishing International, LTD. is not responsible for content written by contributing authors. The information on this site is provided for educational and entertainment purposes only. It is not intended as a substitute for professional advice of any kind. Truth Publishing assumes no responsibility for the use or misuse of this material. Your use of this website indicates your agreement to these terms and those published here. All trademarks, registered trademarks and servicemarks mentioned on this site are the property of their respective owners. Backup IP: http://45.89.97.6

SHANGHAI—In a bustling factory on the outskirts of Shanghai, a new breed of robot is quietly revolutionizing manufacturing. AgiBot, a Chinese startup founded by former Huawei engineer Peng Zhihui, is deploying AI-powered robots that learn tasks through reinforcement learning, marking a significant shift in industrial automation. This technology allows machines to adapt and improve on the fly, potentially transforming labor-intensive industries. According to a recent article in WIRED, AgiBot employs a unique training method combining AI algorithms with human teleoperation. Workers remotely control robots to perform tasks, generating data that trains AI models. This approach has enabled AgiBot to achieve what many consider a breakthrough: the first real-world deployment of reinforcement learning in industrial settings. From Huawei to Robotics Pioneer Peng Zhihui, a graduate of the University of Electronic Science and Technology of China, gained fame for inventions like a self-driving bicycle and an Iron Man-inspired robotic arm, as detailed on Wikipedia. He joined Huawei in 2020, earning a high salary, but left in December 2022 to launch AgiBot in February 2023. Backed by investors including HongShan, Hillhouse Investment, and BYD, the company quickly established a manufacturing facility in Shanghai by January 2024. By August 2024, AgiBot’s factory began deliveries, shipping 200 bipedal and 100 wheeled robots by year’s end, according to reports from The Paper. This rapid scaling underscores China’s aggressive push in AI-robotics integration, with AgiBot emerging as a key player. Mass Production Milestones Recent posts on X highlight AgiBot’s mass production of general-purpose humanoid robots for factories, labs, and homes. One post from user tphuang notes the production of wheeled robots with adjustable bodies, emphasizing data collection through showrooms. Another from The Humanoid Hub describes the startup’s progress since 2023, including mass production of 1,000 units earlier this year and plans for 5,000 more. Disclose.tv on X reported the beginning of mass production for humanoid robots, while RT mentioned launches for warehouse and store applications. These developments align with AgiBot’s official site, which lists products like AgiBot A2, X/D1, Genie, and C5, designed for diverse tasks. Breakthrough in Reinforcement Learning AgiBot achieved a historic milestone with the first real-world deployment of reinforcement learning in industrial robotics, as announced in a press release covered by RoboticsTomorrow. Jianlan Luo, AgiBot’s Chief Scientist, explained that the system integrates advanced algorithms with hardware, enabling stable learning on physical robots. This was demonstrated on a pilot production line with Longcheer Technology. The collaboration plans to expand to precision manufacturing in consumer electronics and automotive components, focusing on modular, deployable solutions. According to The Robot Report, robots can learn new skills in minutes on the factory floor, bridging academic research and industrial application. Unleashing Massive Datasets In December 2024, AgiBot released the largest humanoid manipulation dataset, AgiBot World, with over 1 million trajectories from 100 robots, as reported by PR Newswire and The Robot Report. This dataset enables large-scale learning, paving the way for general-purpose robots in everyday life. The diversity and complexity of the data support training for tasks like precision nailing, as shown in WIRED’s coverage. Schneider from WIRED notes that other companies are exploring similar reinforcement learning for manufacturing, but AgiBot’s approach stands out in China’s booming AI-robotics scene. Deployments and Partnerships AgiBot secured a deal to deploy 100 robots at car parts factories, with A2-W models meeting monthly production targets in a single shift, according to the South China Morning Post. Additionally, CyberRobo on X mentioned deployments in commercial scenarios like shopping malls and auto dealerships, with mass production scaling to 5,000 units. Recent news from Ubergizmo and Gizmochina emphasizes AgiBot’s reinforcement learning deployment, allowing robots to learn tasks in minutes rather than weeks. This self-learning capability could redefine manufacturing, as Eyisha Zyer’s X post suggests. Global Implications and US Comparisons WIRED highlights that AgiBot’s technology may be crucial for US companies aiming to reshore manufacturing. US startups like Physical Intelligence and Skild are developing similar robo-learning algorithms, spun out from UC Berkeley and Carnegie Mellon research. However, AgiBot’s integration of AI models for humanoids and fixed robot arms positions it ahead in practical deployments. A post from EyeingAI on X calls this the moment embodied AI powers the real world, while Alif Hossain notes robots learning tasks in minutes without human intervention after initial training. Innovative Platforms and Future Visions AgiBot unveiled LinkCraft, a robot content creation platform allowing users to upload human movement videos for humanoid mimicry, as reported by The Information. This tool democratizes robot programming, requiring no expertise. At IROS 2025, AgiBot debuted and concluded the AgiBot World Challenge, showcasing advancements, per PR Newswire. WIRED Science on X notes how smarter machines could transform physical labor in China, with AgiBot leading the charge. Challenges and Industry Sentiment Despite successes, scaling reinforcement learning poses challenges, including data quality and hardware integration. Industrial Automation Magazine on X reported on the deployment, emphasizing its groundbreaking nature. Current sentiment on X, from users like tphuang, questions if the market will be dominated by one player like DJI in drones. AgiBot’s factory aims for robots cheaper than family cars—under $20,000—at scale, potentially accelerating adoption. Pushing Boundaries in AI Robotics AgiBot’s blend of reinforcement learning and human-assisted training addresses longstanding robotics hurdles, enabling adaptability in dynamic environments. As WIRED describes, a ‘small army of workers’ aids in data generation, combining human intuition with AI efficiency. Looking ahead, expansions with Longcheer signal broader applications, from electronics to automotive. This positions AgiBot at the forefront of a manufacturing renaissance, where AI not only automates but evolves with the tasks at hand. Subscribe for Updates Help us improve our content by reporting any issues you find. Get the free daily newsletter read by decision makers Get our media kit Deliver your marketing message directly to decision makers.

Geeky Gadgets The Latest Technology News 10:14 am August 29, 2025 By Julian Horsey What if robots could learn to adapt to their surroundings as effortlessly as humans do? The rise of quadruped robots, like Boston Dynamics’ Spot, is turning this vision into reality. By integrating reinforcement learning (RL)—a innovative machine learning technique, these robots are not only mastering practical tasks like industrial inspections but also pushing the boundaries of agility and resilience. Imagine a robot navigating a hazardous construction site, climbing uneven stairs, or recovering gracefully after a slip, all without human intervention. This isn’t science fiction; it’s the result of advanced programming and iterative development that’s redefining what robotics can achieve in real-world applications. Boston Dynamics provide more insights into the fantastic role of RL in enhancing the behavior and performance of quadruped robots. From the meticulous programming that balances practicality with high-performance maneuvers to the use of simulation environments for perfecting their adaptability, we’ll uncover how Spot is evolving into a versatile tool for industries and beyond. You’ll discover how hardware optimization, robustness testing, and iterative debugging contribute to Spot’s reliability in unpredictable conditions. By the end, you’ll see how these advancements are not just improving robotics but reshaping the future of automation itself, one step, leap, or backflip at a time. TL;DR Key Takeaways : Spot’s programming is carefully designed to address two primary objectives: executing practical tasks for real-world applications and performing extreme maneuvers that push its operational boundaries. This dual focus ensures that Spot is both a reliable tool for everyday tasks and a robust performer in extreme conditions. The development of Spot relies heavily on simulation modeling and hardware optimization, which together create a smarter and more efficient robot. By combining advanced simulation techniques with precise hardware refinement, Spot achieves seamless integration between its software intelligence and physical capabilities. Watch this video on YouTube. Explore further guides and articles from our vast library that you may find relevant to your interests in humanoid robots. Spot’s reliability is the result of a rigorous development process that emphasizes iterative debugging and comprehensive robustness testing. This meticulous process guarantees that Spot is prepared to handle the diverse demands of real-world applications, making it a dependable asset in various industries. Spot’s versatility and adaptability make it a valuable resource across a wide range of industrial sectors, where precision and reliability are paramount. Spot’s ability to adapt to diverse scenarios underscores its potential to transform workflows, improve safety, and enhance productivity across multiple domains. The ongoing advancements in reinforcement learning and hardware engineering are paving the way for a new era in quadruped robotics. As these technologies continue to evolve, robots like Spot are expected to become even more capable, addressing challenges that were once considered insurmountable. From conducting industrial inspections to participating in creative and innovative projects, the potential applications for quadruped robots are vast and varied. By combining innovative machine learning techniques with robust engineering, quadruped robots are poised to play a fantastic role across industries. Their ability to perform complex tasks with precision, adapt to changing conditions, and operate in challenging environments positions them as key contributors to the future of automation and innovation. As these machines become more advanced, they will redefine what is possible in both practical and creative domains, ushering in a new era of technological progress. Media Credit: Boston Dynamics Latest Geeky Gadgets Deals Disclosure: Some of our articles include affiliate links. If you buy something through one of these links, Geeky Gadgets may earn an affiliate commission. Learn about our Disclosure Policy.

A study published on arXiv details how researchers at the University of Bonn have developed a reinforcement learning framework that enables robots to manipulate granular media such as sand into target shapes. The system trains a robotic arm with a cubic end-effector and a stereo camera to reshape loose material into forms including rectangles, L-shapes, polygons, and negatives of archaeological fresco fragments. Experiments showed millimeter-level accuracy, with the trained agent outperforming two baseline approaches and transferring successfully from simulation to a physical robot without additional training. Granular materials pose difficulties for robotics because of their high-dimensional configuration space and unstable dynamics. Rule-based approaches often fail, while particle simulations are computationally expensive. Researchers addressed these challenges by designing compact observation spaces and reward functions that guided learning. Visual policies were trained using Truncated Quantile Critics (TQC), an off-policy reinforcement learning algorithm. Depth images from a ZED 2i stereo camera were converted into height maps, allowing the robot to compare current and goal structures in a form suitable for efficient training. The system was evaluated against a random policy and a Boustrophedon Coverage Path Planning baseline. Across 400 goal shapes, the learned agent consistently outperformed both methods. Using the delta reward (DELTA) formulation, the robot achieved a mean height difference of 3.4 millimeters compared with 4.8 millimeters for the planning method and 7.2 millimeters for random motion. Execution time was shorter as well, averaging 23.5 steps versus 44 for the path planning baseline. The agent also modified 97 percent of relevant cells in the goal area, compared with 54 percent for random motion. Execution steps were defined as the number of actions until the end-effector left the granular medium for three consecutive steps. Statistical testing confirmed that the DELTA policy significantly outperformed all alternatives. The project involved the Humanoid Robots Lab, the Autonomous Intelligent Systems Lab, and the Center for Robotics at the University of Bonn, working with the Lamarr Institute for Machine Learning and Artificial Intelligence. Funding came from the European Commission’s RePAIR program under Horizon 2020 and from Germany’s Federal Ministry of Education and Research through the Robotics Institute Germany initiative. Further experiments examined design choices. When the goal-area movement reward was removed, agents avoided manipulation behaviors entirely, performing no better than random baselines. Feature extractor ablations showed that the proposed gating-based encoder achieved the best performance, with an average error of 3.4 millimeters compared with 4.6 millimeters when relying directly on depth images. Algorithm comparisons confirmed that TQC achieved stable convergence, whereas Soft Actor-Critic lagged and Twin Delayed Deep Deterministic Policy Gradient failed to converge. A supplementary site linked in the paper provides additional details, videos, and code. Deployment on a UR5e robotic arm validated the approach outside simulation. Despite sensor noise and an uneven starting surface, the robot reproduced target shapes such as rectangles with results similar to those seen in simulation. The ability to transfer directly from synthetic training environments to real-world execution demonstrated the robustness of the framework. Research into granular media manipulation spans excavation, grading, and extraterrestrial soil handling. Many approaches depend on computationally demanding finite or discrete element simulations or on imitation learning pipelines tailored to specific tasks. By combining efficient height map representations with carefully designed reward formulations, the Bonn team demonstrated that reinforcement learning can adaptively shape granular media without handcrafted rules.The authors conclude that their method consistently outperforms traditional baselines and establishes a viable route for adaptive robotic manipulation of deformable materials. Limited spaces remain for AMA:Energy 2025. Register now to join the conversation on the future of energy and additive manufacturing. Ready to discover who won the 2024 3D Printing Industry Awards? Subscribe to the 3D Printing Industry newsletter and follow us on LinkedIn to stay updated with the latest news and insights. Featured image shows a training process is employed to enable agents to manipulate granular media using sensory inputs. Image via University of Bonn. Anyer Tenorio Lara is an emerging tech journalist passionate about uncovering the latest advances in technology and innovation. With a sharp eye for detail and a talent for storytelling, Anyer has quickly made a name for himself in the tech community. Anyer's articles aim to make complex subjects accessible and engaging for a broad audience. In addition to his writing, Anyer enjoys participating in industry events and discussions, eager to learn and share knowledge in the dynamic world of technology. © Copyright 2017 | All Rights Reserved | 3D Printing Industry

In this thesis , the use of reinforcement learning for controlling Cable-driven parallel robots has been investigated. This class of robots is known with its complex dynamics and the nonlinearity of the system, which offers an interesting environment for the implementation of reinforcement learning algorithms. These algorithms requires a lot of data to learn the optimal policy, which is not always possible in real world applications. To overcome this issue, we propose a sim-to-real approach. First, the newton euler equation of the robot is used to derive the dynamic model of the robot, and by setting the parameters to the real values of the robot, we validated the model by comparing the simulation results with the real data. To ensure a high precision of the simulation data and a reduced execution time, it was implemented using Matlab/Simulink and then converted to exttt{C++} library for easier integration with the extit{gym} environment in exttt{python}. Additionally, in order to learn the optimal policy using reinforcement learning, the objective of the controller should be specified. As most of the use cases of cable-driven parallel robots could be summarized as a trajectory tracking problem, a reward function that align with this objective is designed and a process for target trajectories generation was developed. In addition to that, a limitation on the action space was introduced to ensure the cable tension is within the limits during the training process. These key components along with the most known reinforcement learning algorithms for continuous space: DDPG, PPO, and SAC make a full-fledged training platform for the generation of reinforcement learning based controllers for cable-driven parallel robots. A comparison between the three algorithms during the training process and the performance of the trained controllers was conducted. A side by side evaluation of the reinforcement learning controller with a PID based controller developed for insect tracking purposes was also performed and many aspects were compared such as the tracking error, the energy consumption, and the robustness of the controller. One of the main challenges of this work is the transition to different configuration of the robot, as the trained policy was developed for a specific configuration, a new training process is required for each different configuration. To overcome this issue, a new method to learn an actuator level policy has been developed and comparative analysis with the conventional policy has been hold. Finally, the trained controller was tested on the robot to ensure the transferability of the policy from the simulation to the real world. Dans cette thèse, l'apprentissage par renforcement appliqué au contrôle des robots parallèles à câbles a été étudié. Cette catégorie de robots se distingue par sa dynamique complexe et la non-linéarité de son système, offrant ainsi un cadre idéal pour l'implémentation d'algorithmes d'apprentissage par renforcement. Toutefois, ces algorithmes nécessitent de vastes quantités de données pour apprendre la politique optimale, ce qui n'est pas toujours réalisable dans des scénarios réels. Pour contourner cette limitation, nous avons proposé une approche sim-to-real. Tout d'abord, l'équation de Newton-Euler a été utilisée pour modéliser la dynamique du robot, et en fixant les paramètres aux valeurs réelles, le modèle a été validé en comparant les résultats de la simulation avec les données expérimentales. Afin d'assurer une grande précision des simulations tout en réduisant les temps de calcul, le modèle a été implémenté sous Matlab/Simulink, puis converti en bibliothèque exttt{C++} pour une intégration plus fluide avec l'environnement extit{gym} en exttt{python}. Par ailleurs, pour déterminer la politique optimale via l'apprentissage par renforcement, l'objectif du contrôleur doit être défini. Étant donné que la majorité des applications des robots parallèles à câbles se rapportent au suivi de trajectoires, une fonction de récompense alignée sur cet objectif a été conçue, accompagnée d'un processus de génération de trajectoires cibles. De plus, une limitation de l'espace d'action a été introduite afin de garantir que la tension des câbles reste dans les limites acceptables durant l'apprentissage. Ces éléments clés, associés aux algorithmes d'apprentissage par renforcement les plus répandus pour les espaces continus - DDPG, PPO et SAC - forment une plateforme complète dédiée à la génération de contrôleurs pour les robots parallèles à câbles. Une comparaison approfondie entre ces trois algorithmes a été réalisée, tant durant l'apprentissage que lors de l'évaluation des performances des contrôleurs entraînés. En parallèle, une comparaison a été effectuée entre le contrôleur d'apprentissage par renforcement et un contrôleur PID développé pour le suivi des insectes, en prenant en compte divers critères tels que l'erreur de suivi, la consommation d'énergie et la robustesse du système. Un des défis majeurs de cette étude concerne la transition vers différentes configurations du robot, car la politique apprise est spécifique à une configuration donnée, nécessitant un nouveau processus d'apprentissage pour chaque configuration différente. Pour pallier cette difficulté, une nouvelle méthode d'apprentissage d'une politique au niveau des actionneurs a été développée et comparée à la méthode conventionnelle. Enfin, le contrôleur entraîné a été testé sur le robot afin de valider la transférabilité de la politique de la simulation au monde réel. Connectez-vous pour contacter le contributeur https://hal.univ-lorraine.fr/tel-05072595 Soumis le : mercredi 10 septembre 2025-11:36:36 Dernière modification le : mercredi 5 novembre 2025-03:24:06 Archivage à long terme le : jeudi 11 décembre 2025-18:23:51 Contact Ressources Informations Questions juridiques Portails CCSD

In the rapidly evolving field of robotics, a groundbreaking advancement is reshaping how machines learn to interact with the world. Researchers have unveiled a new system that integrates reinforcement learning with advanced robotic vision, enabling robots to master complex manipulation tasks with remarkably less reliance on human-provided demonstration data. This innovation not only accelerates learning but also allows robots to innovate beyond their initial training, discovering more efficient movement patterns that humans might not have anticipated. At the core of this breakthrough is a sophisticated algorithm that rewards robots for successful actions while penalizing failures, all processed through visual inputs. By simulating trial-and-error in real-time, the system scales up vision-action skills, turning raw pixel data into precise motor commands. This approach marks a significant leap from traditional methods that demand extensive pre-programmed examples, potentially revolutionizing industries from manufacturing to healthcare. Unlocking Autonomous Discovery in Robotics Recent reports highlight how this technology empowers robots to explore uncharted territories of motion. For instance, in tasks like grasping irregular objects or navigating cluttered environments, the system doesn’t just mimic; it evolves. According to an article from Quantum Zeitgeist, the integration of reinforcement learning with vision allows for “learning complex manipulation tasks with less human demonstration data and even discovers new, more efficient movement patterns beyond those it was initially taught.” This self-improvement capability is akin to how animals adapt in the wild, but engineered for mechanical precision. Industry insiders note that such advancements address long-standing bottlenecks in robotics, where data scarcity has hindered scalability. By minimizing the need for human oversight, this method could democratize robot deployment in small-scale operations, from warehouse automation to personalized assistive devices. Insights from Recent Academic and Industry Developments Building on this, a study published in the International Journal of Robotics Research, as detailed in a 2021 review by Tengteng Zhang and Hongwei Mo in SAGE Journals, underscores the potential of reinforcement learning to endow robots with “humanoid perception and decision-making wisdom.” Fast-forward to today, and we’re seeing practical applications emerge. A recent piece from TechXplore describes work at UC Berkeley where AI-driven robots learn tasks faster with human feedback, stacking Jenga blocks with a single limb—demonstrating how vision-guided reinforcement learning handles delicate, real-world interactions. Moreover, posts on X (formerly Twitter) from robotics experts like Russell Mendonca reveal ongoing excitement, with one noting that reinforcement learning enables robots to “learn skills via real-world practice, without any demonstrations or simulation engineering,” using language and vision models for rewards. This sentiment echoes broader innovations, such as Google DeepMind’s framework for coordinating multiple robot arms without collisions, as reported in Science Robotics, where up to 40 tasks run simultaneously in crowded spaces. Challenges and Future Implications for Scalability Yet, scaling these vision-action skills isn’t without hurdles. Training in dynamic environments requires immense computational power, and ensuring safety in unpredictable settings remains a priority. As outlined in a 2018 paper from Proceedings of Machine Learning Research on scalable deep reinforcement learning for vision-based manipulation, the key lies in balancing exploration with exploitation to avoid catastrophic failures during learning. For industry leaders, this breakthrough signals a shift toward more adaptive systems. Imagine assembly lines where robots self-optimize workflows, reducing downtime and costs. A Neuroscience News article from two weeks ago highlights robots integrating sight and touch for human-like object handling, further amplified by reinforcement learning’s trial-and-error ethos. Bridging Theory to Real-World Deployment Experts predict this will accelerate adoption in sectors like logistics and elder care. A post on X by AK discusses “RoboGen,” a generative simulation approach for learning diverse skills at scale, pointing to infinite data generation as a game-changer. Similarly, a DVIDS news release from six days ago reports the U.S. Naval Research Laboratory’s successful reinforcement learning control of a free-flyer in space, extending these principles beyond Earth. As these technologies mature, ethical considerations loom—ensuring equitable access and mitigating job displacement. Still, the fusion of vision and action through reinforcement learning promises a future where robots aren’t just tools, but intelligent partners, continually evolving to meet human needs. With ongoing research from institutions like Carnegie Mellon University, as referenced in their 2013 publication, the trajectory is clear: robotics is entering an era of unprecedented autonomy. Subscribe for Updates Help us improve our content by reporting any issues you find. Get the free daily newsletter read by decision makers Get our media kit Deliver your marketing message directly to decision makers.

TAIPEI, Aug. 25, 2025 /PRNewswire/ — Advantech, a global leader in industrial edge computing, is proud to announce the launch of the Advantech MIC-742-AT Robotics Development Kit accelerated by NVIDIA Jetson Thor modules with NVIDIA Holoscan platform support. Advantech MIC-742-AT Robotics Development Kit Designed for next-generation robotics and physical AI applications, the platform combines NVIDIA Jetson Thor’s ultra-high AI compute power with built-in support for the NVIDIA Holoscan platform—enabling robots to sense, understand, and act with real-time reasoning and ultra low latency sensor processing at the edge. Next-Level AI Compute for Robotics Accelerated by the NVIDIA Jetson Thor module, the Advantech MIC-742-AT delivers up to 2,070 FP4 TFLOPS AI performance with 128GB high-bandwidth LPDDR5X memory. Designed for industrial-grade reliability, it offers a total system power consumption of just 150W, wide operating temperature support from -10°C to 60°C, and a compact form factor for space-constrained deployments. The platform also features 8-channel GMSL 2.0 support, enabling high-speed, low-latency sensor connectivity for advanced perception systems. These capabilities make it ideal for humanoid robots, autonomous mobile robots (AMRs), surgical robotics, and other high-performance edge AI applications. Future-Ready Sensor Fusion and Integration Aligning with the industry trend toward sensor standardization and Ethernet packet-based transmission, Advantech is adopting NVIDIA Holoscan platform with NVIDIA GPUDirect RDMA, allowing sensor data to stream directly into GPU memory—bypassing CPU bottlenecks and reducing latency—to accelerate AI inference and real-time decision-making. For developers requiring real-time multimodal sensor synchronization, Advantech offers optional MIC-FG-HSBA1 powered by Holoscan Sensor Bridge and EKI-2712X-SPE switch accessories. These enable low-latency, high bandwidth, and sub-millisecond alignment of data from multiple sensors—such as stereo cameras, Lidar, radar, ultrasound, and IMUs—ensuring AI models operate on precisely timed, multimodal data for more reliable perception and decision-making in robotics. Availability The Advantech MIC-742-AT Jetson Thor Robotics Development Kit will officially launch in September 2025. For full specifications, please visit the product page at Advantech website. Advantech Holoscan Sensor Bridge (MIC-FG-HSBA1) and Holoscan Switch (EKI-2712X-SPE) accessories will be offered as add-on modules for advanced robotics applications. About Advantech: Advantech’s corporate vision is to enable an intelligent planet. The company is a global leader in the fields of IoT intelligent systems and embedded platforms. To embrace the trends of IoT, big data, and artificial intelligence, Advantech promotes IoT hardware and software solutions with the Edge Intelligence WISE-PaaS core to assist business partners and clients in connecting their industrial chains. Advantech is also working with business partners to co-create business ecosystems that accelerate the goal of industrial intelligence. (www.advantech.com) ​ Your email address will not be published. Required fields are marked * Comment * Name * Email * Website Save my name, email, and website in this browser for the next time I comment. Copyright © 2024 Money Compass Media (M) Sdn Bhd. All Rights Reserved Login to your account below Remember Me Please enter your username or email address to reset your password. Copyright © 2024 Money Compass Media (M) Sdn Bhd. All Rights Reserved

Get daily business news from the region delivered straight to your inbox The lab will integrate NVIDIA’s accelerated computing platforms with TII’s multidisciplinary research in AI, robotics, autonomous systems, and high-performance computing The Technology Innovation Institute (TII), the applied research arm of Abu Dhabi’s Advanced Technology Research Council (ATRC), and NVIDIA, the global leader in accelerated computing and artificial intelligence (AI), have launched the Middle East’s first joint laboratory dedicated to AI and robotics. The TII-NVAITC (NVIDIA AI Technology Centre) Joint Lab for AI and Robotics aims to develop next-generation AI models, robotics platforms, and humanoid technologies, with the goal of accelerating innovation across multiple industries. The agreement was signed at TII’s headquarters in Abu Dhabi. “This collaboration with NVIDIA marks a major step toward building AI-enhanced robotic systems capable of reasoning, adapting, and acting in complex environments,” Dr Najwa Aaraj, CEO of TII. She added that combining TII’s robotic platforms with AI models and accelerated computing would accelerate the convergence of perception, control, and language, laying the foundation for intelligent machines. The lab will integrate NVIDIA’s accelerated computing platforms with TII’s multidisciplinary research in AI, robotics, autonomous systems, and high-performance computing. It will be the first NVIDIA AI Technology Centre lab in the Middle East, with research spanning robotic learning and control at scale, large language models including TII’s Falcon AI models, and hardware for real-time robotic systems. Carlo Ruiz, VP – Enterprise Solutions & Operations EMEA at NVIDIA, said the lab expands the scope of NVIDIA’s global AI Technology Centre network into robotics for the Middle East, helping researchers accelerate breakthroughs in intelligent systems. The initiative aligns with Abu Dhabi’s long-term strategy to advance technological sovereignty and the UAE’s wider ambition to establish itself as a global AI and robotics hub. The lab will also support TII’s open innovation strategy, including joint research, open-source initiatives, and cross-network learning through the global NVAITC community. TII’s existing modular robotic platforms, including robotic arms and delivery robots, will provide a foundation for research focused on technical excellence and practical readiness. © 2021 MOTIVATE MEDIA GROUP. ALL RIGHTS RESERVED.

News Summary: SANTA CLARA, Calif., Aug. 25, 2025 (GLOBE NEWSWIRE) -- NVIDIA today announced the general availability of the NVIDIA Jetson AGX Thor™ developer kit and production modules, powerful new robotics computers designed to power millions of robots across industries including manufacturing, logistics, transportation, healthcare, agriculture and retail. Early adopters include industry leaders Agility Robotics, Amazon Robotics, Boston Dynamics, Caterpillar, Figure, Hexagon, Medtronic and Meta, while 1X, John Deere, OpenAI and Physical Intelligence are evaluating Jetson Thor to advance their physical AI capabilities. “We’ve built Jetson Thor for the millions of developers working on robotic systems that interact with and increasingly shape the physical world,” said Jensen Huang, founder and CEO of NVIDIA. “With unmatched performance and energy efficiency, and the ability to run multiple generative AI models at the edge, Jetson Thor is the ultimate supercomputer to drive the age of physical AI and general robotics.” The Ultimate Platform for Next-Generation Robotics Powered by an NVIDIA Blackwell GPU and featuring 128GB of memory, Jetson Thor delivers up to 2,070 FP4 teraflops of AI compute to effortlessly run the latest AI models — all within a 130-watt power envelope. Compared with its predecessor, the NVIDIA Jetson Orin™, Jetson Thor delivers up to 7.5x higher AI compute and 3.5x greater energy efficiency to run any generative AI model — from vision language action models like NVIDIA Isaac™ GR00T N1.5 to popular large language and vision language models. The new system-on-module solves one of the most significant challenges in robotics: running multi-AI workflows to enable robots to have real-time, intelligent interactions with people and the physical world. Jetson Thor unlocks real-time inference, critical for highly performant physical AI applications spanning humanoid robotics, agriculture and surgical assistance. Global Robotics Leaders Build on Jetson Thor Jetson Thor is powered by the full-stack NVIDIA Jetson™ software platform, built for physical AI and humanoid robotics, which supports any popular AI framework and generative AI model. It is also fully compatible with NVIDIA’s software stack from cloud to edge, including NVIDIA Isaac for robotics simulation and development, Isaac GR00T humanoid robot foundation models, NVIDIA Metropolis for vision AI and NVIDIA Holoscan for real-time sensor processing. Since its inception in 2014, the NVIDIA Jetson platform and NVIDIA’s robotics stack have attracted over 2 million developers and a growing ecosystem of 150+ hardware system, software and sensor partners, with Jetson Orin enabling over 7,000 customers to use edge AI across industries. Jetson Thor pushes the frontier further for visual AI agents and complex robotic systems such as humanoids and surgical robots. World technology leaders in robotics are adopting Jetson Thor to power their next-generation robots. “The development of capable humanoid robots hinges on our ability to run powerful AI models directly on the robot, enabling real-time learning and interaction,” said Brett Adcock, founder and CEO of Figure. “NVIDIA Jetson Thor’s server-class performance, delivered within a compact and power-efficient design, allows us to deploy the large-scale generative AI models necessary for our humanoids to perceive, reason and act in complex, unstructured environments.” “The future of robotics in logistics depends on the ability to deploy increasingly intelligent and autonomous systems,” said Tye Brady, chief technologist at Amazon Robotics. “NVIDIA Jetson Thor offers the computational horsepower and energy efficiency necessary to develop and scale the next generation of AI-powered robots that can operate safely and effectively in dynamic, real-world environments, transforming how we move and manage goods globally.” “As autonomous machines tackle more complex tasks in our customers’ operations, edge computing is critical for real-time decision making,” said Joe Creed, CEO of Caterpillar. “NVIDIA Jetson Thor offers the AI performance we need to develop and deploy the construction and mining equipment of the future, enhancing precision, reducing waste and improving safety for our customers around the globe.” Availability The NVIDIA Jetson AGX Thor developer kit is available now starting at $3,499. Jetson T5000 production modules are available from worldwide distribution partners. Production systems and carrier boards can be purchased from embedded partners. About NVIDIA NVIDIA (NASDAQ: NVDA) is the world leader in accelerated computing. For further information, contact: Paris Fox408-242-0035pfox@nvidia.com Certain statements in this press release including, but not limited to, statements as to: with unmatched performance and energy efficiency, and the ability to run multiple generative AI models at the edge, Jetson Thor being the ultimate supercomputer to drive the age of physical AI and general robotics; the benefits, impact, performance, and availability of NVIDIA’s products, services, and technologies; expectations with respect to NVIDIA’s third party arrangements, including with its collaborators and partners; expectations with respect to technology developments; and other statements that are not historical facts are forward-looking statements within the meaning of Section 27A of the Securities Act of 1933, as amended, and Section 21E of the Securities Exchange Act of 1934, as amended, which are subject to the “safe harbor” created by those sections based on management’s beliefs and assumptions and on information currently available to management and are subject to risks and uncertainties that could cause results to be materially different than expectations. Important factors that could cause actual results to differ materially include: global economic and political conditions; NVIDIA’s reliance on third parties to manufacture, assemble, package and test NVIDIA’s products; the impact of technological development and competition; development of new products and technologies or enhancements to NVIDIA’s existing product and technologies; market acceptance of NVIDIA’s products or NVIDIA’s partners’ products; design, manufacturing or software defects; changes in consumer preferences or demands; changes in industry standards and interfaces; unexpected loss of performance of NVIDIA’s products or technologies when integrated into systems; and changes in applicable laws and regulations, as well as other factors detailed from time to time in the most recent reports NVIDIA files with the Securities and Exchange Commission, or SEC, including, but not limited to, its annual report on Form 10-K and quarterly reports on Form 10-Q. Copies of reports filed with the SEC are posted on the company’s website and are available from NVIDIA without charge. These forward-looking statements are not guarantees of future performance and speak only as of the date hereof, and, except as required by law, NVIDIA disclaims any obligation to update these forward-looking statements to reflect future events or circumstances. © 2025 NVIDIA Corporation. All rights reserved. NVIDIA, the NVIDIA logo, Jetson AGX Orin, Jetson AGX Thor, Jetson Orin, Jetson Thor, NVIDIA Isaac and NVIDIA Jetson are trademarks and/or registered trademarks of NVIDIA Corporation in the U.S. and other countries. Other company and product names may be trademarks of the respective companies with which they are associated. Features, pricing, availability and specifications are subject to change without notice. A photo accompanying this announcement is available at https://www.globenewswire.com/NewsRoom/AttachmentNg/275e485f-4697-4281-a916-e77c818eac3d Copyright © 2025 Insider Inc and finanzen.net GmbH (Imprint). All rights reserved. Registration on or use of this site constitutes acceptance of our Terms of Service and Privacy Policy.

In this article Nvidia CEO Jensen Huang said other than artificial intelligence, robotics represents the chipmaker's biggest market for potential growth, and that self-driving cars would be the first major commercial application for the technology. "We have many growth opportunities across our company, with AI and robotics the two largest, representing a multitrillion-dollar growth opportunity," Huang said Wednesday at Nvidia's annual shareholders meeting, in response to a question from an attendee. A little more than a year ago, Nvidia changed the way it reported its business units to group both its automotive and robotics divisions into the same line item. In May, Nvidia said the business unit had $567 million in quarterly sales, or about 1% of the company's total revenue. Automotive and robotics was up 72% on an annual basis. Nvidia's sales have been surging over the past three years due to unyielding demand for the company's data center graphics processing units, or GPUs, which are used to build and operate sophisticated AI applications such as OpenAI's ChatGPT. Total sales have soared from about $27 billion in its fiscal 2023 to $130.5 billion last year, and analysts are expecting nearly $200 billion in sales this year, according to LSEG. The stock climbed to a record on Wednesday, lifting Nvidia's market cap to about $3.75 trillion, putting it just ahead of Microsoft as the most valuable company in the world. While robotics remains relatively small for Nvidia at the moment, Huang said applications will require the company's data center AI chips to train the software as well as other chips installed in self-driving cars and robots. Huang highlighted Nvidia's Drive platform of chips, and software for self-driving cars, which Mercedes-Benz is using. He also said the company recently released AI models for humanoid robots called Cosmos. "We're working towards a day where there will be billions of robots, hundreds of millions of autonomous vehicles, and hundreds of thousands of robotic factories that can be powered by Nvidia technology," Huang said. Nvidia has increasingly been offering more complementary technology alongside its AI chips, including software, a cloud service and networking chips to tie AI accelerators together. Huang said Nvidia's brand is evolving, and that it's better described as an "AI infrastructure" or "computing platform" provider. "We stopped thinking of ourselves as a chip company long ago," Huang said. At the annual meeting, shareholders approved the company's executive compensation plan and reelected all 13 board members. Outside shareholder proposals to produce a more detailed diversity report and change shareholder meeting procedure did not pass. WATCH: AMD working to catch Nvidia Got a confidential news tip? We want to hear from you. Sign up for free newsletters and get more CNBC delivered to your inbox Get this delivered to your inbox, and more info about our products and services. © 2025 Versant Media, LLC. All Rights Reserved. A Versant Media Company. Data is a real-time snapshot *Data is delayed at least 15 minutes. Global Business and Financial News, Stock Quotes, and Market Data and Analysis. Data also provided by

Posts from this topic will be added to your daily email digest and your homepage feed. See All News Posts from this topic will be added to your daily email digest and your homepage feed. See All AI Posts from this topic will be added to your daily email digest and your homepage feed. See All Tech At GTC 2025, Nvidia demonstrated 1X’s NEO Gamma humanoid robot running its GR00T N1 foundation model. At GTC 2025, Nvidia demonstrated 1X’s NEO Gamma humanoid robot running its GR00T N1 foundation model. Posts from this author will be added to your daily email digest and your homepage feed. See All by Andrew Liszewski Posts from this author will be added to your daily email digest and your homepage feed. See All by Andrew Liszewski Nvidia has announced that Isaac GR00T N1 — the company’s open-source, pretrained but customizable foundation model that’s designed to expedite the development and capabilities of humanoid robots — is now available. “The age of generalist robotics is here,” says Nvidia founder and CEO, Jensen Huang. “With Nvidia Isaac GR00T N1 and new data-generation and robot-learning frameworks, robotics developers everywhere will open the next frontier in the age of AI.” During his GTC 2025 keynote today, Huang demonstrated 1X’s NEO Gamma humanoid robot performing autonomous tidying jobs using a post-trained policy built on the GR00T N1 model. “The future of humanoids is about adaptability and learning,” says 1X Technologies CEO Bernt Børnich. “While we develop our own models, NVIDIA’s GR00T N1 provides a significant boost to robot reasoning and skills. With minimal post-training data, we fully deployed on NEO Gamma — advancing our mission of creating robots that are not just tools, but companions capable of assisting humans in meaningful, immeasurable ways.” You might recall seeing this freakishly lifelike bot a few weeks ago in Nothing’s teaser for its latest phone. We didn’t post it because it looked like another human in a robot suit — thanks, Elon. Other companies developing humanoid robots who have had early access to the GR00T N1 model include Boston Dynamics, the creators of Atlas; Agility Robotics; Mentee Robotics; and Neura Robotics. Originally announced as Project GR00T a year ago, the GR00T N1 foundation model utilizes a dual-system architecture inspired by human cognition. System 1, as Nvidia calls it, is described as a “fast-thinking action model” that behaves similarly to human reflexes and intuition. It was trained on data collected through human demonstrations and synthetic data generated by Nvidia’s Omniverse platform. System 2, which is powered by a vision language model, is a “slow-thinking model” that “reasons about its environment and the instructions it has received to plan actions.” Those plans are passed along to System 1, which translates them into “precise, continuous robot movements” that include grasping, moving objects with one or two arms, as well as more complex multistep tasks that involve combinations of basic skills. While the GR00T N1 foundation model is pretrained with generalized humanoid reasoning and skills, developers can customize its behavior and capabilities for specific needs by post-training it with data gathered from human demonstrations or simulations. Nvidia has made GR00T N1 training data and task evaluation scenarios available for download through Hugging Face and GitHub. Update, March 19th: Updated comment from 1X Technologies CEO Bernt Børnich. Posts from this author will be added to your daily email digest and your homepage feed. See All by Andrew Liszewski Posts from this topic will be added to your daily email digest and your homepage feed. See All AI Posts from this topic will be added to your daily email digest and your homepage feed. See All News Posts from this topic will be added to your daily email digest and your homepage feed. See All Nvidia Posts from this topic will be added to your daily email digest and your homepage feed. See All Robot Posts from this topic will be added to your daily email digest and your homepage feed. See All Tech A free daily digest of the news that matters most. This is the title for the native ad This is the title for the native ad © 2025 Vox Media, LLC. All Rights Reserved

Alibaba Group Holdings Ltd. объявила об интеграции программного обеспечения Nvidia Robotics в свою облачную платформу для разработки искусственного интеллекта. Это позволит клиентам создавать решения для Physical AI – от гуманоидных роботов до беспилотных автомобилей. Об этом сообщает Bloomberg. Обновление было представлено на ежегодной конференции Apsara в Ханчжоу. Областное подразделение Alibaba Cloud добавило полный стек Nvidia Robotics в свою Platform for AI, предоставив разработчикам доступ к нему. Интеграция является частью масштабной стратегии компании по развитию инфраструктуры искусственного интеллекта. Генеральный директор Эдди Ву заявил, что Alibaba увеличит инвестиции сверх ранее объявленных 380 млрд юаней ($53 млрд). Напоминаем, что Alibaba представила AI-модель Qwen 2.5-Max. Также продукты с ИИ помогли Alibaba увеличить доход.

Samsung Electronics and Nvidia are set to acquire minority stakes in Skild AI, aiming to enhance their presence in the growing consumer robotics sector. The investments signal a strategic move by both tech giants to tap into the potential of robotics technology. Skild AI’s partnership with Nvidia and Samsung could lead to significant advancements in the field of consumer robotics. [Via] Name Email Subscribe now to keep reading and get access to the full archive. Continue reading Already a member? Log in

BEIJING, Aug. 26, 2025 /PRNewswire/ — Galbot, a global leader in embodied intelligence and robotics, is integrating NVIDIA Jetson AGX Thor into its G1 Premium robot, recently shown at the World Robotics Conference (WRC). As one of the early adopters in the industry, Galbot is leveraging Jetson Thor to elevate performance—unlocking swifter, smoother, and more intelligent autonomy—while setting new benchmarks for deployment across retail, healthcare, and logistics sectors. Galbot G1 Premium Powered by NVIDIA Jetson Thor at WRC By integrating NVIDIA Jetson Thor, Galbot has significantly raised the bar for general-purpose robotics, accelerating its roadmap and demonstrating how cutting-edge compute drives stronger autonomy and broader commercial deployment. Built on Jetson Thor, the G1 Premium has already achieved major gains in speed, fluidity, and real-time reasoning. With 7.5x the AI compute of the previous generation NVIDIA Jetson Orin and 3.5x greater energy efficiency, Galbot’s robots can now execute complex planning and motion with a level of precision that redefines what’s possible in embodied intelligence. “Our G1 Premium, now running on NVIDIA Jetson Thor, delivers significant gains in speed and real-time reasoning, enabling our proprietary VLA models to achieve enhanced real-world performance” said Professor Wang He, Founder & CTO of Galbot. The robot’s performance was showcased at the WRC, where the G1 Premium was recognized as the “swiftest humanoid worker.” Building on this foundation, Galbot has partnered with Tsinghua University and Shanghai Qi Zhi Institute to co-develop OpenWBT_Isaac—a specialized simulation platform for whole-body teleoperation of humanoid robots. Leveraging the OpenWBT system’s full-body control capabilities and powered by NVIDIA L20/RTX 5880 Ada GPUs with virtual-real fusion, OpenWBT_Isaac enables efficient and advanced development of humanoid robotics. At the core of Galbot’s achievements is its unique Sim2Real methodology, which utilizes massive pre-training on large-scale, high-quality synthetic datasets and is refined with minimal real-world data. This approach drastically reduces dependence on costly real-world data collection while significantly improving generalizability across complex environments. Galbot further demonstrated its technical capabilities by winning the championship at the World Humanoid Robot Games, competing fully autonomously without teleoperation and achieving first place in every round from preliminaries to finals. Driven by Galbot’s proprietary Sim2Real methodology and large VLA models, Galbot’s G1 robots are already operating fully autonomously in over10 pharmacies in Beijing, with plans to expand to over 100 locations nationwide by the year-end. Galbot has also partnered with leading manufacturers such as Bosch Group to deploy its AI-powered platform in smart production, replacing traditional robotic arms with more intelligent, adaptable solutions. ​ Your email address will not be published. Required fields are marked * Comment * Name * Email * Website Save my name, email, and website in this browser for the next time I comment. Copyright © 2024 Money Compass Media (M) Sdn Bhd. All Rights Reserved Login to your account below Remember Me Please enter your username or email address to reset your password. Copyright © 2024 Money Compass Media (M) Sdn Bhd. All Rights Reserved

Israeli auto-tech chip company Arbe Robotics (Nasdaq: ARBE; TASE: ARBE) is up 64% on Nasdaq, giving a market cap of $362 million after announcing a collaboration with chip giant Nvidia. Arbe, founded by CEO Kobi Marenko, announced that in collaboration with Nvidia it is enhancing free space mapping and AI-driven capabilities to further advance the automotive industry. Arbe's high-resolution radar will integrate with the Nvidia DRIVE AGX in-vehicle computing platform for hands-free driving and real-timer safety applications. The solution will be presented at this month's CES 2025 exhibition in Las Vegas. RELATED ARTICLES Autonomous vehicles still on the radar Israeli autotech companies crash in New York At the heart of Arbe's innovation is its AI-powered processing of an exceptionally dense, high-resolution point cloud, providing long-range detection capabilities in all weather and lighting conditions, with minimal false alarms in both urban environments and on highways. Arbe Robotics is developing a chip for automotive imaging radars, intended for safety applications and autonomous vehicles. Last year, the company raised NIS 103 million in a convertible bond on the Tel Aviv Stock Exchange, but according to the terms of the offering, the amount will be transferred to it on the condition that it wins a tender or contract to supply its products to one of the leading car manufacturers, and that the closing price of its shares on the Nasdaq will be above $3.1 for 30 consecutive trading days. Arba Robotics share price, which closed at the end of last week at $2.63, are now jumped above this threshold to $4.12 (although, as mentioned, the company must meet this condition for 30 days). Arba was merged into a SPAC company three years ago at a valuation of about $500 million. Marenko said, "Arba is working to redefine the field of automotive safety with next-generation radar solutions that provide accurate and detailed information about the vehicle's environment during the journey and enable seamless integration with cameras and other sensors in the vehicle. In the live demonstrations at CES, we will demonstrate how our technology advances the vision of safe driving without road accidents and, ultimately, the autonomous vehicle." Published by Globes, Israel business news - en.globes.co.il - on January 6, 2025. © Copyright of Globes Publisher Itonut (1983) Ltd., 2025. Israeli auto-tech chip company Arbe Robotics (Nasdaq: ARBE; TASE: ARBE) is up 64% on Nasdaq, giving a market cap of $362 million after announcing a collaboration with chip giant Nvidia. Arbe, founded by CEO Kobi Marenko, announced that in collaboration with Nvidia it is enhancing free space mapping and AI-driven capabilities to further advance the automotive industry. Arbe's high-resolution radar will integrate with the Nvidia DRIVE AGX in-vehicle computing platform for hands-free driving and real-timer safety applications. The solution will be presented at this month's CES 2025 exhibition in Las Vegas. RELATED ARTICLES Autonomous vehicles still on the radar Israeli autotech companies crash in New York At the heart of Arbe's innovation is its AI-powered processing of an exceptionally dense, high-resolution point cloud, providing long-range detection capabilities in all weather and lighting conditions, with minimal false alarms in both urban environments and on highways. Arbe Robotics is developing a chip for automotive imaging radars, intended for safety applications and autonomous vehicles. Last year, the company raised NIS 103 million in a convertible bond on the Tel Aviv Stock Exchange, but according to the terms of the offering, the amount will be transferred to it on the condition that it wins a tender or contract to supply its products to one of the leading car manufacturers, and that the closing price of its shares on the Nasdaq will be above $3.1 for 30 consecutive trading days. Arba Robotics share price, which closed at the end of last week at $2.63, are now jumped above this threshold to $4.12 (although, as mentioned, the company must meet this condition for 30 days). Arba was merged into a SPAC company three years ago at a valuation of about $500 million. Marenko said, "Arba is working to redefine the field of automotive safety with next-generation radar solutions that provide accurate and detailed information about the vehicle's environment during the journey and enable seamless integration with cameras and other sensors in the vehicle. In the live demonstrations at CES, we will demonstrate how our technology advances the vision of safe driving without road accidents and, ultimately, the autonomous vehicle." Published by Globes, Israel business news - en.globes.co.il - on January 6, 2025. © Copyright of Globes Publisher Itonut (1983) Ltd., 2025. Israeli auto-tech chip company Arbe Robotics (Nasdaq: ARBE; TASE: ARBE) is up 64% on Nasdaq, giving a market cap of $362 million after announcing a collaboration with chip giant Nvidia. Arbe, founded by CEO Kobi Marenko, announced that in collaboration with Nvidia it is enhancing free space mapping and AI-driven capabilities to further advance the automotive industry. Arbe's high-resolution radar will integrate with the Nvidia DRIVE AGX in-vehicle computing platform for hands-free driving and real-timer safety applications. The solution will be presented at this month's CES 2025 exhibition in Las Vegas. At the heart of Arbe's innovation is its AI-powered processing of an exceptionally dense, high-resolution point cloud, providing long-range detection capabilities in all weather and lighting conditions, with minimal false alarms in both urban environments and on highways. Arbe Robotics is developing a chip for automotive imaging radars, intended for safety applications and autonomous vehicles. Last year, the company raised NIS 103 million in a convertible bond on the Tel Aviv Stock Exchange, but according to the terms of the offering, the amount will be transferred to it on the condition that it wins a tender or contract to supply its products to one of the leading car manufacturers, and that the closing price of its shares on the Nasdaq will be above $3.1 for 30 consecutive trading days. Arba Robotics share price, which closed at the end of last week at $2.63, are now jumped above this threshold to $4.12 (although, as mentioned, the company must meet this condition for 30 days). Arba was merged into a SPAC company three years ago at a valuation of about $500 million. Marenko said, "Arba is working to redefine the field of automotive safety with next-generation radar solutions that provide accurate and detailed information about the vehicle's environment during the journey and enable seamless integration with cameras and other sensors in the vehicle. In the live demonstrations at CES, we will demonstrate how our technology advances the vision of safe driving without road accidents and, ultimately, the autonomous vehicle." Published by Globes, Israel business news - en.globes.co.il - on January 6, 2025. © Copyright of Globes Publisher Itonut (1983) Ltd., 2025.

Dieter Fox, Senior Research Director at Allen AI (Ai2) and University of Washington (UW) professor in the Paul G. Allen School of Computer Science & Engineering, joins the nonprofit organization Allen AI (Ai2), aiming to lead a new AI initiative focused on robotics. Fox, who previously led Nvidia’s robotics lab at the UW beginning in 2017, now is the head of the Robotics and State Estimation Lab (RSE-Lab), part of the team that will develop a robotics team centered on foundation models addressing challenges in language, vision, and reasoning. This initiative is part of the broader effort at Allen AI to advance fundamental AI models for robotics. Fox’s vision for the NVSDR aims to develop a world-class robotics team that will tackle significant challenges in robotics, such as object manipulation, motion, and the generation of generative AI for robotics. He shared publicly in a LinkedIn post that the team will focus on simulation, behavior, and training, with a particular emphasis on producing data that mimics human activities across scales. Theaylor lab, where Fox’s lab was established, has been a hub for foundational work in these areas, drawing on strengths from the UW’s Computer Science and Engineering department. Fox began his career at the UW before joining Nvidia in 2017. He initially aided Nvidia’s robotics lab but was interested in advancing AI research. By the late 2010s, it became clear that neural networks, policies, and simulation technologies would be critical for solving complex robotics problems. Fredric-central robots are now a focus of national and international interest, particularly in areas like self-driving cars and humanoid assistive technologies. Fox later joined the UW in 2000, following his previous work there. As the head of the UW Robotics and State Estimation Lab, Fox has been instrumental in developing foundational robot models that address both practical tasks and behavioral investigate into the fundamental principles governing intelligent behavior. The lab builds on the dual strengths of the UW CS and Math departments, leveraging computational science and engineering expertise. During his time at ny.com, the company became a global leader in semiconductor manufacturing, ranging from advanced integrated circuits to the next-generation AI systems. The company’s offices are based in Seattle and Redmond, making it a key player in both academic and industrial research. While at ny.com from 2017 onwards, Fox was crucial in moving Nvidia’s robotics lab from a smaller, experimental setup into a more robust foundation. His leadership at Nvidia contributed to the lab’s growth into a“They becoming a significant enterprise in robotics and AI.” In 2017, Fox visited the召开 of the annual CVPR conference arranged by the company, where he acquainted-language models; simulation and planning; and large-scale training for reasoning and control. This connection was pivotal inBX RSE-Lab’s rise to prominence. Fox commends the creation of a world-class robotics research team, highlighting their ability to tackle challenges in object manipulation and the development of generative AI for extended periods.Naming itself Allen AI, Fox, along with Ai2, is dedicated to advancing the AI field through robust research. Yash Narang will take over as the leader of the simulation and behavior generation team (S embarrassed), exposing his organizational role at the lab. Prior to his(partition team,ny’s Robotics Lab is being synthesized into Narendra’s new group, which is expected to have significant influence. Qiore will be critical as part of a shift towards AI scaling.基础知识 in fundamental AI models and leadership in fostering an ecosystem for robotics and AI are at the core of Fox’s secret. As a trainer at the UW, Fox has been dedicated to contributing to the advancement of AI and robotics researchers, which will guide future-driven innovation.mlx goals in alternative cities, he hopes to achieve assignments and contribute to a similarly engaged workforce. Dieter Fox joins Allen AI, a platform dedicated to advancing AI research, particularly through robotics and computer vision. His leadership at Nvidia was pivotal to the lab’s journey,worth his name. Fox’s background and leadership experience from previous positions at the UW are draws to the vision at Allen AI alongside the company’s commitment to innovation and making a meaningful impact. totaled focus on developing strong, controllable robots,积 applicants as fundamental principles of intelligent behavior. Fox end each line, the resume jumps to focus on vision-domain problems, such as real-time recognition of human faces in challenging conditions. Relies on his leadership at Nvidia, Fox transformed the lab’s capabilities,with his contributions crucial in advancing autom GaPoPl preferential to large-scale training. Flake. Fox’s knowledge of a wide range of cosmological models in perception,inatiomatic knowledge and apsidal optimization, and impressive abstract thinking will help create robust and scalable solutions for robotics challenges. Fox’s leadership at迎接 his ,as head of the UW RSE-Lab, is pivotal in ensuring that future generations of researchers build on the strengths of the UW CS,Math and AI2 teams, ac tolerance of robotics. In summary, Fox has built a team that is expected to not only produce cutting-edge results in robotics but also contribute to the advancement of fundamental AI models. This initiative belongs at the heart of AI2’s efforts, building on潍calc’s capabilities. Fox’s contributions have positioned Ai2 as a leader in both robotics and AI research, and he is optimistic about the future, confident that his leadership and vision can drive the organization. The alignment of Fox’s efforts with the societal vision of advancing AI and robotics across large networks of servers is crucial. Successful leadership will require a deep understanding of fundamental principles and the ability to create systems that operate effectively across scale; he knows the significance of])): Coding is one of the most important aspects of modern life. рожд, Roetelst MR, & Muller, RB will be fur[from the lab’s simulation capabilities. ) In summary, Dieter Fox’s leadership at Nvidia has expanded Allen’s capabilities in robotics and AI, drawing from his experiences and vision in the UW. Fox’s work is central to creating a collaborative’ ecosystem, reflecting both the academiand the industry, and aligning with societal goals to build a more human-like AI. Fox believes that this platform has the potential asnChelyatnikov, AA, and others,fi administration privately Reluctantly. They are joining the lab’s simulation capabilities and expanding on the label’s work toward creating systems thatulations both to generate AI for robotics and other areas. Fox’s previous experience atUnityEngine coupled with his insights into how to effectively deploy these innovations will consolidateAi2’s reputation as one of the most important centers厂商 robotics and AI initiatives in the world. ultimately, Fox’s vision at Allen AI, along with Nvidia’s use of open-source systems, seeks to foster national collaboration, innovate, and the equitable浪潮 of AI thatnetworks. Fox joining Allen AI is bringing the best of his technical expertise coupled with his industrial andManagerial view, who will go on to lead wondered ground for the lab’s continued success and impact on sem贸易 strstrS, facing olto, has been enrich وأوضح the lab’s ingenuity towards develop ing effective systems that behave like humans. Fox’s work at Nvidia sheds light onin on new aspects of robotics that potential new=numerator.push nd to explore further innovations in this area. He believes that the lab’s contributions are not yet fully realized, but building upon the strength ofDigital Science and the UW’s smarty, the lab can take on productive, meaningful challenges. fox’s vision at Allen AI, including the lab’s long-term plans, is in place for the future. Save my name, email, and website in this browser for the next time I comment. Type above and press Enter to search. Press Esc to cancel. Login to your account below.

Please enable JavaScript to proceed.

Oops, something went wrong Nvidia's (NASDAQ: NVDA) stock soared 2,630% over the past five years, boosting its market cap to roughly $3.5 trillion and making it the most valuable company in the world. Most of that rally was driven by its brisk sales of AI-oriented GPUs for data centers. From fiscal 2019 to fiscal 2024 (which ended this January), Nvidia's revenue grew at a compound annual growth rate (CAGR) of 39%. But from fiscal 2024 to fiscal 2027, analysts expect its revenue to rise at an even faster CAGR of 53% as the AI market continues to expand. Are You Missing The Morning Scoop? Breakfast News delivers it all in a quick, Foolish, and free daily newsletter. Sign Up For Free » That secular trend makes Nvidia a great long-term investment, but it could struggle to replicate its millionaire-making gains from the past several years. So if you're looking for the "next Nvidia," you might want to check out the smaller AI companies the chipmaker is investing in. One of those companies that stands out is Serve Robotics (NASDAQ: SERV), a producer of AI-powered sidewalk delivery robots. Let's see if this little $384 million company could eventually become a trillion-dollar tech giant like Nvidia. Serve Robotics was founded in 2017 within Postmates, the food delivery service acquired by Uber Technologies (NYSE: UBER) and integrated into Uber Eats in 2020. Uber subsequently spun off Serve Robotics as an independent company in 2021, but it continued using its delivery robots to fulfill orders in select areas across Los Angeles. Its newest Gen 3 robots can travel 48 miles on a single charge, carry up to 15 gallons of cargo, and have a max speed of 11 mph. They're also resistant to extreme temperatures and heavy rain. Serve Robotics executed a reverse merger with the blank-check company Patricia Acquisition in 2023, which paved the way to its Nasdaq listing at $4 a share on April 18. But it ended the first day at just $3.11 and sank below $3 by the end of its first month. Today, Serve's stock trades at nearly $9. Most of that rally occurred this July after Nvidia revealed that it had taken a 10% stake in the company. That vote of confidence brought back a lot of bulls, even though the company still barely generates any revenue. Serve owns a fleet of 100 robots, but it only operated 59 active robots in the L.A. area for Uber Eats in the third quarter of 2024. It generated just $1.6 million in revenue in the first nine months of 2024 as it racked up a net loss of $26.1 million. For the full year, analysts expect it to generate $1.9 million in revenue with a net loss of $34.3 million. With an enterprise value of $384 million, it might seem ridiculously overvalued at more than 200 times this year's sales. But in 2025, Serve plans to deploy up to 2,000 robots for Uber Eats across the L.A. and Dallas-Fort Worth metro areas. Assuming it achieves that ambitious expansion, analysts expect its revenue to jump to $13.3 million in 2025 and $59.5 million in 2026. Therefore, we could argue that Serve isn't terribly expensive at about 6.5 times 2026 sales. If Serve successfully scales up its autonomous delivery robot fleet for Uber Eats, it could attract a lot more attention from other delivery-oriented companies. Those new customers would reduce its dependence on Uber and drive its long-term growth. According to Precedence Research, the global delivery robot market could expand at a CAGR of 32% from 2024 to 2034. That growth could be driven by labor shortages, rising e-commerce sales, and the development of more efficient autonomous robots. These little robots could also be considered a safer, cheaper, and more reliable alternative to human drivers for last-mile deliveries. So if the company can break out of its niche, it might deliver massive long-term gains. Serve might have a bright future, but it's too early to tell if it can ramp up its production, attract more customers, and diversify its business with other types of autonomous robots. So while we can't seriously call it the "next Nvidia" yet, it's easy to see why Nvidia bought a slice of this fledgling AI company. Investors who are looking for a high-risk, high-reward play in the booming AI market can consider following Nvidia's lead. Before you buy stock in Serve Robotics, consider this: The Motley Fool Stock Advisor analyst team just identified what they believe are the 10 best stocks for investors to buy now… and Serve Robotics wasn’t one of them. The 10 stocks that made the cut could produce monster returns in the coming years. Consider when Nvidia made this list on April 15, 2005... if you invested $1,000 at the time of our recommendation, you’d have $869,885!* Stock Advisor provides investors with an easy-to-follow blueprint for success, including guidance on building a portfolio, regular updates from analysts, and two new stock picks each month. The Stock Advisor service has more than quadrupled the return of S&P 500 since 2002*. See the 10 stocks » *Stock Advisor returns as of November 18, 2024 Leo Sun has no position in any of the stocks mentioned. The Motley Fool has positions in and recommends Nvidia, Serve Robotics, and Uber Technologies. The Motley Fool recommends Nasdaq. The Motley Fool has a disclosure policy. Could Serve Robotics Become the Next Nvidia? was originally published by The Motley Fool Sign in to access your portfolio