智能体与机械臂融合技术研究进展及展望
Advances and Future Outlook in the Convergence of Agents and Robotic Manipulators
面向复杂环境下机械臂操作任务对自主决策、长时序执行与多约束交互的需求,智能体(Agent)与机械臂系统的融合逐渐成为具身智能的重要技术路径。为支撑该方向的研究梳理与工程落地,本文旨在构建Agent ‒ 机械臂融合研究的结构化技术框架,明确方法体系、数据评测脉络及关键挑战。本文首先回顾机械臂智能化的技术基础与控制范式演进;随后从强化学习型智能体、规划与认知型智能体以及大模型驱动具身智能体三个层级,对代表性方法的系统结构、决策机制与典型应用进行归纳;进一步梳理典型公开数据集,给出数据与选型要点。研究表明,现有Agent ‒ 机械臂融合研究已形成以学习型策略、规划推理机制与大模型驱动闭环为核心的三类技术路线,不同路线在任务泛化能力、交互复杂度适应性与部署成本方面呈现差异;同时,数据分布覆盖不足、评测协议不统一等问题制约了结论的可复现性与可比性。面向通用操作智能体的发展,未来研究需重点关注数据与任务覆盖扩展、跨域迁移与鲁棒性提升、安全可靠机制以及工程化部署流程等方向,为相关方法由原型验证走向规模化应用提供支撑。
To address the needs of robotic manipulators in complex environments for autonomous decision-making, long-horizon execution, and multi-constraint interactions, the integration of intelligent agents with robotic manipulator systems is increasingly recognized as a key technical pathway for embodied intelligence. To support research review and engineering implementation in this direction, this study seeks to establish a structured technical framework for agent‒manipulator integration, clarifying the methodological landscape, data-and-evaluation context, and key challenges. First, the study reviews the technical foundations of manipulator intelligence and the evolution of control paradigms. Then, it summarizes representative methods in terms of system architectures, decision-making mechanisms, and typical applications from three levels: reinforcement-learning-based agents, planning and cognition-oriented agents, and large-model-driven embodied agents. Furthermore, we examine representative public datasets and provide key considerations for dataset selection. Our analysis shows that existing agent‒manipulator integration research has formed three major technical routes centered on learning-based policies, planning-and-reasoning mechanisms, and large-model-driven closed-loop frameworks, and these routes differ in task generalization capability, adaptability to interaction complexity, and deployment cost. Meanwhile, insufficient coverage of data distributions and non-unified evaluation protocols hinder the reproducibility and comparability of reported findings. Toward the development of general-purpose manipulator agents, future research should prioritize expanding data and task coverage, improving cross-domain transfer and robustness, developing safety and reliability mechanisms, and establishing engineering-oriented deployment pipelines, thereby supporting the transition of relevant methods from prototype validation to scalable application.
| [1] |
Li D Z,Jin Y X,Sun Y H,et al. What foundation models can bring for robot learning in manipulation:A survey[J]. The International Journal of Robotics Research,2026,45(7):1091-1142. |
| [2] |
Liu Y,Chen W X,Bai Y J,et al. Aligning cyber space with physical world:A comprehensive survey on embodied AI[J]. IEEE/ASME Transactions on Mechatronics,2025,30(6):7253-7274. |
| [3] |
Zitkovich B,Yu T,Xu S,et al.RT-2:Vision-Language-Action models transfer web knowledge to robotic control[C]//The 7th Conference on Robot Learning,2023:2165-2183. |
| [4] |
Driess D,Xia F,Sajjadi M S M,et al. PaLM-E:An embodied multimodal language model[C]//The 40th International Conference on Machine Learning,2023,202:8469-8488. |
| [5] |
Collaboration O X,Padalkar A,Pooley A,et al. Open X-embodiment:Robotic learning datasets and RT-X models[C]//IEEE International Conference on Robotics and Automation (ICRA),2024:1-10. |
| [6] |
Ghosh D,Walke H R,Pertsch K,et al. Octo:An open-source generalist robot policy[C]//Robotics:Science and Systems XX,2024:90. |
| [7] |
Walke H R,Black K,Zhao T Z,et al. BridgeData V2:A dataset for robot learning at scale[C]//The 7th Conference on Robot Learning,2023:1723-1736. |
| [8] |
Fu Z P,Zhao T Z,Finn C. Mobile ALOHA:Learning bimanual mobile manipulation using low-cost whole-body teleoperation[C]//The 8th Conference on Robot Learning,2025:4066-4083. |
| [9] |
Mees O,Hermann L,Rosete-Beas E,et al. CALVIN:A benchmark for language-conditioned policy learning for long-horizon robot manipulation tasks[J]. IEEE Robotics and Automation Letters,2022,7(3):7327-7334. |
| [10] |
Liu B,Zhu Y F,Gao C K,et al. LIBERO:Benchmarking knowledge transfer for lifelong robot learning[C]//The 37th Annual Conference on Neural Information Processing Systems,2023:44776-44791. |
| [11] |
Xu Z Y,Wu K,Wen J J,et al. A survey on robotics with foundation models:Toward embodied AI[PP/OL]. v1. arXiv(2024-02-04)[2026-03-31]. https://arxiv.org/abs/2402.02385. |
| [12] |
Ichter B,Brohan A,Chebotar Y,et al. Do as I can,not as I say:Grounding language in robotic affordances[C]// The 6th Conference on Robot Learning,2022:287-318. |
| [13] |
ISO 8373:2021 Robotics—Vocabulary[S]. |
| [14] |
International Federation of Robotics. World robotics 2025—Industrial robots:Sources and methods[R]. Frankfurt:VDMA Services GmbH,2025:30-31. |
| [15] |
Denavit J,Hartenberg R S. A kinematic notation for lower-pair mechanisms based on matrices[J]. Journal of Applied Mechanics,1955,22(2):215-221. |
| [16] |
Nakamura Y. Advanced robotics:Redundancy and optimization[M]. Reading:Addison-Wesley Publishing Company,1991:128-137. |
| [17] |
Spong M W,Hutchinson S,Vidyasagar M. Robot modeling and control[M]. Hoboken:John Wiley & Sons,Inc.,2006:187-224. |
| [18] |
Slotine J J E,Li W. Applied nonlinear control[M]. Englewood Cliffs:Prentice Hall,1991:212-213,290-302. |
| [19] |
Hogan N. Impedance control:An approach to manipulation:Part I—Theory[J]. Journal of Dynamic Systems,Measurement,and Control,1985,107(1):1-7. |
| [20] |
Zhu M L,Gong D W,Zhao Y Y,et al. Compliant force control for robots:A survey[J]. Mathematics,2025,13(13):2204. |
| [21] |
LaValle S M. Rapidly-exploring random trees:A new tool for path planning[R]. Ames:Department of Computer Science,Iowa State University,1998:1-4. |
| [22] |
Kavraki L E,Svestka P,Latombe J C,et al. Probabilistic roadmaps for path planning in high-dimensional configuration spaces[J]. IEEE Transactions on Robotics and Automation,1996,12(4):566-580. |
| [23] |
Zucker M,Ratliff N,Dragan A D,et al. CHOMP:Covariant hamiltonian optimization for motion planning[J]. The International Journal of Robotics Research,2013,32(9/10):1164-1193. |
| [24] |
Åström K J,Hägglund T. PID controllers:Theory,design,and tuning[M]. 2nd ed. Research Triangle Park,NC:Instrument Society of America,1995:59. |
| [25] |
Luh J,Walker M,Paul R. Resolved-acceleration control of mechanical manipulators[J]. IEEE Transactions on Automatic Control,1980,25(3):468-474. |
| [26] |
Liu J,Yap H J,Khairuddin A S M. Review on motion planning of robotic manipulator in dynamic environments[J]. Journal of Sensors,2024,2024:5969512. |
| [27] |
Argall B D,Chernova S,Veloso M,et al. A survey of robot learning from demonstration[J]. Robotics and Autonomous Systems,2009,57(5):469-483. |
| [28] |
Ng A Y,Russell S J. Algorithms for inverse reinforcement learning[C]//The Seventeenth International Conference on Machine Learning (ICML 2000),2000:663-670. |
| [29] |
Ross S,Gordon G J,Bagnell J A. A reduction of imitation learning and structured prediction to No-regret online learning[C]//The Fourteenth International Conference on Artificial Intelligence and Statistics,2011,15:627-635. |
| [30] |
Correia A,Alexandre L A. A survey of demonstration learning[J]. Robotics and Autonomous Systems,2024,182:104812. |
| [31] |
Kober J,Bagnell J A,Peters J. Reinforcement learning in robotics:A survey[J]. The International Journal of Robotics Research,2013,32(11):1238-1274. |
| [32] |
Lillicrap T P,Hunt J J,Pritzel A,et al. Continuous control with deep reinforcement learning[C/OL]//The 4th International Conference on Learning Representations (ICLR 2016),2016:1-14. https://arxiv.org/abs/1509.02971v6. |
| [33] |
Schulman J,Wolski F,Dhariwal P,et al. Proximal policy optimization algorithms[PP/OL]. v2. arXiv(2017-08-28)[2026-03-31]. https://arxiv.org/abs/1707.06347. DOI:10.48550/arXiv.1707.06347. |
| [34] |
Sutton R S,Precup D,Singh S. Between MDPs and semi-MDPs:A framework for temporal abstraction in reinforcement learning[J]. Artificial Intelligence,1999,112(1/2):181-211. |
| [35] |
Tobin J,Fong R,Ray A,et al. Domain randomization for transferring deep neural networks from simulation to the real world[C]//2017 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS),IEEE,2017:23-30. |
| [36] |
Da L,Turnau J,Kutralingam T P,et al. A survey of sim-to-real methods in RL:Progress,prospects and challenges with foundation models[PP/OL]. v3. arXiv(2025-03-08)[2026-03-31]. https://arxiv.org/abs/2502.13187. DOI:10.48550/arXiv.2502.13187. |
| [37] |
Liang J,Huang W L,Xia F,et al. Code as policies:Language model programs for embodied control[C]//2023 IEEE International Conference on Robotics and Automation (ICRA). IEEE,2023:9493-9500. |
| [38] |
Fujimoto S,van Hoof H,Meger D. Addressing function approximation error in actor-critic methods[C]//The 35th International Conference on Machine Learning,2018:1587-1596. |
| [39] |
Haarnoja T,Zhou A,Abbeel P,et al. Soft actor-critic:Off-policy maximum entropy deep reinforcement learning with a stochastic actor[C]//The 35th International Conference on Machine Learning,2018:1861-1870. |
| [40] |
Dasari S,Ebert F,Tian S,et al. RoboNet:Large-scale multi-robot learning[C]//The Conference on Robot Learning,2020:885-897. |
| [41] |
Brohan A,Brown N,Carbajal J,et al. RT-1:Robotics transformer for real-world control at scale[C]//Robotics:Science and Systems XIX. Daegu,Republic of Korea,2023:25. |
| [42] |
Shridhar M,Manuelli L,Fox D. Perceiver-actor:A multi-task transformer for robotic manipulation[C]//The 6th Conference on Robot Learning,2023:785-799. |
| [43] |
Jang E,Irpan A,Khansari M,et al.BC-Z:Zero-shot task generalization with robotic imitation learning[C]//The 5th Conference on Robot Learning,2022:991-1002. |
| [44] |
Jiang Y,Gupta A,Zhang Z,et al. VIMA:Robot manipulation with multimodal prompts[C]//The 40th International Conference on Machine Learning,2023:14975-15022. |
| [45] |
Chi C,Xu Z J,Feng S Y,et al. Diffusion policy:Visuomotor policy learning via action diffusion[J]. The International Journal of Robotics Research,2025,44(10/11):1684-1704. |
| [46] |
Shafiullah N M,Cui Z,Altanzaya A,et al. Behavior transformers:Cloning k modes with one stone[C]//The 36th Annual Conference on Neural Information Processing Systems,2022:22955-22968. |
| [47] |
Eysenbach B,Gupta A,Ibarz J,et al. Diversity is all you need:Learning skills without a reward function[C/OL]//International Conference on Learning Representations, 2019:1. https://openreview.net/forum? id=SJx63jRqFm. |
| [48] |
Lynch C,Khansari M,Xiao T,et al.Learning latent plans from play[C]// The Conference on Robot Learning,2020:1113-1132. |
| [49] |
Bousmalis K,Vezzani G,Rao D,et al. RoboCat:A self-improving generalist agent for robotic manipulation[J/OL]. Transactions on Machine Learning Research,2024:1-60. https://openreview.net/forum? id=vsCpILiWHu. |
| [50] |
Lowe R,Wu Y,Tamar A,et al. Multi-agent actor-critic for mixed cooperative-competitive environments[C]//Advances in Neural Information Processing Systems 30,2017:6379-6390. |
| [51] |
Makoviychuk V,Wawrzyniak L,Guo Y,et al. Isaac Gym:High performance GPU based physics simulation for robot learning[C/OL]//The Neural Information Processing Systems Track on Datasets and Benchmarks,2021:1. https://datasets-benchmarks-proceedings.neurips.cc/paper_files/paper/2021/hash/28dd2c7955ce926456240b2ff0100bde-Abstract-round2.html. |
| [52] |
Levine S,Kumar A,Tucker G,et al. Offline reinforcement learning:Tutorial,review,and perspectives on open problems[PP/OL]. v3. arXiv(2020-11-01)[2026-03-31]. https://arxiv.org/abs/2005.01643. DOI:10.48550/arXiv.2005.01643. |
| [53] |
Zhao T Z,Kumar V,Levine S,et al. Learning fine-grained bimanual manipulation with low-cost hardware[C]//Robotics:Science and Systems XIX,2023:16. |
| [54] |
Silver T,Chitnis R,Tenenbaum J B,et al. Learning symbolic operators for task and motion planning[C]//2021 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS),2021:3182-3189. |
| [55] |
Huang D A,Nair S,Xu D,et al. Neural task graphs:Generalizing to unseen tasks from a single video demonstration[C]//The IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2019:8565-8574.. |
| [56] |
Huang W L,Abbeel P,Pathak D,et al. Language models as zero-shot planners:Extracting actionable knowledge for embodied agents[C]//The 39th International Conference on Machine Learning,2022:9118-9147. |
| [57] |
Garrett C R,Chitnis R,Holladay R,et al. Integrated task and motion planning[J]. Annual Review of Control,Robotics,and Autonomous Systems,2021,4:265-293. |
| [58] |
Envall J,Poranne R,Coros S. Differentiable task assignment and motion planning[C]//2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS),2023:2049-2056. |
| [59] |
Dantam N T,Kingston Z K,Chaudhuri S,et al. Incremental task and motion planning:A constraint-based approach[C]//Robotics:Science and Systems XII,2016. |
| [60] |
Cheng S,Xu D. LEAGUE:Guided skill learning and abstraction for long-horizon manipulation[J]. IEEE Robotics and Automation Letters,2023,8(10):6451-6458. |
| [61] |
Hafner D,Pasukonis J,Ba J,et al. Mastering diverse control tasks through world models[J]. Nature,2025,640(8059):647-653. |
| [62] |
Lu G X,Jia B X,Li P H,et al. GWM:Towards scalable gaussian world models for robotic manipulation[C]//The IEEE/CVF International Conference on Computer Vision (ICCV),2025:9263-9274. |
| [63] |
Zhu F Q,Wu H T,Guo S,et al. IRASim:A fine-grained world model for robot manipulation[C]//The IEEE/CVF International Conference on Computer Vision (ICCV),2025:9834-9844. |
| [64] |
Pang J C,Tang N,Li K Y,et al. Learning view-invariant world models for visual robotic manipulation[C/OL]//International Conference on Learning Representations (ICLR),2025:54853-54876. https://proceedings.iclr.cc/paper_files/paper/2025/hash/8882d370cdafec9885b918a8cfac642e-Abstract-Conference.html. |
| [65] |
NVIDIA,Agarwal N,Ali A,et al. Cosmos world foundation model platform for physical AI[PP/OL]. v3. arXiv(2025-07-09)[2026-03-31]. https://arxiv.org/abs/2501.03575. DOI:10.48550/arXiv.2501.03575. |
| [66] |
Kim M J,Pertsch K,Karamcheti S,et al.OpenVLA:An open-source vision-language-action model[C]//The 8th Conference on Robot Learning,2025:2679-2713. |
| [67] |
Li X H,Liu M H,Zhang H B,et al. Vision-language foundation models as effective robot imitators[C/OL]//International Conference on Learning Representations (ICLR),2024:26703-26721. https://proceedings.iclr.cc/paper_files/paper/2024/hash/71639c317fb0bf398835627b4418693e-Abstract-Conference.html. |
| [68] |
Liu J M,Liu M Z,Wang Z Y,et al. RoboMamba:Efficient vision-language-action model for robotic reasoning and manipulation[C]//Advances in Neural Information Processing Systems,2024:40085-40110. |
| [69] |
Black K,Brown N,Driess D,et al. π0:A Vision-language-action flow model for general robot control[C]//Robotics:Science and Systems XXI,2025:10. |
| [70] |
Black K,Brown N,Darpinian J,et al. π0.5:A vision-language-action model with open-world generalization[C]//The 9th Conference on Robot Learning,2025:17-40. |
| [71] |
NVIDIA,Bjorck J,Castañeda F,et al. GR00T N1:An open foundation model for generalist humanoid robots[PP/OL]. v2. arXiv(2025-03-27)[2026-03-31]. https://arxiv.org/abs/2503.14734. DOI:10.48550/arXiv.2503.14734. |
| [72] |
Shukor M,Aubakirova D,Capuano F,et al. SmolVLA:A vision-language-action model for affordable and efficient robotics[PP/OL]. v1. arXiv(2025-06-02)[2026-03-31]. https://arxiv.org/abs/2506.01844. DOI:10.48550/arXiv.2506.01844. |
| [73] |
Lee A C,Chuang I,Chen L Y,et al. InterACT:Inter-dependency aware action chunking with hierarchical attention transformers for bimanual manipulation[C]//The 8th Conference on Robot Learning,2025:1730-1743. |
| [74] |
George A,Barati Farimani A. One ACT Play:Single demonstration behavior cloning with action chunking transformers[PP/OL]. v1. arXiv(2023-09-18)[2026-03-31]. https://arxiv.org/abs/2309.10175. DOI:10.48550/arXiv.2309.10175. |
| [75] |
Huang W L,Xia F,Xiao T,et al. Inner monologue:Embodied reasoning through planning with language models[C]//The 6th Conference on Robot Learning,2022:1769-1782. |
| [76] |
Nasiriany S,Maddukuri A,Zhang L,et al. RoboCasa:Large-scale simulation of everyday tasks for generalist robots[C]//Robotics:Science and Systems XX,2024:50. |
| [77] |
Gu J Y,Xiang F B,Li X L,et al. ManiSkill2:A unified benchmark for generalizable manipulation skills[C/OL]//International Conference on Learning Representations (ICLR),2023:1-30. https://iclr.cc/virtual/2023/poster/12152. |
| [78] |
Luo Y L,Fan C K,Dong M H,et al. RoboBench:A comprehensive evaluation benchmark for multimodal large language models as embodied brain[C/OL]//European Conference on Computer Vision (ECCV),2026:501-520. https://eccv.ecva.net/virtual/2026/poster/4213. |
| [79] |
Zhang L,Dong J,Bai K,et al. ResponsibleRobotBench:Benchmarking responsible robot manipulation using multi-modal large language models[PP/OL]. v1. arXiv(2025-12-03)[2026-03-31]. https://arxiv.org/abs/2512.04308. DOI:10.48550/arXiv.2512.04308. |
| [80] |
Zhao E,Raval V,Zhang H,et al.ManipBench:Benchmarking vision-language models for low-level robot manipulation[C]//The 9th Conference on Robot Learning,2025:3413-3462. |
| [81] |
Mandlekar A,Zhu Y K,Garg A,et al. Roboturk:A crowdsourcing platform for robotic skill learning through imitation[C]//Conference on Robot Learning,2018:879-893. |
| [82] |
Fu J,Kumar A,Nachum O,et al. D4RL:Datasets for Deep Data-Driven Reinforcement Learning[PP/OL]. v4. arXiv(2021-02-06)[2026-03-31]. https://arxiv.org/abs/2004.07219. |
国家自然科学基金项目(62401062)
中央高校科研业绩经费(FRF-TP-25-038)
Funding project: National Natural Science Foundation of China Project(62401062)
Fundamental Research Funds for the Central Universities(FRF-TP-25-038)
/
| 〈 |
|
〉 |