文献类型：期刊文献

英文题名：Behavioral planning and parameter meta learning for embodied intelligence robots in adaptive assembly

作者：Chen, Baotong[1,2];Xu, Guangjun[1,2];Wang, Lei[1,2];Jiang, Chun[3,4];Zhang, Zelin[1,2];Wang, Zhaohui[1,2];Xia, Xuhui[1,2]

机构：[1]Wuhan Univ Sci & Technol, Key Lab Met Equipment & Control Technol, Minist Educ, Wuhan 430081, Peoples R China;[2]Wuhan Univ Sci & Technol, Hubei Key Lab Mech Transmiss & Mfg Engn, Wuhan 430081, Peoples R China;[3]Guangdong Ocean Univ, Sch Mech Engn, Zhanjiang 524088, Peoples R China;[4]Guangdong Key Lab Precis Equipment & Mfg Technol, Guangzhou 510641, Peoples R China

年份：2026

卷号：49

外文期刊名：JOURNAL OF INDUSTRIAL INFORMATION INTEGRATION

收录：SCI-EXPANDED(收录号：WOS:001621075400002)、、WOS

基金：This work was supported by the National Natural Science Foundation of China (Grant No. 62203340, No. 72471181, and No. 52275503) , the Natural Science Foundation Innovation Group Project in Hubei Province (Grant No. 2024AFA026) , the Scientific Research Program of Depart-ment of Education of Hubei Province (Grant No. Q20241106) , the Guangdong Key Laboratory of Precision Equipment and Manufacturing Technology Open Research Project (Grant No. PEMT202404) , and the Program for Scientific Research Start-up Funds of Guangdong Ocean University (Grant No. 060302062314) .

语种：英文

外文关键词：Embodied intelligence (EI) robots; Parameter meta-learning; Behavioral planning; Low-code/no-code; Adaptive assembly

外文摘要：Embodied intelligence (EI) is an emerging frontier in robotics that tightly integrates perception, action, and cognition. By continuously interacting with their environments, EI robots can self-evolve and adapt to uncertainties in flexible assembly tasks, thereby enhancing adaptability and execution efficiency. This paper proposes a behavioral planning and parameter meta learning approach for EI robots in adaptive assembly, with the aims of enabling low-code/no-code execution in complex assembly scenarios. This method leverages sensors to capture real-time environmental data and adopts a blackboard mechanism for information storage and sharing, thereby ensuring seamless data flow. The synergistic integration of PDDL-based reasoning with behavior tree orchestration is deployed to achieve dynamic behavior planning. Furthermore, a motion feedback-driven closed loop for parameter meta learning and behavior evolution is constructed based on the PEARL (Probabilistic Embedding for Actor-Critic Reinforcement Learning) and SAC (Soft Actor-Critic) algorithms. The proposed method was validated through a series of hole-and-axis assembly simulations under interference conditions. In addition, we evaluated robustness under different tolerances. The framework maintained a success rate of over 94% and stable adaptive latency under all tolerance levels, with faster adaptation speed, higher precision, and better efficiency.