Robotic Control for Human–Robot Collaborative Assembly Based on Digital Human Model and Reinforcement Learning

Bitao Yao, Hongzhou Ai, Wenjun Xu, Zude Zhou, Lihui Wang

In human–robot collaboration (HRC), safe and efficient robotic control requires a deep understanding and realistic modeling of human motion behavior. However, current human motion data collection is inefficient in real HRC and constrained by safety considerations, resulting in limited data coverage of HRC processes. This paper proposes a method for generating human trajectories in collaborative assembly based on a digital human model and reinforcement learning. A muscle‐driven musculoskeletal dynamic model of the digital human is developed, and reinforcement learning is employed to generate synthetic human motion trajectories that conform to biomechanical constraints. Furthermore, these trajectories are used to train collaborative robot control policies by reinforcement learning, enabling adaptive trajectory planning and real‐time obstacle avoidance. The synthetic trajectories are evaluated using Fitts’ law and compared with experimental data. The experimental results show that the synthetic trajectories exhibit consistency with real human motion. The proposed robotic control method can effectively improve the efficiency and safety of human–robot collaborative assembly.