Motion planning framework based on dual-agent DDPG method for dual-arm robots guided by human joint angle constraints

• DOI: 10.3389/fnbot.2024.1362359
• Publication Date: 2024-02-22T04:44:17Z
• Authors: Keyao Liang; Fusheng Zha; Wei Guo; Shengkai Liu; Pengfei Wang; Lining Sun
• Keywords: reinforcement learning; 0202 electrical engineering, electronic engineering, information engineering; human experience constrains guidance; Neurosciences. Biological psychiatry. Neuropsychiatry; 02 engineering and technology; motion parameter mapping; trajectory planning; dual-agent depth deterministic strategy gradient; RC321-571; Neuroscience

Introduction Reinforcement learning has been widely used in robot motion planning. However, for multi-step complex tasks of dual-arm robots, the trajectory planning method based on reinforcement still some problems, such as ample exploration space, long training time, and uncontrollable process. Based dual-agent depth deterministic strategy gradient (DADDPG) algorithm, this study proposes a framework constrained by human joint angle, simultaneously realizing humanization content style. It quickly plans coordinated tasks. Methods The proposed mainly includes two parts: one is modeling angle constraints. calculated from arm data measured inertial measurement unit (IMU) establishing human-robot kinematic mapping model. Then, range constraints are extracted multiple groups demonstration expressed inequalities. Second, segmented reward function designed. constraint guides exploratory process form step reward. Therefore, space reduced, speed accelerated, controllable to certain extent. Results discussion effectiveness was verified gym simulation environment Baxter robot's reach-grasp-align task. results show that framework, experience knowledge significant impact guidance learning, can more plan