Dynamic human-like motion planning method of human–robot collaboration based on movement primitives and double DQN

Purpose The purpose of this paper is to address the low acceptance and declining collaborative willingness caused by hard-to-understand, unpredictable robot motions and neglected human feelings in physical human–robot collaborative motion (pHRCM). To this end, this paper proposes a dynamic human-like motion planning (DHMP) method based on movement primitives and double deep Q-network (DQN). Design/methodology/approach In the method, the data sets of physical human–human collaborative motion (pHHCM) and pHRCM are established, and the human arm movement primitives of pHHCM are analyzed. On this basis, a double DQN model based on long short-term memory (LSTM) is established. The model can autonomously decide the primitives type based on perceiving and understanding the collaborator’s motion. Then, a DHMP method, including admittance control, movement primitives decision-making, primitives parameter estimation and joint design, is designed to realize the robot’s dynamic motion planning. Finally, the pHRCM experiment based on DHMP is carried out. Findings The experimental results show that compared with the previous methods, the motion planning method proposed in this paper has significant improvement in the aspects of heart rate change, mood fluctuation and acceptance of robot motion. Originality/value A double DQN model based on LSTM is proposed to solve the problem of autonomous selecting movement primitives based on feedback from collaborators. Meanwhile, A four-module DHMP framework is proposed to address the problem of robot motion dynamic replanning caused by real-time target change in pHRCM.