Research On Role Adaptive Assignment Method For Human-Robot Cooperative System Based On Reinforcement Learning

With the development of science and technology,robots have shown an increasingly high level of intelligence.Robots become more actively communicate with human partner in order to achieve common task objectives.Therefore,how to achieve effective interaction between human partners and robot partners has become a new research issue in the field of intelligent control.Given the human-robot interaction ability and arbitration ability of intelligent control system is the way to achieve the cooperative task between human partners and robot partners,which has important academic value and practical significance.This paper is based on the coordinated and flexible movement of point-to-point between human and robot in transferring or gripping tasks.Firstly,aiming at how to realizes the perception and understanding of the human intentions in human-robot cooperation,the force information perception and human-robot coordinated motion control strategy are constructed.Secondly,to accomplish the cooperative tasks more flexibly and naturally,a role adaptive assignment method based on reinforcement learning is designed,which realizes the dynamic allocation of control rights between human and robot in the process of human-robot collaboration.Finally,the human-robot collaboration of point-to-point experiment verifies the research results of this paper.This paper mainly studies the following contents:(1)Firstly,to describe and assign the role relationship between the human partners and the robot partners,we established a shared control model,which can utilize both the intelligent knowledge of the human partners and the flexibility of the robot partners.Based on this model,the evaluation index and method of system collaboration performance are put forward.A comprehensive collaboration performance model in the process of collaboration is established to determine the allocation rules in deciding the role of control right in the system.(2)Secondly,in order to achieve coordinated motion,a human-robot coordinated motion control method based on force information is proposed for an effective interaction between human partners and the robot partners.In contact human-robot interaction,the control of the robot is accompanied by the interaction of forces,so a communication interface based on force information acquisition is established.A human-robot coordinated motion admittance control method based on force information is designed,and the influence of parameters in admittance control is simulated and analyzed.(3)Finally,aiming at how to adjust the roles between human partners and the robot partners in the human-robot collaboration system to achieve a flexible and natural movement form,an adaptive role assignment method for the human-robot collaboration system based on fuzzy reinforcement learning is proposed.The reinforcement learning model is constructed and a fuzzy reinforcement learning algorithm based on ? greedy strategy is designed.By constantly expanding knowledge and accumulating experience,the system achieves the dynamic allocation of control rights,and ultimately achieves the dual goals of improving sport compliance and reducing the efforts of human partners.