Robotic joint space variable impedance controller design method based on reinforcement learning

Abstract

The application discloses a robot joint space variable impedance controller design method based on reinforcement learning, the method obtains a robot joint space impedance control law by analyzing a robot joint space dynamics model and combining a physical representation of a second-order system of an expected robot joint dynamic response, and designs robot joint impedance control parameters on the basis; in combination with joint position, speed, torque value and torque change speed and other information acquired by a robot sensor, a reward function is designed on demand, an approximate Q function is obtained through a linear method, a feature matrix parameter is obtained by using a time difference method, and the designed variable impedance parameter is obtained. The method can be used for impedance control of a robot system, can better realize adaptive variable impedance dynamic characteristics of the robot facing unknown environment physical interaction, and is an important, effective and low-cost means for improving flexible interaction between the robot and the environment.

Images