Variable Impedance Control of Manipulator Based on DQN

For traditional constant impedance control, the robot suffers from constant stiffness, poor flexibility, large wear and high energy consumption in the process of movement. To address these problems, a variable impedance control method based on reinforcement learning (RL) algorithm Deep Q Network (DQN) is proposed in this paper. Our method can optimize the reference trajectory and gain schedule simultaneously according to the completion of task and the complexity of surroundings. Simulation experiments show that, compared with the constant impedance control, the proposed algorithm can adjust impedance in real time while manipulator is executing the task, which implies a better compliance, less wear and less control energy.