Grasping Force Optimization and DDPG Impedance Control for Apple Picking Robot End-Effector

Abstract

To minimize mechanical damage caused by an apple picking robot end-effector during the apple grasping process, and on the basis of optimizing the minimum stable grasping force of apple, a variable impedance control strategy based on a reinforcement learning deep deterministic policy gradient (DDPG) algorithm is proposed to achieve compliant grasping control for apples. Firstly, according to the apple contact force model, the gradient flow algorithm is adopted to optimize grasping force in terms of the friction cone, force balancing condition, and stability assessment index and to obtain a minimum stable grasping force for apples. Secondly, based on the analysis of the influence of impedance parameters on the control system, a variable impedance control based on the DDPG algorithm is designed, with the reward function adopted so as to improve the control performance. Then, the improved control strategy is used to train the optimized impedance control. Finally, simulation and experimental results indicate that the proposed variable impedance control outperforms the traditional impedance control by reducing the peak grasping force from 4.49 N to 4.18 N while achieving a 0.6 s faster adjustment time and a 0.24 N narrower grasping force fluctuation range. The improved impedance control successfully tracks desired grasping forces for apples of varying sizes and significantly reduces mechanical damage during apple harvesting.