Energy-Efficient Motion Planning and Control for Robotic Arms via Deep Reinforcement Learning

Abstract

With the optimization of China’s energy supply structure, the issue of efficient energy use has attracted widespread attention from researchers [1]. The robotic arm, which is widely used in industrial production, comes with high energy usage. In order to reduce energy usage, we propose an energy-efficient motion planning and control method based on Reinforcement Learning (RL) to reduce energy consumption during operation. The motion planning and control policy is learned in a model-free manner using RL, making it effective even in complex industrial environments. The main goal can be divided into three parts: moving to the target location, obstacle avoidance, and energy saving. From this, a comprehensive and effective reward function is designed using a distance reward, a velocity reward, an obstacle avoidance reward, and an energy-saving reward. Using this reward function, a control policy is trained which balances the aforementioned goals. Finally, the algorithm is verified by using the trained policy to avoid obstacles in a simulation environment. Extensive experimental results have demonstrated the effectiveness of the algorithm proposed, and show that the learned control policy can make the robot move with less energy whilst avoiding obstacles.