Motion Control of Flexible-Joint Robotic Arms for Variable-Station Warehouse Sorting Based on Proximal Policy Optimization

Abstract

In variable-station warehouse sorting scenarios, the motion control of flexible-joint robotic arms must simultaneously address the triple-coupled challenges of flexible joint characteristics, dynamic workstation environments, and sorting task requirements, making arm motion control of the robotic arm highly complex. To address this, a motion control method for flexible-joint robotic arms in variable-station warehouse sorting is proposed, based on Proximal Policy Optimization (PPO algorithm). After analyzing the variable-station warehouse sorting model and the robotic arm's control system architecture, a motion control model based on Proximal Policy Optimization is constructed. This model maps the robotic arm as an agent, by designing a multidimensional state space that encompassing station coordinates and cargo status. It divides the action space into overall arm movement and end-effector rotation, establishing a reward function incorporating continuous rewards, sparse rewards, and penalties. An LSTM is introduced to capture temporal motion correlations, predicting advantageous function values under different actions as workstation coordinates change. The PPO algorithm obtains the robotic arm motion control commands with the highest cumulative reward value—such as angular velocity and torque for each joint, along with gripper opening degree (gripping force)—for robotic arm motion control. Experiments demonstrate that this method achieves position control errors as low as 0.1 mm and gripping force errors reduced to 0.05N for flexible-joint robotic arms in variable-workstation warehouse sorting. Sorting speeds reach 30 items per minute, meeting the high-precision and high-robustness control demands of variable-workstation warehouse sorting.