Coordinated Transportation of Dual-arm Robot Based on Deep Reinforcement Learning

In the transportation task of the dual-arm robot, a closed-chain tightly coordinated constraint is formed between the two arms and the object, which faces many challenges for motion planning because of the complex kinematic and dynamic coupling relationship. Therefore, this paper proposes a method for dual-arm robot cooperative transporting tasks based on deep reinforcement learning (DRL), which avoids the complicated and time-consuming motion planning process. In this paper, we establish a closed chain model of a dual-arm transport system, which trained by Soft Actor-Critic (SAC) algorithm. The simulation shows the feasibility together with the advantages of short decision time and high success rate. This method provides a theoretical basis for further research on dual-arm coordinated transportation by DRL.