Research on Intelligent Control Decision of Manipulator Based on Deep Reinforcement Learning

For the general reinforcement learning algorithm, the training time will be too long due to the need for constant trial and error to provide sufficient data in the control of the robot arm. This paper uses unity3d engine to build a simulation environment, including UR robot and intelligent electricity meter. M-SAC algorithm, which combines multi-agent with SAC (Soft Actor Critical) algorithm, speeds up the training speed and shortens the training agent time. The efficiency of M-SAC algorithm is verified by analyzing the average reward value after training.