Manipulator Reinforcement Learning with Mask Processing Based on Residual Network

In the field of intelligent manufacturing, manipulators are expected to have a higher level of learning ability to master skills. In this paper, the method of visual support is adopted to enable the UR5 manipulator to learn the skills of grasping object blocks in a random environment. The deep reinforcement learning algorithm can effectively learn visual support mapping to guide the grasping action of the manipulator. To verify the designed algorithm, a manipulator grasping system and test scenario are built on the CoppeliaSim simulation platform. The simulation results show that the deep reinforcement learning algorithm combined with the residual network (ResNet) obtains good success rate for the manipulator to grasp the target block. The experimental results show that the method significantly improves the grasping efficiency of the manipulator.