Self-Supervised Robotic Arm Sorting Control Method Based on Dual-Channel VPG

The gripping problem of the robotic arm has been widely studied, and the control task after gripping is still important in this paper. This paper proposes a dual-channel predictive network structure based on the VPG (Visual Pushing for Grasping) algorithm, which improves the stability of the policy compared with VPG, and then extends the control task of the robotic arm to sorting and designs a self-supervised sorting algorithm without recognition and segmentation. The paper proposes a dual-channel fusion VPG network structure to improve the stability of the deep reinforcement learning policy. The simulation results show that the optimized grasping algorithm achieves 83.3%-97.8% grasping success rate in an environment with several randomly placed objects, which improves the stability and grasping success rate of the policy compared with the original algorithm VPG, and achieves 87%-100% grasping success rate and 83.5%-96.29% sorting success rate in the sorting task. It also reduces the probability of dropping the robot arm during the sorting and placing of objects.