Sorting operation method of manipulator based on deep reinforcement learning

Abstract

Radioactive waste sorting often faces an unstructured and locally radioactive working environment. At present, remote operation sorting has problems such as low sorting efficiency, greater difficulty in operation, longer training periods for personnel, and poor autonomous control capabilities. Based on the premise of improving the adaptability and autonomous operation ability of robots in an unstructured environment, this paper uses the dual deep Q learning algorithm to optimize the classic deep Q learning algorithm to improve training speed and improve sorting efficiency and stability. Secondly, the sorting algorithm model of deep reinforcement learning is used to determine the optimal behavior in this state. Set up multiple sets of simulations and physical experiments to verify the sorting method. The results show that the robotic arm can autonomously complete sorting tasks under complex conditions and can significantly improve work efficiency when pushing and grasping collaborative operations and will preferentially grasp objects with high radioactivity in the radioactive area. The algorithm has migration ability and good generalization.