Multi-target Approaching Control of Hyper-redundant Manipulators using Reinforcement Learning.

Hyper-redundant manipulators have superior dexterity. However, conventional control based on accurate modeling is difficult to achieve. This paper proposes a control method for the hyper-redundant manipulator to complete target approaching tasks using deep reinforcement learning. A hyper-redundant manipulator simulation model was first built and the corresponding forward kinematics equation was analyzed. Then, we used reinforcement learning methods to construct controllers for single target tasks, during which the impact of several key factors on the training was analyzed and optimized accordingly. On this basis, the multi-target task training was achieved by expanding the state to a generalized state. Finally, the proposed control method was verified in a dynamic simulation environment. Simulation results show the effectiveness of the proposed control method.