Research on Robot Path Planning Algorithm Based on Improved TD3

This paper is a research on robot path planning based on Improve Twin Delayed Deep Deterministic policy gradient (I-TD3) algorithm. Aiming at the problems of sparse rewards and waste of experience data in complex environments, the TD3 algorithm proposes to use the robot's explored experience as a reference point to guide the robot in path planning, which effectively improves the robot's action selection strategy in path planning in complex environments; meanwhile, it combines the reference point with the idea of artificial potential field to design a combined reward function, which mitigates the problem of sparse rewards; and it integrates the preferred hierarchical experience playback mechanism to reduce the waste of experience and improve the training efficiency of the algorithm. Experiments show that I-TD3 effectively improves the convergence speed of the network and its effectiveness in the path planning task.