Robot Path Planning Method Based on an Improved TD3 Algorithm

Traditional path planning methods often rely heavily on environmental maps, while deep reinforcement learning (DRL) algorithms face challenges in achieving stable policy convergence in complex environments. To overcome these challenges, this study introduces an enhanced Twin Delayed Deep Deterministic Policy Gradient (TD3) algorithm for autonomous path planning in unknown environments. First, temporally correlated the algorithm integrates Ornstein-Uhlenbeck (OU) noise to improve action exploration. Second, an Uncertainty-Weighted TD3 (UW-TD3) mechanism is introduced, which fuses Q-value estimates based on confidence weights to enhance stability and accelerate convergence. Finally, Hindsight Experience Replay (HER) is integrated to improve sample efficiency, enabling the agent to learn effectively from unsuccessful experiences. Simulation results under static as well as dynamic obstacle scenarios demonstrate that the proposed method significantly improves planning success rates and training efficiency, outperforming conventional TD3 and DDPG algorithms.