Computer Science · Computer Vision and Pattern Recognition
Mobile Robot Path Planning Based on Optimized TD3 Neural Network
Wenwei Li, Zhongliang Deng, Pinhua Xie, Guoqing Xu
See on the mapAbstract
This paper explores the core issues of path planning for mobile unmanned robots and evaluates the performance of three path planning algorithms based on deep reinforcement learning (TD3, Opt_TD3, Opt_TD3+LSTM) in different simulation environments. Experimental results show that the traditional TD3 algorithm tends to get stuck in local optima, while the Opt_TD3, which incorporates obstacle avoidance modules and heuristic functions, shows improvements but lacks stability. In contrast, the Opt_TD3+LSTM algorithm, which integrates LSTM, demonstrates superior performance in terms of policy selection and stability, effectively adapting to complex environments.