Path planning for mobile robot based on deep reinforcement learning with double experience replay

摘要

To further improve the convergence speed of the deep deterministic policy gradient (DDPG) algorithm in the mobile robot path planning task. In this paper, an improved DDPG algorithm with dual experience replay (DER) buffer based on prioritized and positive experience screening mechanism is proposed. Firstly, a prioritized experience replay buffer is introduced to store the basic samples. Then, an additional positive experience replay buffer is added to store samples with greater learning value by utilizing the positive experience filtering mechanism. The double experience replay buffer solves the low sample utilization, thereby effectively improves the stability of the algorithm in dynamic environments. Simulation experiments show that in the path planning task, compared with the traditional DDPG algorithm, the reward value obtained by the proposed algorithm is significantly improved, and the network convergence speed is further improved, which provides a new scheme for path planning of mobile robot.