Mobile Robot Path Planning Based On Improved DDQN Algorithm

摘要

A deep Reinforcement Learning (RL) algorithm integrating Prioritized Experience Replay (PER) & Double Deep Q-Network (DDQN) is proposed. It boosts training data sample quality via adjusting experience replay priority, brings in N-step bootstrap for more accurate Q-value estimation, & combines with dual Q-learning to solve traditional Q-learning's over-estimation. Compared with traditional DDQN & DDQN with PER, the DDQN-NstepPER algorithm is more stable in training. The Webots simulation experiment with Pioneer 3-DX robot shows it can achieve mobile robot path planning in complex environment with good effect.