Path planning algorithm based on improved DDQN in gridded environments

To address the issues of slow convergence speed and low path quality in path planning, this study proposes a novel robot path planning algorithm in a grid environment based on the Double Deep Q-Network (DDQN) algorithm. First, convergence speed is enhanced by designing a dual experience pool with positive and negative samples and adjusting the sampling ratio to achieve a balance between exploration and exploitation. Second, the artificial potential field method is refined by optimizing the direction of repulsive force generation, and convergence speed is further improved by replacing random exploration with the enhanced artificial potential field approach. Finally, path quality is elevated by eliminating redundant nodes through an analysis of the positional relationship between the path and obstacles, followed by path smoothing using Brazier curves. Simulation results conducted in a MATLAB environment demonstrate that, compared to the original DDQN algorithm, the improved DDQN algorithm achieves a 52.9 % increase in convergence speed and a 27.39 % reduction in path length. Furthermore, relative to the traditional DDQN algorithm, the proposed method exhibits faster convergence and produces smoother paths.