Retrospective-Based Deep Q-Learning Method for Autonomous Pathfinding in Three-Dimensional Curved Surface Terrain

Path planning in complex environments remains a challenging task for unmanned vehicles. In this paper, we propose decoupled path-planning algorithm with the help of deep reinforcement learning that separates evaluation paths from to facilitate vehicles real-time consideration environmental factors. We use 3D surface map represent path cost, where elevation information represents integrated cost. The peaks function simulates which is processed and used as algorithm’s input. Furthermore, improved double Q-learning (DDQL), called retrospective-double DDQL (R-DDQL), improve performance. R-DDQL utilizes global incorporates retrospective mechanism employs fuzzy logic evaluate quality selected actions identify better states inclusion memory. Our simulation studies show proposed has training speed stability compared algorithm. demonstrate effectiveness under both static dynamic tasks.