Enhancing Navigation Efficiency in Robotics with PRM-DDPG

Abstract

This study describes a new way to plan the paths of mobile robots. It combines the large-scale, global planning power of Probabilistic Roadmaps (PRM) with the local, flexible decision-making power of Deep Reinforcement Learning (DRL). While PRM focuses on waypoint delineation, Deep Deterministic Policy Gradient (DDPG) focuses on real-time obstacle avoidance. By integrating these two approaches, the proposed PRM-DDPG algorithm significantly enhances the robot's navigation capabilities, allowing it to effectively handle both structured and complex environments. In the performed simulations, PRM-DDPG outperforms sampling-based methods, such as PRM and RRT in terms of path length, time efficiency, and obstacle avoidance, especially in difficult environments. In addition, the PRM-DDPG algorithm produced the shortest path of 27.0182 m with only six corners, while methods, such as ID3QN and Genetic Algorithm (GA), produced longer paths with more corners. Fewer corners indicate a smoother and more direct path. The results show that using both PRM and DDPG together produces paths that are faster and smoother than those produced by classical or pure machine learning methods alone. The proposed PRM-DDPG algorithm will advance mobile robotics by enabling smarter, more flexible, and more effective self-navigation systems for real-world applications.