Mobile Robot Navigation System Using Reinforcement Learning with Path Planning Algorithm

E.W. Andargea, A. Ordysa, Y.M. Abebeb aFaculty of Mechatronics, Institute of Micromechanics and Photonics, Warsaw University of Technology, św. A. Boboli 8, 02-525 Warsaw, Poland bCollege of Engineering, Department of Electrical and Computer Engineering, Addis Ababa Science and Technology University, Akaki Kality Sub-City, Addis Ababa, Ethiopia

Full Text PDF

Navigation systems are critical for the practical implementation of autonomous vehicles. While classical and modern control methods have been explored, simultaneous localization and mapping have limitations due to their complexity and sensor requirements. As a result, there is growing interest in using artificial intelligence algorithms. This paper proposes to use a reinforcement learning algorithm agent as a local planner for obstacle avoidance, coupled with rapidly exploring random trees for global path planning. Unlike prior methods relying solely on reinforcement learning or point of interest for global path planning, this approach integrates reinforcement learning with random trees to ensure efficient navigation in complex environments. By mapping the environment and generating optimal global paths, divided into intermediate waypoints, the proposed method facilitates efficient navigation toward the goal. Experimental results demonstrate the effectiveness of the approach, particularly in navigating through intricate environments, offering promising advancements in autonomous navigation systems.

DOI:10.12693/APhysPolA.146.452 topics: simultaneous localization and mapping (SLAM), artificial intelligence (AI), reinforcement learning (RL), rapidly exploring random trees (RRT)