AUTONOMOUS MOBILE ROBOT PATH PLANNING USING EINFORCEMENT LEARNING IN MATLAB

Abstract

Path planning is a critical component of autonomous mobile robots, allowing them to navigate their environment efficiently while avoiding obstacles. Traditional algorithms such as A* and Dijkstra’s provide optimal solutions based on predefined maps but struggle with adaptability in dynamic and unknown environments. These limitations necessitate the adoption of machine learning-based approaches such as Reinforcement Learning (RL). RL, particularly Q-learning and Deep Q-Networks (DQN), enables robots to learn optimal navigation strategies through continuous interaction with their surroundings, allowing them to adapt to environmental changes. This research implements and evaluates Q-learning and DQN in MATLAB, training autonomous robots in a simulated environment to navigate obstacles while optimizing path efficiency. The study compares RL-based path planning with traditional algorithms (A* and Dijkstra’s) by assessing computational complexity, adaptability, and performance in dynamic scenarios. Key findings highlight the effectiveness of RL-based methods in improving realtime decision-making and adaptability, but also expose challenges such as high computational demands and convergence issues. The research provides insights into the applicability of RL for autonomous robot navigation and proposes potential optimizations for real-world deployment.