Backtracking-Aware Path Planning for Mobile Robots via an Improved Q-Learning Approach

Path planning is a fundamental component of mobile robot navigation, directly impacting the robot’s ability to move autonomously, avoid obstacles, and perform tasks efficiently. Among existing approaches, reinforcement learning particularly Q-learning has shown promising results in enabling robots to learn navigation policies through interaction with the environment. However, traditional Q-learning-based methods often suffer from issues such as sub-optimal path selection and redundant movements, which can lead to undesirable behaviors like backtracking or turning back along previously visited paths. To overcome these limitations, this study proposes an improved Q-learning-based method to prevent backtracking behavior in mobile robot path planning. In our work, several strategies are proposed including a step penalty mechanism and a refined action selection policy to discourage path repetition. Proposed approach also considers flexible motion in a two-dimensional plane, where the robot can move in eight possible directions, offering greater maneuverability. Experimental evaluations conducted in both confined and large-scale environments demonstrate that the proposed method effectively reduces backtracking and improves path efficiency, outperforming conventional Q-learning in both accuracy and consistency.