Ant Colony Enhanced Q-Learning Algorithm for Mobile Robot Path Planning

Abstract

Q-learning algorithm is a common method for solving robot path planning problems. However, it is not only significantly time-consuming, but also prone to getting stuck in local optima. This paper presents an enhanced approach to the Q-learning algorithm by incorporating ant colony optimization, which leads to better convergence of the Q-learning algorithm and reduces the number of iterations and computational time. Firstly, the Q-table and initial paths are constructed through interactive training between the agents and the environment. Secondly, the ant colony algorithm dynamically updates the pheromone matrix and iteratively adapts the path search strategy in the Q-table. Experimental results illustrate that the enhanced Q-learning algorithm exhibits superior performance compared to the original Q-learning algorithm in path planning simulations, as indicated by three evaluation metrics. Furthermore, improvements are observed in terms of the length of the shortest path and the quantity of turning points, yielding smoother paths.