Robotic Path Planning by Q Learning and a Performance Comparison with Classical Path Finding Algorithms

Abstract

—Q Learning is a form of reinforcement learning for path finding problems that does not require a model of the environment. It allows the agent to explore the given environment and the learning is achieved by maximizing the rewards for the set of actions it takes. In the recent times, Q Learning approaches have proven to be successful in various applications ranging from navigation systems to video games. This paper proposes a Q learning based method that supports path planning for robots. The paper also discusses the choice of parameter values and suggests optimized parameters when using such a method. The performance of the most popular path finding algorithms such as A* and Dijkstra algorithm have been compared to the Q learning approach and were able to outperform Q learning with respect to computation time and resulting path length.