Optimal path planning method based on epsilon-greedy Q-learning algorithm

Path planning in an environment with obstacles is an ongoing problem for mobile robots. Q-learning algorithm increases its importance due to its utility in interacting with the environment. However, the size of state space and computational cost are the main parts to be improved. Hence, this paper proposes an improved epsilon-greedy Q-learning (IEGQL) algorithm to enhance efficiency and productivity regarding path length and computational cost. It is important to determine an effective reward function and adjust the agent's next action to ensure exploitation and exploration. We present a new reward function to ensure the environment's knowledge in advance for a mobile robot. Additionally, novel mathematical modeling is proposed to provide the optimal selection besides ensuring a rapid convergence. Since a mobile robot has difficulty moving through the path with sharp corners, the smooth path is formed after obtaining the optimal skeleton path. Furthermore, a real-world experiment is given based on the multi-objective function. The benchmark of the proposed IEGQL algorithm with the classical EGQL and A-star algorithms is presented. The experimental results and performance analysis indicate that the IEGQL algorithm generates the optimal path based on path length, computation time, low jerk, and staying closer to the optimal skeleton path.