Improved Q-Learning Algorithm Based on Flower Pollination Algorithm and Tabulation Method for Unmanned Aerial Vehicle Path Planning
Lan Bo, Tiezhu Zhang, Hongxin Zhang, Jian Yang, Zhen Zhang, Caihong Zhang, Mingjie Liu
Abstract
Planning a path is crucial for safe and efficient Unmanned aerial vehicle flights, especially in complex environments. While the Q-learning algorithm in reinforcement learning performs better in handling such environments, it suffers from slow convergence speed and limited real-time capability. To address these problems, this study proposes an enhanced initialization process using the flower pollination algorithm and employs a tabulation method to improve local obstacle avoidance ability. An improved Q-learning algorithm based on the flower pollination algorithm and tabulation method (IQ-FAT) is proposed, which can perform both global and local path planning, enhance the convergence time of Q-learning, and expedite obstacle avoidance. Evaluation results on various obstacle maps demonstrate that the modified algorithm has a significant improvement convergence speed of approximately 40% compared to the original algorithm while enabling global path planning and local obstacle avoidance. Furthermore, the algorithm demonstrates superior path-planning capabilities in complex environments and enhances the dynamic response time of UAVs by approximately 90% compared to the artificial potential field method.