Enhanced Multi-UAV Path Planning in Complex Environments With Voronoi-Based Obstacle Modelling and Q-Learning
Wenjia Su, Min Gao, Xinbao Gao, Zhaolong Xuan
Abstract
To tackle the challenge of obstacle avoidance path planning for multiple unmanned aerial vehicles (UAVs) in intricate environments, this study introduces a Voronoi graph–based model to represent the obstacle-laden environment and employs a Markov decision process (MDP) for single UAV path planning. The traditional Q-learning algorithm is enhanced by adjusting the initial state of the Q-table and fine-tuning the reward and penalty values, enabling the acquisition of efficient obstacle avoidance paths for individual UAVs in complex settings. Leveraging the improved Q-learning algorithm for single UAVs, the Q-table is iteratively refined for a fleet of UAVs, with dynamic modifications based on the waypoints chosen by each UAV. This approach ensures the generation of collision-free paths for multiple UAVs, as validated by simulation results that showcase the algorithm’s effectiveness in learning from past training data. The proposed method offers a robust framework for practical UAV trajectory generation in complex environments.