UAV path planning via fusion of enhanced RRT* and path-guided Q-learning

This study proposes an unmanned aerial vehicle path-planning framework that integrates improved Rapidly-exploring Random Tree Star (improved RRT*) and Q-learning algorithms (Improved Rapidly-exploring Random Tree Star (IRRT*)-IQL) to address challenges such as low path-planning efficiency and susceptibility to local optima in complex environments, while also mitigating inter-unmanned aerial vehicle conflicts and enabling efficient multi-unmanned aerial vehicle cooperation. First, the RRT* algorithm is enhanced through the introduction of dynamic step size, dynamic neighborhood search radius, and goal-biased sampling strategies, which significantly improve its environmental exploration efficiency. Second, the resultant path point information is transferred to initialize the Q-table, providing the improved IQL algorithm with high-value state-action estimates. On this basis, the Q-learning algorithm is further enhanced with a dynamic learning rate, dynamic exploration rate, multiple reward functions, and an enhanced ε-greedy strategy to boost its path optimization capability. The planned path is smoothed using Bezier curves to generate a safe and smooth trajectory. Finally, path-planning experiments conducted in diverse static complex environments and varying numbers of unmanned aerial vehicles in dynamic complex scenarios show that the proposed algorithm effectively blends the rapid search capability and the autonomous learning ability, demonstrating competitive path quality together with improved convergence efficiency and effective multi-unmanned aerial vehicle coordination in the evaluated scenarios.