3D PG-RRT: UAV Path Planning Using a Gaussian-Mixture-Model-Driven, Kinematically Constrained RRT

Efficient navigation of unmanned aerial vehicles (UAVs) in three-dimensional (3D) environments requires advanced path-planning strategies that account for vehicle kinematic constraints to ensure smooth and feasible waypoint generation in complex and cluttered spaces. Such planning is critical to enable accurate trajectory tracking and stable control performance within the overall navigation pipeline. For real-time autonomous UAV operations, computationally efficient algorithms are essential to meet onboard processing limitations. This paper presents 3D PG-RRT algorithm, a probabilistic framework using Gaussian mixture model (GMM) based on Rapidly Exploring Random Tree (RRT) planner, designed to generate kinematically feasible waypoints for multirotor UAVs. The proposed approach integrates a Gaussian mixture model (GMM) into the RRT node-generation process to introduce a dynamic goal bias and reduce redundant node expansion. In addition, a special node mechanism is employed to enhance 3D search-space exploration and accelerate convergence. This targeted formulation significantly improves both sampling efficiency and path smoothness. The performance of the proposed planner is evaluated across two simulated complex 3D scenarios, demonstrating its capability to generate paths that adhere to vehicle constraints while minimizing unnecessary node expansion. Comparative analysis with state-of-the-art planning algorithms, RRT and RRT*, in the selected test scenarios shows that 3D PG-RRT achieves substantial improvements in both iteration count and runtime efficiency, highlighting its potential for real-time UAV navigation in realistic environments.