A Probabilistic Inference-Based Efficient Path Planning Method for Quadrotors

This article proposes the probabilistic inference-based local path planner, a local trajectory planning method for quadrotor unmanned aerial vehicles (UAVs). The trajectory planning problem is formulated as the maximum a posteriori (MAP) problem. The Gaussian process (GP) is utilized, and various distribution functions are designed to construct a comprehensive probabilistic model that meets the quadrotor's local trajectory planning requirements. The model is then constructed as a factor graph for the implementation of the inference algorithm. A marginal inference method named belief propagation (BP) is employed to solve the desired trajectory from the factor graph model. Utilizing the chain structure of the trajectory and the sparse property of the GP, the BP method could guarantee efficient and exact marginal computation. Besides, a trajectory inference framework is designed to deploy the algorithm on the resource-constrained quadrotor platform. Validated through numerical simulation and practical flight experiments, the proposed strategy enables the rapid computation of smooth and safe local trajectories for quadrotor UAVs. It can ensure more reliable real-time trajectory planning compared with existing quadrotors' trajectory planning methods.