Dynamic Obstacle Avoidance for UAV Motion Planning in Unknown Environments Based on Probabilistic Inference

Safe, agile, and reliable Unmanned Aerial Vehicle (UAV) autonomous navigation in the unknown environment requires the motion planning module's fast reaction to the potential collision caused by obstacles. However, avoiding dynamic obstacles remains a challenge in most existing work. In this paper, we propose a modified dynamic rapidly exploring random tree star (RRT*) algorithm to obtain a fine time allocation initial solution connected to the goal state while reducing sampling failures. We further optimize the solution by probabilistic inference-based Gaussian Process Motion Planning (GPMP2), which infers a feasible trajectory on the factor graph, and our method uses a novel dynamic obstacle factor to adapt the predicted information's uncertainty. Finally, the simulation results demonstrate our method's capability to travel in unknown environments when dynamic obstacles appear.