Real-Time Pose-Aware Dynamic Path Planning

This work investigates efficient means of generating low-risk paths for unmanned aerial vehicles (UAVs) in the presence of radar sites. This problem is difficult to solve in real time due to the probabilistic nature of the radar detection model as well as the anisotropy of radar cross section of the vehicle. This couples the vehicle's path and orientation into the objective and constraints, leading to challenging optimization problems. In this work, we propose a novel approach that is capable of generating risk-aware maneuvers in real time in these environments. The proposed approach consists of three modules: a global planner, a local planner, and an imitation learning module. The global planner aims to generate a goal-reaching path which is then refined by the local planner to reduce the risk. Then imitation learning technique is applied to reduce the inference time of the path generation process. Simulation experiments are performed to demonstrate both the efficiency (93.6% reduction in computation time on average) and the generalization capability of the proposed method.