LAUSR

Motion planning in unknown environments suffers from the limited sensing range of cameras and occlusions. Therefore, perception awareness (PA) should be fully considered to ensure flight safety. Specifically, the unknown part of the planned trajectory should be sensed in advance (visibility constraint) at a safe distance (safety constraints). However, most of the optimization-based methods considered PA merely in the optimization stage in terms of the occlusions and ignored the limitation of the sensing model. As a result, the PA constraints remain violated due to poor initialization. Meanwhile, existing sampling methods can only ensure the visibility constraint rather than the safety constraint. In this paper, an optimization-based planning framework is proposed to solve the raised problems in real-time. Specifically, in the path-finding stage, a fast PA checking algorithm based on incremental convex sensing hulls is proposed and combined with the motion primitive propagation to find the initial path that satisfies the PA constraints. Subsequently, in the trajectory optimization stage, perception awareness with respect to occlusion and sensing model are jointly considered and formulated into the nonlinear optimization problem based on a grid map. Benchmarks with the state-of-the-art and physical experiments show that the proposed planner can generate safe trajectories with the highest success rate in different occluded environments and achieve comparable efficiency and aggressiveness simultaneously.