Real-time Perception-Limited Motion Planning Using Sampling-based MPC

Abstract

Motion planning with visual perception is a hot topic for autonomous flight of Micro Aerial Vehicles (MAVs). However, many existing works fail to be implemented in realistic scenarios in real-time due to practical constraints, such as the limited field of view (FOV) of the onboard camera and the limited computational capability. Compared to existing methods, the proposed approach solves the optimization of motion and perception at the same time. A sampling-based Model Predictive Control (MPC) framework is explored as a local planner to generate trajectories which are dynamically feasible and collision-free with limited perception. The sampling-based local planning framework is extended to two independent scenarios for MAVs: a) planning safe trajectories with limited FOV constraint; b) planning trajectories with effective perception of point of interest (POI). The effectiveness of the proposed method is demonstrated through both simulation and real flight experiments.