Integrated perception and control at high speed

Abstract

We present a method for robust high-speed quadrotor flight through unknown cluttered environments using integrated perception and control. Motivated by experiments in which the difficulty of accurate state estimation was a primary limitation on speed, our method forgoes maintaining a map in favor of using only instantaneous depth information in the local frame. This provides robustness in the presence of significant state estimate uncertainty. We compare the method against a benchmark approach using a simulated quadrotor race through a forest at high speeds in the presence of increasing state estimate noise. We then present hardware validation experiments in both indoor and outdoor environments, performing robust obstacle avoidance at speeds of up to 10 m/s, including sustained flight through a forest at 6 m/s. Finally, we add to the memoryless method, and develop a robust obstacle avoidance approach that uses memory without resorting to a maximum-likelihood mapping framework.