UavisBug: vision-based 3D motion planning and obstacle avoidance for a mini-UAV in an unknown indoor environment

Abstract

This paper introduces a new class of Bug algorithm that combines local planning with global information to guarantee convergence. Existing Bug algorithms are typically designed for 2D robot navigation in 2D environments and do not integrate vision information within their planning algorithm. In this paper a new algorithm, uavisBug, is proposed for small quadrotor UAV navigation in a 3D environment. On-board vision perception is incorporated to enable 3D way-finding behaviors in an initially unknown environment. The proposed algorithm shows non-oscillatory behaviors in the UAV's flight in proximity of obstacles. It is also immune to potential local minima exhibited by artificial force field approaches. In addition to system and sensors modeling, UAV control, UAV pose estimation, and fast environment mapping are also included. Simulation results are used to validate the proposed approach.