Huili Yu and Randal W. Beard
path planning, collision avoidance, computer vision, miniature air vehicles
This paper presents a vision-based three-tiered path planning and collision avoidance scheme for small and miniature air vehicles (MAVs), which consists of global, local, and reactive path planners. The global path planner generates inertial paths to the goal configuration based on a known world map that may not necessarily be up-to-date. The local path planner builds maps in the local-level frame of the MAV based on vision data and plans paths to manoeuvre around obstacles with time-to-collision horizon around 3–10 s. The reactive path planner reacts to visually sensed obstacles with time-to-collision horizon less than 3 s by pushing the images of the obstacles to the edges of the image plane, thus avoiding obstacles. A strategy that effectively combines the three path planners is designed to globally guide the MAV to the goal configuration while locally avoiding obstacles. Simulation results show the scheme is successful in solving path planning and obstacle avoidance problems for fixed-wing MAVs in urban terrain.
Important Links:
Go Back
