Precision UAV Landing in Unstructured Environments

Abstract

Autonomous landing of a drone is a necessary part of autonomous flight. One way to have high certainty of safety in landing is to return to the same location the drone took-off from. Implementations of return-to-home functionality fall short when relying solely on GPS or odometry as inaccuracies in the measurements and drift in the state estimate guides the drone to a position with a large offset from the initial position. This can be particularly dangerous if the drone took-off next to something like a body of water. Current work on precision landing relies on localizing to a known landing pattern, which requires the pilot to carry a landing pattern with them. We propose a method using a downward facing fisheye lens camera to accurately land a UAV from where it took off on an unstructured surface, without a landing pattern. Specifically, this approach uses a position estimate relative to the take-off path of the drone to guide the drone back. With the large Field-of-View provided by the fisheye lens, our algorithm can provide visual feedback starting with a large position error at the beginning of the landing, until 25cm above the ground at the end of the landing. This algorithm empirically shows it can correct the drift error in the state estimation and land with an accuracy of 40cm.