Autonomous navigation for large Unmanned Aerial Vehicles (UAVs) is fairlystraight-forward, as expensive sensors and monitoring devices can be employed.In contrast, obstacle avoidance remains a challenging task for Micro AerialVehicles (MAVs) which operate at low altitude in cluttered environments. Unlikelarge vehicles, MAVs can only carry very light sensors, such as cameras, makingautonomous navigation through obstacles much more challenging. In this paper,we describe a system that navigates a small quadrotor helicopter autonomouslyat low altitude through natural forest environments. Using only a single cheapcamera to perceive the environment, we are able to maintain a constant velocityof up to 1.5m/s. Given a small set of human pilot demonstrations, we use recentstate-of-the-art imitation learning techniques to train a controller that canavoid trees by adapting the MAVs heading. We demonstrate the performance of oursystem in a more controlled environment indoors, and in real natural forestenvironments outdoors.
translated by 谷歌翻译