Towards Autonomous UAV Flight in Forests

This paper presents results demonstrating real-time six degree of freedom localization, mapping, navigation and obstacle avoidance in an outdoor environment using only a lowcost off-the-shelf inertial measurement unit and a monocular camera. This navigation system is intended for operation of small unmanned aerial vehicles when GPS signals are unavailable due to obstructions or jamming and when operating in cluttered environments such as urban canyons or forests. A small radio-controlled car is used as a test bed. A bearings-only Simultaneous Localization and Mapping algorithm was implemented to localize both the vehicle and obstacles. This paper describes: (a) the hardware used; (b) a two-step approach for data association (image frame to image frame followed by image frame to map); (c) a technique for landmark initialization for the case where landmarks are located on the ground. Hardware test results demonstrating navigation to a goal in an obstacle strewn environment are presented. The effect of unmodelled sensor biases is examined in simulation.