Small UAV High-speed Obstacle and Collision Avoidance

Currently fielded UAVs cannot robustly detect and avoid obstacles on their own, which leaves the bulk of their capacity unused. While ideally one would like to build UAVs which can “see and avoid” as well as human pilots, in practice this level of autonomy cannot be reached in the short-term. Current technology, however, when carefully integrated and refined for UAVs, has enough sophistication to achieve the level of guidance for the bulk of UAV missions of today. We pursue a multi-level approach where sensing, planning, and control form the on-board guidance system. At the sensing level we will combine radar, laser, and vision sensors, chosen based on their strong complementary nature when it comes to 3D reconstruction, all-weather operation, field-of-view, and resolution. At the perception level we pursue a planning and route deconfliction approach combining a local kinodynamic planner, a global D*-based route planner, and integration with high-level ATC-like flight management systems. At the control level we will build simulation models of different UAVs to be able to customize each sensor configuration to the dynamics of the vehicle. We expect to deliver a prototype system that is practical and applicable to a large range of small UAVs.