Grant: Engage Grants Program

Vision-based control of unmanned systems is gaining momentum within the research community making themomni present. As an example, utilization of high-resolution cameras on board unmanned systems can providepose estimations comparable to that in military-grade Inertial Measurement Units (IMUs) at a much lower costfactor. Furthermore, the information obtained via these sensors can be used for mapping and measurement(AKA, photogrammetry) purposes as well. However, since they would provide only relative information, ratherthan absolute like that in inertial systems, extra care must be exercised to account for dynamic features in thescene. Simply put, the optical flow in sequential images could be due to the movement of the seen objectsand/or the camera/UAV motion. To distinguish between these two one would have to either (1) fuseinertia-based measurements obtained via on board inertial sensors as an information filter, or (2) developstatistical methods to distinguish static from dynamic features in a video stream.Two major foreseeable challenges in vision-based control of UAVs are: (1) detection, tracking, and sequentialmatching of static and dynamic features, and (2) real-time operation.This proposal aims at developing vision-based control/navigation algorithms for the dual-ducted VTOL UAVdeveloped by the 4Front Robotics. Major application domains are: (1) detecting safe landing spots, (2) visualservoing the UAV to a safe landing, (3) visual odometry, (4) tracking a mobile target, and (5) providing metricmeasurements of static/dynamic objects on fly. This research focuses on indoor liftoff within a possiblycollapsible environment. It also assumes that the environment would consist of a mix of objects with staticand/or dynamic features located in any direction w.r.t. the UAV and potentially surrounding the aircraft.