VAMAV: Visual Collision Detection and Avoidance for a Micro Air Vehicle

This proposal presents the VAMAV system (Visual Awareness/Avoidance for a Micro Air Vehicle) for monocular visual collision detection and avoidance. Real time collision detection will be implemented using a time-to-collision approach based on area moment estimates of expansion from globally optimal image segmentation. This approach does not require explicit optical flow computations, and avoids noisy flow correspondence and differentiation. Real time collision avoidance will be implemented using rapidly exploring random trees (RRTs), which generate feasible trajectories given known obstacles and popup threats, while taking into account differential constraints arising from vehicle dynamics. The objectives of the phase II investigation are to (i) demonstrate open loop system performance on flight video, (ii) characterize operational envelope in simulation, (iii) demonstrate simulated closed loop system performance, (iv) demonstrate closed loop system performance on the ground station in an urban flight experiment, (v) predict the achievable framerate and power consumption for an ASIC hardware implementation. Scientific Systems Company Inc. (SSCI) is joined in this investigation by team members Sarnoff Corporation and Brigham Young University.