Collision Avoidance and Local Guidance Based on Insect Visual Motion Processing

Flying insects display remarkable capabilities for local guidance and obstacle avoidance based on their visual sense. If similar performance could be realized in autonomous flying vehicles and weapons, it would greatly enhance their effectiveness by enabling near-ground flight in urban, forest, and indoor/in-cave environments. We have investigated the integration of biomimetic visual motion detection and flight control for collision avoidance in Phase I. Visual processing is based on adaptive motion detection as observed in flies. Adaptation has the effect of reducing detector dependence on non-motion-related parameters of visual stimuli, improving estimates of optic flow. The control theory is based on a model for wide-field neurons (tangential cells) that spatially integrate local optic flow estimates. The outputs of these integrators are applied as feedback signals for direct control of vehicle dynamics. When combined, these algorithms provide a reflexive or inner control loop for avoiding collisions and centering in clutter. In Phase II we propose to develop a demonstration of the combined algorithms in autonomous vehicles, develop an implementation suitable for use in a small UAV, integrate the lower-level reflexive guidance with higher- or mission-level control, and establish a road map for integration into an operational micro-UAV in Phase III.