Dynamic Estimation of Obstacle Postion with Vision Sensing for Reactive Collision Avoidance of UAVs

Abstract

It is very crucial for unmanned aerial vehicles to have autonomous obstacle detection and avoidance capability for their survivability during flight. This paper proposes and validates the application of extended Kalman filter for online obstacle position estimation with a vision based sensor and the usefulness of this information with two recently developed guidance algorithms for collision avoidance. The vision sensor is assumed to continuously sense the environment in front of the vehicle during flight. In case any obstacle is detected, the information from this sensor is then utilized in the filter to estimate the obstacle position online. Simultaneously, the collision cone approach is applied to predict any potential collision in future and, in case of a potential threat, to steer away the vehicle in order to avoid the collision. This is done by first computing a suitable ‘aiming point’ towards which the velocity vector of the vehicle must be aligned as soon as possible and then by using either of two recently proposed guidance laws, namely nonlinear geometric guidance and differential geometric guidance (which are identically same with appropriate gain correlation, but otherwise are different) to achieve this objective. Exhaustive simulation studies show that this overall strategy is fairly successful.