Autonomous Path Planning and Estimation using UAVs

The main contribution of this work is an online path planning framework for cooperative search and localization using unmanned aerial vehicles. In this work, a team of vehicles is used to cooperatively search for a single stationary target. A vision-based sensing system, that incorporates position and attitude uncertainty, is used to develop probabilistic likelihood functions. By accounting for image frames where the target is not detected, a single framework may be used for both search and localization. Likelihood functions are fused using a Bayesian framework, over a grid. Cooperative sensing is accomplished by exchanging likelihood functions between vehicles. A receding horizon approach is used to plan paths. At each planning step, a gradient based optimization routine is used to maximize the probability of detecting the target in the image. A separate long-term planning strategy is used for cases where the vehicle is far from the target distribution. For the sake of cooperative planning, vehicles exchange expected likelihood functions, representing planned future measurements. Rather than developing a team-optimal solution at each time step, a greedy approach is used. This system has been implemented on the Berkeley UAV platform. Sensing and control are performed in a distributed manner, on each aircraft, in real-time. Data for cooperative sensing and planning is exchanged over a wireless ad-hoc network. Results from field trials demonstrate the viability of the receding horizon path-planning, in both single aircraft and multi-aircraft scenarios.