| This thesis explores mobile cooperative sensing using teams of unmanned aerial vehicles. Much of the existing work involving mobile sensing leverages advances in estimation; while these estimation schemes are well developed, the cooperation and control strategies to steer these platforms are not equally mature. This work is aimed at developing a framework for multi-agent path planning to optimize information-based objective functions. The goal is to develop a cooperative system, that directly addresses the perceptual and action uncertainty of each vehicle, to handle a wide variety of sensing problems.;A cooperative, receding-horizon path planning approach is used to allow for the incorporation of vehicle constraints, sensor models and information objectives. Uncertainty is considered from a sensing standpoint, but also with respect to the path of the vehicle. The use of an information-based objective function allows for the path planning scheme to be tightly coupled with the sensing system. By stochastically modeling the sensor, information-maximizing paths can be planned, that explicitly account for the goal of uncertainty minimization.;To account for a variety of sensing problems, a variable horizon planner is developed, that maintains a common planning horizon across the team of vehicles. To ensure computational feasibility, the trajectory generation algorithm may vary online, as a function of the planning horizon. Cooperation is maintained through the exchange of predicted actions. A fully optimal approach is eschewed in favor of a greedy algorithm. In support of a general path planning scheme, this work also develops a vision-based sensor model, and considers different methods of characterizing uncertainty, for particle filters and grid-based schemes.;These techniques have been applied to multi-aircraft search and localization problems, in both simulation and flight tests. Flight results illustrate that receding horizon path planning is a viable method for multi-vehicle cooperative estimation. The use of a common path planning framework is effective, despite different sensing problems. These results indicate the value of cooperation, variable horizon planning and the need for stochastic sensor modeling. |
