Control and design of multiple unmanned air vehicles for persistent surveillance

Control of multiple autonomous aircraft for search and exploration, is a topic of current research interest for applications such as weather monitoring, geographical surveys, search and rescue, tactical reconnaissance, and extra-terrestrial exploration, and the need to distribute sensing is driven by considerations of efficiency, reliability, cost and scalability. Hence, this problem has been extensively studied in the fields of controls and artificial intelligence. The task of persistent surveillance is different from a coverage/exploration problem, in that all areas need to be continuously searched, minimizing the time between visitations to each region in the target space. This distinction does not allow a straightforward application of most exploration techniques to the problem, although ideas from these methods can still be used. The use of aerial vehicles is motivated by their ability to cover larger spaces and their relative insensitivity to terrain. However, the dynamics of Unmanned Air Vehicles (UAVs) adds complexity to the control problem. Most of the work in the literature decouples the vehicle dynamics and control policies, but their interaction is particularly interesting for a surveillance mission. Stochastic environments and UAV failures further enrich the problem by requiring the control policies to be robust, and this aspect is particularly important for hardware implementations. For a persistent mission, it becomes imperative to consider the range/endurance constraints of the vehicles. The coupling of the control policy with the endurance constraints of the vehicles is an aspect that has not been sufficiently explored. Design of UAVs for desirable mission performance is also an issue of considerable significance. The use of a single monolithic optimization for such a problem has practical limitations, and decomposition-based design is a potential alternative. In this research high-level control policies are devised, that are scalable, reliable, efficient, and robust to changes in the environment. Most of the existing techniques that carry performance guarantees are not scalable or robust to changes. The scalable techniques are often heuristic in nature, resulting in lack of reliability and performance. Our policies are tested in a multi-UAV simulation environment developed for this problem, and shown to be near-optimal in spite of being completely reactive in nature. We explicitly account for the coupling between aircraft dynamics and control policies as well, and suggest modifications to improve performance under dynamic constraints. A smart refueling policy is also developed to account for limited endurance, and large performance benefits are observed. The method is based on the solution of a linear program that can be efficiently solved online in a distributed setting, unlike previous work. The Vehicle Swarm Technology Laboratory (VSTL), a hardware testbed developed at Boeing Research and Technology for evaluating swarm of UAVs, is described next and used to test the control strategy in a real-world scenario. The simplicity and robustness of the strategy allows easy implementation and near replication of the performance observed in simulation. Finally, an architecture for system-of-systems design based on Collaborative Optimization (CO) is presented. Earlier work coupling operations and design has used frameworks that make certain assumptions not valid for this problem. The efficacy of our approach is illustrated through preliminary design results, and extension to more realistic settings is also demonstrated.