Distributed DDDAS through Receding Horizon Control

UAVs [1, 2], Disaster Management [3], and more recently the Internet of Things (IoT) [4]. Merging of the DDDAS paradigm with the IoT raises multiple new issues. IoT is based on the basic philosophy of distributed sensing, control, and actuation. In other words, there is no central designer or controller that can serve as a repository of information and source of commands. Rather, each device is equipped with sensors, processors, and communication devices and multiple such devices need to coordinate to achieve the desired task. In DDDAS terms, neither the application system nor the data generation is at a central location. Multiple platforms need to coordinate. This coordination raises interesting challenges and opportunities. As a problem domain, we choose an ISR (Intelligence, Surveillance, and Reconnaissance) UAV-based application where, as in the IoT domain, distributed sensors and information processing produce new capabilities and challenges [5-7]. It is clear that if a centralized processor could access all the information being generated by the various sensors and then relay to the vehicles the points they should move to, we will be in a \"classical\" DDDAS paradigm. However, this will lead to immense data flow and processing load at the central processor. Instead a distributed (or at least hierarchical) implementation is desired. In this implementation, local information flow must lead to coordinated decisions being made by the UAVs that lead to guarantees on global performance for the ISR mission. We set up such a problem and identify the various challenges. In this discussion, we consider two specific challenges. The first is the issue of distributed / hierarchical control of UAV trajectories to achieve global performance. We propose the use of distributed receding horizon control for this purpose. Receding horizon control is a time-honored methodology in which control inputs are calculated and implemented in a sliding window fashion. Although distributed versions of the method have been proposed, they often make strict assumptions about the cost function being decomposable (e.g. in formation control) and the constraints being local (e.g. collision avoidance). The challenge of relating local optimization problems to a global cost function remains open. We present a hierarchical solution to the problem. The second issue we consider is that of processing overload at each processor that may not have been designed to carry out computation intensive DDDAS applications. We present some initial results. Specifically, we identify methods to carry out control computations only when desired and to the accuracy that is required for desired levels of performance. Simulation results are provided to show the efficacy of the two approaches. We also present some initial experimental data and plans to validate the theory using both simulation and a physical swarm of UAVs.