This project is to develop a dynamical systems model of distributed computation motivated from recent work on the distributed computation of averages. The key idea is that static optimization problems (particularly convex optimization problems) can be solved by designing a dynamic system that stabilizes around the optimal solution of the problem. Moreover, when the optimization problem is separable, then the designed dynamic system decomposes into a set of locally-interacting dynamic systems. This is expected to open a door to a host of new computational approaches that take advantage of recent developments in control engineering including robust control (providing a mechanism for errors introduced by discretization), Markovian Jump Linear Systems (providing a mechanism for random discretization time), event-driven control (providing a mechanism for assured asynchronous execution), control over networks (providing a mechanism for improved performance of distributed computational systems in general). The new approach is essential in emerging applications, where the optimization runs on physically separated agents, operating in a noisy environment and communicating over unreliable channels. As a test bed, the project will make use of a two-vehicle robotic system developed by the PI designed to monitor a crop of corn plants, where the dynamic systems perspective of this grant will, for example, allow for distributed optimal estimation toward the goal of optimal station-keeping. By studying how natural systems can collectively compute and optimize, this research has potential to impact many disciplines involving networked systems, from controlling the electric power grid, to modeling the behavior of social, biological or economic systems. It is directly applicable to cooperative networked multi-agent systems like robotic search and rescue missions and disaster-relief operations, distributed machine learning problems, and intelligent systems. An intriguing mix of motivating applications and theoretical problems offer a unique multidisciplinary educational opportunity to students who will be involved in the project, and provide exciting innovative material for courses and labs. Software developed will be distributed as open source via the CPS Virtual Organization.

This project explores balancing performance considerations and power consumption in cyber-physical systems, through algorithms that switch among different modes of operation (e.g., low-power/high-power, on/off, or mobile/static) in response to environmental conditions. The main theoretical contribution is a computational, hybrid optimal control framework that is connected to a number of relevant target applications where physical modeling, control design, and software architectures all constitute important components. The fundamental research in this program advances state-of-the-art along four different dimensions, namely (1) real-time, hybrid optimal control algorithms for power management, (2) power-management in mobile sensor networks, (3) distributed power-aware architectures for infrastructure management, and (4) power-management in embedded multi-core processors. The expected outcome, which is to enable low-power devices to be deployed in a more effective manner, has implications on a number of application domains, including distributed sensor and communication networks, and intelligent and efficient buildings. The team represents both a research university (Georgia Institute of Technology) and an undergraduate teaching university (York College of Pennsylvania) in order to ensure that the educational components are far-reaching and cut across traditional educational boundaries. The project involves novel, inductive-based learning modules, where graduate students team with undergraduate researchers.