Multiple-Level Predictive Control of Mobile Cyber Physical Systems with Correlated Context

Abstract:

With the increasing popularity of mobile computing, cyber physical systems are merging into major mobile systems of our society, such as public transportation, supply chain systems, and taxi networks. Researchers have accumulated abundant knowledge for designing cyber physical systems, such as military surveillance, infrastructure protection, scientific exploration, and smart environments, mostly in relatively stationary settings, i.e., where spatial diversity is limited. Differently, Mobile CPSs interact with phenomena of interest at different locations and environments, and where the context information (e.g., network availability and connectivity) about these physical locations might not be available. This unique feature calls for new solutions to seamlessly integrate mobile computing with physical world processes by sharing information among networked mobile CPSs. Specifically, this work is motivated by one important observation: contextual information are spatiotemporal correlated. Although such correlation decays over distance and time, it is possible (i) to predict context information with an evaluated uncertainty by dynamic sharing of data collected by mobile units that are spatiotemporally close, (ii) to coordinate and control mobile CPSs using predicted context information, and (iii) to employ predictive control for performance assurance based on correlated observations and user specifications. In this work, we choose taxi control and coordination networks and home health care as two primary application domains. The objectives of future taxi control and dispatch systems include: (i) reducing time for drivers to find customers; (ii) reducing time for passengers to wait, (iii) avoiding and preventing traffic congestion; (iv) reducing gas consumption and operating cost; and (v) improving driver and vehicle safety. To achieve these goals, rich contextual information (GPS readings, the amount of gas remaining, destination of current passengers, state of driver and vehicle, pictures and video taken from vehicle carried cameras, wireless link quality, and a priori knowledge of the normal traffic patterns per day and time of day) need to be collected and analyzed. Various control commands (route changes, safety warnings, and driving directions) must be delivered to taxi drivers. Using real data sets we have develop a passenger demand model for a large city, solutions for minimizing taxi cruising miles, and a scheme for sharing cabs when demand exceeds supply. For home health care we are interested in human-in-the-loop control and systems of systems, two key underlying aspects of CPS. A challenge paper in this area of work has been published. New preliminary models of human-in-the-loop control have been constructed. A new architecture for CPS systems of systems has been designed. Further design, implementation and evaluation of these latter two topics are planned in the near future. In this first year, papers supported, in part, from this project have appeared or have been accepted to IEEE RTSS, ACM Sensys, IEEE BigData 2013, IEEE Transactions on Emerging Topics in Computing, and the 8th International Workshop on Feedback Computing.