Jan 26, 2018
Lopata Hall, Room 101
Two Challenge Problems with Intelligent Physical Systems
This two-part talk describes opportunities at the intersection of machine learning, control systems and the use of formal methods for a new generation of safe and scalable Intelligent Physical Systems.
1. Energy Systems: Bridging Machine Learning and Control Systems for Volatile Energy Markets
In January 2014, the east coast (PJM) electricity grid experienced an increase in the price of electricity from $31/MWh to $2,680/MWh MWh - an 83x increase in 5mins. Demand response (DR) is becoming increasingly important as the volatility on the grid continues to increase. Current DR approaches are predominantly manual and rule-based or involve deriving first principles based models which are extremely cost and time prohibitive to build. To this end, we develop data-driven approaches that bridge machine learning and controls for volatile energy markets. Specifically, we present data-driven methods (1) for optimal experiment design of functional tests to learn dynamics of a real building subject to stringent operational constraints, (2) to synthesize control-oriented models for receding horizon control, and (3) to continuously improve the learned model in closed-loop with a real-time controller. Our algorithms generate predictive models using Random Forests and Gaussian Processes - where we can not only predict the state of the building but also generate control strategies with high confidence using only historical weather, schedule, set-points and electricity consumption data. We call this approach Data Predictive Control (DPC). We show that, for a realistic building model, control strategies generated by DPC are remarkably similar to Model Predictive Control (MPC), while being scalable, unlike MPC.
2. Autonomous Systems: A Driver's License Test for Driverless Vehicles
Autonomous vehicles (AVs) have already driven millions of miles on public roads, but even the simplest maneuvers such as a lane change or vehicle overtake have not been certified for safety. Current methodologies for testing and verification of Advanced Driver Assistance Systems (ADAS) such as Adaptive Cruise Control cannot be directly applied to determine AV safety as the AV actively makes decisions using its perception, planning and control systems for both longitudinal and lateral motion. These systems increasingly use machine learning for which it is fundamentally hard to derive safety guarantees across a range of driving scenarios and environmental conditions. New approaches are needed to bound and minimize the risk of AVs to assure the public, determine liability and insurance pricing and ensure the long term growth of the domain.
So what type of evidence should we require before giving a driver's license to an autonomous vehicle? I will describe our research in the design of an autonomous vehicle computer-aided design toolchain, which captures formal descriptions of driving scenarios in order to develop an AV safety case. Rather than focus on a particular component of the AV, like adaptive cruise control, the toolchain models the end-to-end dynamics of the AV in a formal way suitable for testing and verification.
Rahul is an Associate Professor in the Dept. of Electrical & Systems Engineering and Dept. of Computer & Information Science at the University of Pennsylvania. His interests are in cyber-physical systems at the intersection of formal methods, machine learning and controls. He is the Penn Director for the Department of Transportation's $14MM Mobility21 University Transportation Center. He also directs Penn's Embedded Systems Masters Program.
Rahul received the 2016 US Presidential Early Career Award (PECASE) from President Obama for his work on Medical Cyber-Physical Systems. He also received the 2016 Department of Energy's CLEANTECH Prize (Regional), the 2014 IEEE Benjamin Franklin Key Award, 2013 NSF CAREER Award, 2012 Intel Early Faculty Career Award and was selected by the National Academy of Engineering for the 2012 US Frontiers of Engineering. He was the Stephen J. Angelo Assistant Professor from 2008-2013. He received his Ph.D. in Electrical & Computer Engineering from Carnegie Mellon University where he also received his MS and BS in 2007, 2002 and 2000 respectively.