UTRC CDS Lecture: Ketan Savla, "Dynamical Analysis and Control Synthesis for Traffic Systems"
Friday, December 11, 2015
1146 A V Williams
301 405 6576
UTRC Control and Dynamical System Invited Lecture
Dynamical Analysis and Control Synthesis for Traffic Systems: from Microscopic to Macroscopic
University of Southern California
The emergence of autonomous mobility and dynamic control mechanisms in traffic systems provide new opportunities to improve efficiency and resilience. Motivated by these technological trends, we present novel analysis and control tools for microscopic and macroscopic modeling regimes.
In the first part, we present throughput analysis for a horizontal traffic queue, where vehicles arrive into a line with directed flow according to a spatio-temporal Poisson process, and depart after traveling a distance which is identically and independently sampled from an exponential distribution. When inside the system, the instantaneous speed of every vehicle is proportional to power m of the distance to the vehicle in front. We provide an exact characterization of the throughput in the linear (m=1) case, and provide tight probabilistic bounds in the nonlinear case. Our analysis relies on workload as a proxy for queue length in the super-linear (m >= 1) case, and on minimum inter-vehicular distance using order statistic in the sub-linear (m<1) case. We conclude this part by providing an interpretation of the proposed traffic queue as a processor sharing queue, and on using appropriate fluid limit arguments to derive the corresponding macroscopic traffic flow model.
In the second part, we consider continuous time optimal control for macroscopic traffic flow over networks, also known as system optimum dynamic network traffic assignment. Such problems are known to be non-convex due to the underlying traffic flow dynamics. We adopt a known relaxation for the uncontrolled dynamics, and design control policies to make the optimal solution to the relaxed problem feasible with respect to the original dynamics, thereby giving zero optimality gap. We also identify scenarios where the optimal solution to the relaxed problem is readily feasible. Time permitting, distributed implementation for the proposed optimal control design will also be presented.
Ketan Savla is an Assistant Professor of Civil and Environmental Engineering, Industrial and Systems Engineering, and Electrical Engineering at the University of Southern California. Prior to joining USC, he was a research scientist in the Laboratory for Information and Decision Systems at MIT. He received his PhD in Electrical Engineering from the University of California at Santa Barbara. His current research interests include distributed robust and optimal control, dynamical networks, state-dependent queueing systems and mechanism design, with applications in infrastructure systems and robotics. His recognitions include CCDC Best Thesis Award from UCSB and NSF CAREER.