Abstract: 

Control systems often operate in environments with undesirable or uncontrolled influences which may potentially be adversarial, seeking to prevent the system from achieving a specified performance objective such as reachability, safety, or optimality. The quality of resilience of a control system informally refers to its ability to continue achieving its performance objectives, despite the presence of such adversarial influences. In this talk, Ram will describe two different approaches to designing a resilient control system. The first approach builds upon a novel constrained nonlinear control strategy using a linear driftless approximation to the original nonlinear system, which suffers a loss in control authority over some of its actuators. Constructing a partition of the time horizon into successively smaller intervals, it is shown that optimal control laws for the linear driftless ‘proxy’ system result in bounded error over each interval, thus converging to zero as the intervals become smaller. The second approach considers systems suffering an abrupt change in dynamics either through adversaries or system faults, modeling this setting through the framework of switched systems. A class of optimal control problems is proposed for such systems and impose a prefix constraint on controller parameters, which encodes causality with respect to the switching signal. Using system level synthesis (SLS), these problems are recast as convex programs, thus synthesizing optimal resilient controllers efficiently. I will also describe possible extensions under nonlinear dynamics, and improving computational effort through clustering and limiting memory.

Biography:  

Ram Padmanabhan is a final-year PhD candidate in Electrical and Computer Engineering at the University of Illinois Urbana-Champaign, advised by Melkior Ornik. His research interests broadly include problems in control, learning, and optimization. His current research focuses on resilience quantification and resilient design for systems whose dynamics are affected by system faults or adversaries. Ram received his BTech from PES University, Bengaluru in 2021 and MS from the University of Michigan, Ann Arbor in 2023. He is a recipient of the Joan and Lalit Bahl Fellowship.