Functional and Distributional Control of Ensemble Systems using Moment Kernel Machines

November 2, 2023, 3:30 pm – 4:30 pm

Location: Student Success Center Cleary Theatre

Jr-Shin Li

Newton R. and Sarah Louisa Glasgow Wilson Professor

 Department of Electrical and Systems Engineering

Washington University in St. Louis

Abstract

Recent years have witnessed a wave of research activities in systems science and engineering toward control and learning for dynamic population systems. This shift was geared by numerous emerging and ever-changing technologies from neuroscience, biology, and quantum physics to robotics, where many control-enabled applications involve manipulating a large ensemble of structurally identical dynamic agents. Such ensemble control tasks are challenged by the underactuated nature where control and observation can only be made at the population level. In this talk, problems of functional and distributional control of ensemble systems with respect to different types of aggregated measurements related to system labels will be discussed. The moment kernel machine transforming these vast-scale (infinite-dimensional in the limit) control problems to a dual space of the state space will be introduced. This moment transform quantizes ensemble systems and reveals distinctive structures which enable tractable systems-theoretic analysis, computation, control design, and dynamic learning. The generalization and connection of the moment kernel method to optimal transport, data-driven control, and control on fibered manifolds will be illustrated. 

Biography

Dr. Jr-Shin Li is Newton R. and Sarah Louisa Glasgow Wilson Professor in the Department of Electrical and Systems Engineering at Washington University in St. Louis, where he also holds a joint appointment in the Division of Computational \& Data Sciences (DCDS) and the Division of Biology \& Biomedical Sciences (DBBS). Dr. Li received his B.S. and M.S. from National Taiwan University, and his Ph.D. in Applied Mathematics from Harvard University in 2006. His research interests lie in the areas and at the intersection of systems, computational, and data sciences, and their applications to biology, neuroscience, quantum physics, brain medicine, public health, and control engineering. He is a recipient of the NSF Career Award in 2008 and the AFOSR Young Investigator Award in 2010. He is currently Associate Editor of the SIAM Journal on Control and Optimization (SICON) and Editorial Member of Nature Scientific Reports.

A Control Lyapunov Function Approach for Particle Nanomanipulation via Optical Tweezers

October 13, 2023 11:00 am – 12:00 pm

Location: TSRB Auditorium, 132

Mark Spong

Professor

Systems Engineering

University of Texas at Dallas

Abstract

In this talk we investigate the problem of stabilization of a spherical particle trapped inside an optical tweezer using a Control Lyapunov Function (CLF) approach.  The proposed CLF framework enables nonlinear optimization-based closed-loop control using optical tweezers and serves as a first step towards design of effective control algorithms for nanomanipulation of biomolecules. After deriving necessary and sufficient conditions for having smooth uniform CLFs for the optical tweezer control system under study, we present a static nonlinear programming problem (NLP) for generation of robustly stabilizing feedback control inputs. An appealing feature of the proposed CLF framework is the ability to constrain temperature increases due to laser heating through proper inequalities encoded in the CLF based NLP. Numerical simulations demonstrate the effectiveness of the proposed control framework in the presence of external disturbances and initial bead positions that are located far away from the laser beam.

Biography

Mark W. Spong received the D.Sc. degrees in systems science and mathematics in 1981 from Washington University in St. Louis. He has held faculty positions at Lehigh University, Cornell University, the University of Illinois at Urbana-Champaign, and the University of Texas at Dallas. He is currently Professor of Systems Engineering and holds the Excellence in Education Chair at the University of Texas at Dallas.  From 2008-2017 he was the Dean of the Erik Jonsson School of Engineering and Computer Science at UT-Dallas. 

Dr. Spong is Past President of the IEEE Control Systems Society and Fellow of both the IEEE and IFAC.  His main research interests are in robotics, mechatronics, and nonlinear control theory. He has authored or coauthored more than 300 technical articles in control and robotics, seven books and holds one patent. He has made fundamental contributions in robust and nonlinear control of robot manipulators, teleoperators, bipedal walking robots, and multi-robot systems.

His notable awards include the 2020 Rufus Oldenberger Medal from the ASME, the 2018 Bode Lecture Prize from the IEEE Control Systems Society, the 2016 Nyquist Lecture Prize from the ASME, the 2011 Pioneer in Robotics Award from the IEEE Robotics and Automation Society, the  first IROS Fumio Harashima Award for Innovative Technologies in 2007, the IEEE Transactions on Control Systems Technology Outstanding Paper Award, the Senior Scientist Research Award from the Alexander von Humboldt Foundation, the Distinguished Member Award from the IEEE Control Systems Society, the John R. Ragazzini and O. Hugo Schuck Awards from the American Automatic Control Council, and the IEEE Third Millennium Medal.

Data-driven Framework for Stability, Performance, and Safety Using Linear Transfer Operators

April 28, 2023 11:00 am – 12:00 pm

Location: TSRB 132

Umesh Vaidya

Professor

Mechanical Engineering

Clemson University

Abstract

This talk will present results on applying linear transfer operator theory involving Perron-Frobenius and Koopman operators for data-driven control problems. In the first part of this talk, I will show the results of developing Koopman theory for control dynamical systems. One of the main challenges in using Koopman theory for control problems arises due to the bilinear lifting of the control dynamical system. We circumvent the bilinear lifting problem by establishing a connection between the spectrum of the Koopman operator and the Hamilton Jacobi (HJ) equation. We show that the solution to the HJ equation can be extracted from the spectrum of the Koopman operator. The HJ equation is the cornerstone of various problems in control theory, including optimal control, robust control, input-output analysis, dissipativity theory, and reachability/safety analysis. The connection between the Koopman spectrum and the HJ solution opens the possibility of exploiting the Koopman spectrum for various control problems. One of the main advantages of using the Koopman theory is that the Koopman operator and its spectrum can be approximated using data. We present novel approaches for computing the Koopman spectrum from data, thereby leading to systematic convex optimization-based methods for solving the HJ equation with application to optimal control design and input-output analysis of a nonlinear system.

In the second part of this talk, we will present results involving the Perron-Frobenius operator for the convex formulation of the optimal control problem with safety constraints. The convex problem is formulated over the space of densities defined only over the state space. The convex formulation is attractive for multiple reasons. First, the convex optimization problem can be constructed based on the data-driven approximation of the Koopman operator, dual to the Perron-Frobenius operator. Second, the convex incorporation of safety constraints allows us to provide a novel approach for the analytical construction of density functions for navigation. The proposed density function is used for navigation in a complex environment and high dimensional configuration space. The proposed construction overcomes the problem associated with navigation based on navigation functions, which are known to exist but challenging to construct, and potential functions that suffer from the existence of local minima. Finally, we demonstrate the application of the developed results for controlling the robotic system and vehicle autonomy.

Biography

Dr. Umesh Vaidya received a Ph.D. in Mechanical Engineering from the University of California at Santa Barbara, Santa Barbara, CA, in 2004. He was a research engineer at the United Technologies Research Center (UTRC), East Hartford, CT. Dr. Vaidya is a Professor of Mechanical Engineering at Clemson University, SC. Before joining Clemson University in 2019, and since 2006, he was a faculty member with the Department of Electrical and Computer Engineering at Iowa State University, Ames, IA. He is the recipient of the 2012 National Science Foundation CAREER award. His current research interests include dynamical systems and control theory with application to power systems, robotic systems, and vehicle autonomy.

What makes learning to control easy or hard?

April 14, 2023 11:00 am – 12:00 pm

Location: TSRB auditorium

Nikolai Matni

Assistant Professor

Department of Electrical and Systems Engineering

University of Pennsylvania

Abstract

Designing autonomous systems that are simultaneously high-performing, adaptive, and provably safe remains an open problem.  In this talk, we will argue that in order to meet this goal, new theoretical and algorithmic tools are needed that blend the stability, robustness, and safety guarantees of robust control with the flexibility, adaptability, and performance of machine and reinforcement learning.  We will highlight our progress towards developing such a theoretical foundation of robust learning for safe control in the context of two case studies: (i) characterizing fundamental limits of learning-enabled control, and (ii) developing novel robust imitation learning algorithms with finite sample-complexity guarantees.  In both cases, we will emphasize the interplay between robust learning, robust control, and robust stability and their consequences on the sample-complexity and generalizability of the resulting learning-based control algorithms.

Biography

Nikolai Matni is an Assistant Professor in the Department of Electrical and Systems Engineering at the University of Pennsylvania, where he is also a member of the Department of Computer and Information Sciences (by courtesy), the GRASP Lab, the PRECISE Center, and the Applied Mathematics and Computational Science graduate group.  He has held positions as a Visiting Faculty Researcher at Google Brain Robotics, NYC, as a postdoctoral scholar in EECS at UC Berkeley, and as a postdoctoral scholar in the Computing and Mathematical Sciences at Caltech. He received his Ph.D. in Control and Dynamical Systems from Caltech in June 2016. He also holds a B.A.Sc. and M.A.Sc. in Electrical Engineering from the University of British Columbia, Vancouver, Canada. His research interests broadly encompass the use of learning, optimization, and control in the design and analysis of autonomous systems.  Nikolai is a recipient of the NSF CAREER Award (2021), a Google Research Scholar Award (2021), the 2021 IEEE CSS George S. Axelby Award, and the 2013 IEEE CDC Best Student Paper Award.  He is also a co-author on papers that have won the 2022 IEEE CDC Best Student Paper Award and the 2017 IEEE ACC Best Student Paper Award.

A Finite-Sample Analysis of Payoff-Based Independent Learning in Zero-Sum Stochastic Games

April 7, 2023 11:00 am – 12:00 pm

Location: TSRB auditorium

Dr. Zaiwei Chen

CMI postdoctoral fellow 

The Computing + Mathematical Sciences (CMS) Department 

California Institute of Technology

Abstract

We study two-player zero-sum stochastic games, and propose a form of independent learning dynamics called Doubly Smoothed Best-Response dynamics, which combines a discrete and doubly smoothed variant of the best-response dynamics with temporal-difference (TD)-learning and minimax value iteration. The resulting dynamics are payoff-based,  convergent, rational, and symmetric among players.  Our main results provide finite-sample guarantees. In particular, we prove the first-known $\tilde{\mathcal{O}}(1/\epsilon^2)$ sample complexity bound for payoff-based independent learning dynamics, up to a smoothing bias. In the special case where the stochastic game has only one state (i.e., matrix games), we provide a sharper $\tilde{\mathcal{O}}(1/\epsilon)$ sample complexity. Our analysis uses a novel coupled Lyapunov drift approach to capture the evolution of multiple sets of coupled and stochastic iterates, which might be of independent interest.

Biography

Dr. Zaiwei Chen is currently a CMI postdoctoral fellow in The Computing + Mathematical Sciences (CMS) Department at California Institute of Technology, hosted by Dr. Adam Wierman and Dr. Eric Mazumdar. Zaiwei obtained a Ph.D. degree in Machine Learning, an M.S. degree in Mathematics, and an M.S. degree in Operations Research from Georgia Institute of Technology, where he was advised by Dr. Siva Theja Maguluri and Dr. John-Paul Clarke. Before that, Zaiwei obtained his B.S. degree in Electrical Engineering at Chu Kochen Honors College, Zhejiang University.

Zaiwei was a recipient of the Simoudis Discovery Prize, and was named a PIMCO Postdoctoral Fellow in Data Science in 2022. His Ph.D. thesis won the Sigma Xi Best Ph.D. Thesis Award, and was selected as a runner-up for the 2022 SIGMETRICS Doctoral Dissertation Award. Before that, Zaiwei received the ARC-TRIAD Student Fellowship in 2021, and was selected as as one of 7 nominees to represent Georgia Institute of Technology at the 2021 Schmidt Science Fellows Award Competition. A proposal based on his research received The IDEaS-TRIAD Research Scholarship in 2020.

Tunable Control Barrier Functions for Multi-Agent Safety via Trust Adaptation

March 31, 2023 11:00 am – 12:00 pm

Location: TSRB 509

Dimitra Panagou

Associate Professor

Department of Robotics

Department of Aerospace Engineering

University of Michigan

Abstract

We will present some of our recent results and ongoing work on safety-critical control synthesis under state, time and input constraints, with applications to non-cooperative multi-agent systems and, time permitting, spacecraft control applications. The proposed framework aims to eventually develop and integrate adaptive, learning and control methods towards provably-correct and computationally-efficient mission synthesis for multi-agent systems in the presence of constraints and uncertainty.

Biography

Dimitra Panagou received the Diploma and PhD degrees in Mechanical Engineering from the National Technical University of Athens, Greece, in 2006 and 2012, respectively. In September 2014 she joined the Department of Aerospace Engineering, University of Michigan as an Assistant Professor. Since July 2022 she is an Associate Professor with the newly established Department of Robotics, with a courtesy appointment with the Department of Aerospace Engineering, University of Michigan. Prior to joining the University of Michigan, she was a postdoctoral research associate with the Coordinated Science Laboratory, University of Illinois, Urbana-Champaign (2012-2014), a visiting research scholar with the GRASP Lab, University of Pennsylvania (June 2013, Fall 2010) and a visiting research scholar with the University of Delaware, Mechanical Engineering Department (Spring 2009).

Dr. Panagou’s research program spans the areas of nonlinear systems and control; multi-agent systems and networks; motion and path planning; human-robot interaction; navigation, guidance, and control of aerospace vehicles. She is particularly interested in the development of provably-correct methods for the safe and secure (resilient) operation of autonomous systems in complex missions, with applications in robot/sensor networks and multi-vehicle systems (ground, marine, aerial, space). Dr. Panagou is a recipient of the NASA Early Career Faculty Award, the AFOSR Young Investigator Award, the NSF CAREER Award, and a Senior Member of the IEEE and the AIAA.

More details: http://www-personal.umich.edu/~dpanagou/research/index.html

Information-Theoretic Approach to Gaussian Belief Space Path Planning for Minimum Sensing Navigation

February 10, 2023 11:00am – 12:00pm

Location: TSRB auditorium

Ali Reza Pedram

Ph.D.

University of Texas at Austin

Abstract

Motion planning and strategic sensing are inseparable problems for autonomous robots navigating in uncertain environments under perceptual resource constraints. In this talk, a new path planning methodology for a mobile robot in an obstacle-filled environment to generate a reference path that is traceable with moderate sensing efforts will be discussed. In this framework, the desired reference path is characterized as the shortest path in an obstacle-filled Gaussian belief manifold equipped with a certain information-geometric distance function. The distance function introduced can be interpreted as the minimum information gain required to steer the Gaussian belief. An RRT*-based numerical solution algorithm is presented to solve the formulated shortest-path problem. The asymptotic optimality of the proposed path planning algorithm will also be discussed. A smoothing algorithm will be presented to remove the possible sharp turns, which are common in sampling-based planners, in the output of the proposed algorithm. Finally, simulation results will be presented demonstrating that the proposed method is effective in various robot navigation scenarios to reduce sensing costs, such as the required frequency of sensor measurements and the number of sensors that must be operated simultaneously.

Biography

Ali Reza Pedram received the B.Sc. degrees in mechanical engineering and applied physics from the Sharif University of Technology, Tehran, Iran, in 2015, and the M.S. degree in mechanical engineering from the Sharif University of Technology in collaboration with the Max Planck Institute for Intelligent Systems, Stuttgart, Germany, in 2017. He is currently working toward the Ph.D. degree in mechanical engineering with the University of Texas at Austin, Austin, TX, USA. His research interests include motion planning, information theory, stochastic control, and optimization.

Building Certifiably Safe and Correct Large-scale Autonomous Systems

November 11, 2022 11:00am – 12:00pm

Location: Technology Square Research Building 118 auditorium

Chuchu Fan

Assistant Professor

Department of Aeronautics and Astronautics

MIT

Abstract

The introduction of machine learning (ML) and artificial intelligence (AI) creates unprecedented opportunities for achieving full autonomy. However, learning-based methods in building autonomous systems can be extremely brittle in practice and are not designed to be verifiable. In this talk, I will present several of our recent efforts that combine ML with formal methods and control theory to enable the design of provably dependable and safe autonomous systems. I will introduce our techniques to generate safety certificates and certified decision and control for complex autonomous systems, even when the systems have a large number of agents, follow nonlinear and nonholonomic dynamics, and need to satisfy high-level specifications.

Biography

Chuchu Fan an Assistant Professor in the Department of Aeronautics and Astronautics and LIDS at MIT. Before that, she was a postdoc researcher at Caltech and got her Ph.D. from the Electrical and Computer Engineering Department at the University of Illinois at Urbana-Champaign in 2019. She earned her bachelor’s degree from Tsinghua University, Department of Automation. Her group at MIT works on using rigorous mathematics including formal methods, machine learning, and control theory for the design, analysis, and verification of safe autonomous systems. Chuchu’s dissertation work “Formal methods for safe autonomy” won the ACM Doctoral Dissertation Award in 2020.

Geometric Characterization of H-property for Step-graphons

October 21, 2022 11:00am – 12:00 pm

Location: Technology Square Research Building 509

Xudong Chen

Assistant Professor

Department of Electrical, Computer, and Energy Engineering

University of Colorado Boulder

Abstract

Graphon has recently been introduced by Lovasz, Sos, etc. to study very large graphs. A graphon can be understood as either the limit object of a convergent sequence of graphs, or, a statistical model from which to sample large random graphs. We take here the latter point of view and address the following problem: What is the probability that a random graph sampled from a graphon has a Hamiltonian decomposition? We have recently observed the following phenomenon: In the asymptotic regime where the size of the random graph goes to infinity, the probability tends to be either 0 or 1, depending on the underlying graphon. In this talk, we establish this “zero-one” property for the class of step-graphons and provide a geometric characterization.

Biography

Xudong Chen is an Assistant Professor in the Department of Electrical, Computer, and Energy Engineering at the University of Colorado Boulder. Prior to that, he was a postdoctoral fellow in the Coordinated Science Laboratory at the University of Illinois, Urbana-Champaign. He obtained the B.S. degree in Electronics Engineering from Tsinghua University, China, in 2009, and the Ph.D. degree in Electrical Engineering from Harvard University, Massachusetts, in 2014. He is an awardee of the 2020 Air Force Young Investigator Program, a recipient of the 2021 NSF CAREER award, and the recipient of the 2021 Donald P. Eckman award. His current research interests are in the area of control theory, stochastic processes, optimization, graph theory and their applications in modeling, analysis, control, and estimation of large-scale complex systems.

Online Optimization and Control using Black-Box Predictions

April 19, 2022 11:00 am – 12:00 pm

Location: Instructional Center 105

Also live streamed at https://gatech.zoom.us/j/96813456832

Zoom Meeting ID: 968 1345 6832

Adam Wierman

Professor

Caltech

Abstract

Making use of modern black-box AI tools is potentially transformational for online optimization and control. However, such machine-learned algorithms typically do not have formal guarantees on their worst-case performance, stability, or safety. So, while their performance may improve upon traditional approaches in “typical” cases, they may perform arbitrarily worse in scenarios where the training examples are not representative due to, e.g., distribution shift or unrepresentative training data. This represents a significant drawback when considering the use of AI tools for energy systems and autonomous cities, which are safety-critical. A challenging open question is thus: Is it possible to provide guarantees that allow black-box AI tools to be used in safety-critical applications? In this talk, I will introduce recent work that aims to develop algorithms that make use of black-box AI tools to provide good performance in the typical case while integrating the “untrusted advice” from these algorithms into traditional algorithms to ensure formal worst-case guarantees. Specifically, we will discuss the use of black-box untrusted advice in the context of online convex body chasing, online non-convex optimization, and linear quadratic control, identifying both novel algorithms and fundamental limits in each case.

Biography

Adam Wierman is a Professor in the Department of Computing and Mathematical Sciences at Caltech. He received his Ph.D., M.Sc., and B.Sc. in Computer Science from Carnegie Mellon University and has been a faculty at Caltech since 2007. Adam’s research strives to make the networked systems that govern our world sustainable and resilient. He is best known for his work spearheading the design of algorithms for sustainable data centers and his co-authored book on “The Fundamentals of Heavy-tails”. He is a recipient of multiple awards, including the ACM Sigmetrics Rising Star award, the ACM Sigmetrics Test of Time award, the IEEE Communications Society William R. Bennett Prize, multiple teaching awards, and is a co-author of papers that have received “best paper” awards at a wide variety of conferences across computer science, power engineering, and operations research.