The Control Theory Seminar is on Mondays at 4pm in Cory 540AB
|Mon., Sep-9||Jeff Shamma, Georgia Tech|
|Stability and Selection in Multiagent Learning|
Recent years have witnessed significant interest in the area of multiagent or distributed architecture control, with applications ranging from autonomous vehicle teams to communication networks to smart grid. The general setup is a collection of multiple decision-making elements interacting locally, perhaps striving to achieve a common collective objective. In multiagent learning, agents dynamically adapt to the actions of other agents, thereby effectively making the environment non-stationary from the perspective of any single agent. The resulting dynamics can exhibit behaviors ranging from chaos to convergence. This talk focuses on the two concerns of stability and selection—i.e., do agents converge, and if so, to what configurations? We discuss "stability" of population games through new connections between passivity theory and evolutionary game theory. We discuss "selection" in evolutionary games using the notion of stochastic stability and demonstrate its broader applicability in the setting of programmable self assembly.
Jeff Shamma is the Julian T. Hightower Chair in Systems & Control in the School of Electrical and Computer Engineering at the Georgia Institute of Technology. Jeff received a BS in Mechanical Engineering from Georgia Tech in 1983 and a PhD in Systems Science and Engineering from the Massachusetts Institute of Technology in 1988. Prior to returning to Georgia Tech in 2007, he held faculty positions at the University of Minnesota, University of Texas-Austin, and University of California–Los Angeles. Jeff is a recipient of the NSF Young Investigator Award (1992) and the American Automatic Control Council Donald P. Eckman Award (1996), and a Fellow of the IEEE (2006). He previously served on the Air Force Scientific Advisory Board (2008–2011) and is currently an Associate Editor for the IEEE Transactions on Cybernetics (2009–present) and Games (2012–present) and a Senior Editor for the newly formed IEEE Transactions on Control of Network Systems.
|Mon., Sep-16||Pushkin Kachroo, University of Nevada, Las Vegas|
|Transportation Feedback Control Problems: Microscopic to Macroscopic perspective leading to Resilient Networks|
All roads lead to Rome! Transportation has been and is the leading economic and social backbone of any successful civilization. Two competing aspects of mobility are throughput and safety. From a modelling and control point of view transportation has a microscopic scale where we study problems relating to vehicles to the macroscopic scale where we study traffic problems. The path from the microscopic to macroscopic perspective can go through a mesoscopic scale. The talk will present various control models, problems, solutions, and challenges in these various scales. It will discuss some weak solutions of dynamic systems in both scales. For instance the work in microscopic scale leads to ODEs with discontinuous right hand side, and the work in the macroscopic scale leads to PDEs with discontinuous solutions in finite time (shock waves). The talk will conclude with current and future work aimed at the theory and design of control theoretic resilient transportation networks in the context of general resilient networks.
Pushkin Kachroo is currently a visiting Professor at the University of California at Berkeley working with Professor Shankar Sastry. He is also the Lincy Professor of Transportation in the department of Electrical and Computer Engineering at the University of Nevada Las Vegas (UNLV). He is the director of the Transportation Research Center at UNLV and also the Associate director of the Mendenhall Innovation Program at the Thomas a Hughes College of Engineering at UNLV. He was an Associate Professor at Virginia Tech before he joined UNLV in 2007. He obtained his Ph.D. in Mechanical Engineering from University of California at Berkeley performing research in Vehicle Control in 1993 under Professor Masayoshi Tomizuka, and obtained another Ph.D. in Mathematics from Virginia Tech in Mathematics in the area of hyperbolic system of partial differential equations with applications to Traffic Control and Evacuation under Professor Joseph A. Ball. He has authored 10 books on traffic and vehicle control. He has authored more than 120 publications that include books, research papers, and edited volumes. He has taught about 30 different courses in Virginia Tech in the areas of electrical and computer engineering, and mathematics. Similarly, he also taught 30 different courses at UNLV since 2007. He has graduated more than 35 graduate students, and has been P.I. or Co P.I. on projects worth more than 4 Million Dollars. He was awarded the most outstanding new professor at Virginia Tech, and also has received many teaching awards and certificates, both at Virginia Tech and UNLV. He also received the faculty excellence award from CSUN UNLV in 2011.
|Mon., Sep-23||Katie Byl, University of California, Santa Barbara|
|Robust Legged Locomotion on Uneven Terrain|
An obvious motivation for using legs instead of wheels is to negotiate rough and intermittent terrain. Our group focuses on underactuated legged robots on such terrain, particularly in the limit where having a small foothold limits (or eliminates) available ankle torque, and we are specifically interested in designing control for and quantifying the performance of such systems when noisy information about stochastic environments is available. In this talk, I will provide an overview of several projects in our group on humanoid control, including planar models for standing balance, for point-footed walking, and for actuated SLIP hopping, and I will also discuss (very briefly) some trade-offs in our use of a non-humanoid robot for the upcoming DARPA Robotics Challenge.
Katie Byl received her B.S., M.S., and Ph.D. degrees in mechanical engineering from MIT. Her research is in dynamic systems and control, with particular interest in modeling and control techniques to deal with the inherent challenges of underactuation and stochasticity that characterize bio-inspired robot locomotion and manipulation in real-world environments. Her research is currently funded in part by DARPA's M3 program, the DARPA Robotics Challenge (with JPL), the Army's Institute for Collaborative Biotechnologies (ICB) and Robotics CTA programs, an NSF CAREER award (2013), the Hellman Foundation (2012), and an Alfred P. Sloan Research Fellowship in Neuroscience (2011). Katie has worked on a wide range of research topics in the control of dynamic systems, including magnetic bearing control, flapping-wing microrobotics, piezoelectic noise cancellation for aircraft, and vibration isolation for gravity wave detection, and she was once a professional gambler on the now-infamous MIT Blackjack Team.
|Mon., Sep-30||Kostas Daniilidis, University of Pennsylvania|
|3D Object Detection and Pose Estimation: from Single Images to Active Viewpoint Selection|
In this talk, we address the problem of detection and localization of 3D objects in cluttered scenes. Object exemplars are given in terms of 3D models without any appearance cues. Deformable part-models are trained on clusters of silhouettes of similar poses and produce hypotheses about possible object locations. Objects are simultaneously segmented and verified inside each hypothesis bounding region using the chordiogram descriptor. A final iteration on the 6-DOF object pose minimizes the distance between the selected image contours and the actual projection of the 3D model. While we demonstrated successful grasps based on single images we believe that selection of class and pose could be optimized if we explore the capability of active viewpoint selection. When an initial static detection phase identifies an object of interest, several hypotheses are made about its class and orientation. We plan a sequence of viewpoints, which balances the amount of energy used to move with the chance of identifying the correct hypothesis. We formulate an active hypothesis testing problem, which includes camera mobility, and solve it using a point-based approximate POMDP algorithm. Experiments using a 3D model database and an RGB-D sensor show a significant improvement both in detection and pose estimation. This is joint work with Menglong Zhu, Matthieu Lecce, Cody Phillips, Kosta Derpanis, Nikolay Atanasov, Jerome Le Ny, and George Pappas.
Kostas Daniilidis is Professor of Computer and Information Science at the University of Pennsylvania where he has been faculty since 1998. He obtained his undergraduate degree in Electrical Engineering from the National Technical University of Athens, 1986, and his PhD in Computer Science from the University of Karlsruhe, 1992, under the supervision of Hans-Hellmut Nagel. His research interests are on visual motion and navigation, active perception, 3D object detection and localization, and panoramic vision. He was Associate Editor of IEEE Transactions on Pattern Analysis and Machine Intelligence from 2003 to 2007. He founded the series of IEEE Workshops on Omnidirectional Vision. In June 2006, he co-chaired with Pollefeys the Third Symposium on 3D Data Processing, Visualization, and Transmission, and he was Program co-Chair of the 11th European Conference on Computer Vision in 2010. He has been the director of the interdisciplinary GRASP laboratory from 2008 to 2013 and he is the Associate Dean for Graduate Education of Penn Engineering since 2013. He is an IEEE Fellow.
|Fri., Oct-4||Christos Cassandras, Boston University|
|Event-Driven Control, Communication, and Optimization|
The event-driven paradigm is an alternative viewpoint, complementary to the time-driven approach, for modeling, sampling, estimation, control, and optimization of dynamic systems. For example, time-driven sampling and communication with energy-constrained wireless devices can be inefficient, unnecessary, and sometimes infeasible. The key idea in event-driven control is that a “clock” should not be dictating actions simply because a time step is taken; rather, an action should be triggered by an “event” which may be a well-defined condition on the system state (including a simple time step) or a random state transition.
We will present results in two areas where event-driven control and optimization have made significant progress. First, in distributed multi-agent systems, we will show how event-driven, rather than synchronous, communication can guarantee convergence in cooperative distributed optimization while provably maintaining optimality. Second, in stochastic hybrid systems, we will show how on-line gradient estimation techniques boil down to a set of event-driven Infinitesimal Perturbation Analysis equations (an “IPA calculus”) where the estimates are (under certain conditions) robust with respect to modeling details and noise, and scalable in the number of observed events. We will illustrate how this IPA calculus is used to solve a large class of stochastic optimization problems
Dr. Cassandras was Editor-in-Chief of the IEEE Transactions on Automatic Control from 1998 through 2009 and has also served as Editor for Technical Notes and Correspondence and Associate Editor. He was the 2012 President of the IEEE Control Systems Society (CSS) and has served as Vice President for Publications and on the Board of Governors of the CSS. He has chaired the CSS Technical Committee on Control Theory, and served as Chair of several conferences. He has been a plenary speaker at many international conferences, including the American Control Conference in 2001 and the IEEE Conference on Decision and Control in 2002, and an IEEE Distinguished Lecturer.
He is the recipient of several awards, including the 2011 IEEE Control Systems Technology Award, the Distinguished Member Award of the IEEE Control Systems Society (2006), the 1999 Harold Chestnut Prize (IFAC Best Control Engineering Textbook) for Discrete Event Systems: Modeling and Performance Analysis, a 2011 prize for the IBM/IEEE Smarter Planet Challenge competition, a 1991 Lilly Fellowship and a 2012 Kern Fellowship. He is a member of Phi Beta Kappa and Tau Beta Pi. He is also a Fellow of the IEEE and a Fellow of the IFAC.
|*TIME AND DATE CHANGE: Seminar on Friday at 3pm in Cory 540AB|
|Mon., Oct-7||Shai Revzen, University of Michigan|
|How Most Things Run—An Oscillator Theory Perspective on Rapid Locomotion|
Legged locomotion on land is often viewed in terms of “gaits” corresponding to repeated patterns of footfalls, suggesting an application of the mathematical theory of limit cycle oscillators. Unlike classical oscillators, the oscillators appearing in locomotion encounter rapid processes that are often modeled as non-smooth “hybrid” transitions.
I present an experimental scientist's perspective on rapid multilegged locomotion that lead us through the development of key tools for the empirical analysis of animal and robot gaits, and on to a discovery of new, fundamental forms of stability unique to hybrid limit cycle oscillators of multiple contact gaits.
|Mon., Oct-14||Jorge Cortés, University of California, San Diego|
|Distributed algorithmic solutions to zero-sum two-network games and network bargaining|
Recent years have seen an increasing interest in the development of novel tools and algorithms for the solution of strategic problems that involve either networked entities as players, forces that operate over networks, or both. In such scenarios, information is distributed across multiple layers and only partially available to individual agents. The nature of these agents, as well as their interactions, might be diverse, ranging from cooperative to adversarial, or anything in-between. This talk is a contribution to this growing body of research. We consider two scenarios: zero-sum two-network games, where two teams of agents have opposing objectives with regards to the optimization of a common objective function, and network bargaining problems, where agents bargain over the possibility of matching and allocating a common good among them. For each scenario, we present progress on the synthesis of provably-correct distributed coordination strategies that find Nash stable outcomes over general interaction topologies. Our technical approach exploits the close connection of these scenarios with distributed optimization problems and linear programming, and combines notions and methods from algebraic graph theory, nonsmooth and convex analysis, consensus algorithms, set-valued dynamical systems, and game theory.
Jorge Cortés is an Associate Professor with the Department of Mechanical and Aerospace Engineering at the University of California, San Diego. He received the Licenciatura degree in mathematics from the Universidad de Zaragoza, Spain, in 1997, and the Ph.D. degree in engineering mathematics from the Universidad Carlos III de Madrid, Spain, in 2001. He held postdoctoral positions at the University of Twente, The Netherlands, and at the University of Illinois at Urbana-Champaign, USA. He was an Assistant Professor with the Department of Applied Mathematics and Statistics at the University of California, Santa Cruz from 2004 to 2007. He is the author of "Geometric, Control and Numerical Aspects of Nonholonomic Systems" (New York: Springer-Verlag, 2002) and co-author of "Distributed Control of Robotic Networks" (Princeton: Princeton University Press, 2009). He received a NSF CAREER award in 2006 and was the recipient of the 2006 Spanish Society of Applied Mathematics Young Researcher Prize. He has co-authored papers that have won the 2008 IEEE Control Systems Outstanding Paper Award, the 2009 SIAM Review SIGEST selection from the SIAM Journal on Control and Optimization, and the 2012 O. Hugo Schuck Best Paper Award in the Theory category. He is a IEEE Control Systems Society Distinguished Lecturer.
|Mon., Oct-28||Richard Murray, Caltech|
|Correct-by-Construction Design of Control Protocols for Hybrid System|
We are investigating the specification, design and verification of distributed systems that combine communications, computation and control in dynamic, uncertain and adversarial environments. Our goal is to develop methods and tools for designing control policies, specifying the properties of the resulting distributed embedded system and the physical environment, and proving that the specifications are met. We have recently developed a promising set of results in receding horizon temporal logic planning that allow automatic synthesis of protocols for hybrid (discrete and continuous state) dynamical systems that are guaranteed to satisfy the desired properties even in the presence of environmental action. The desired properties are expressed in the language of temporal logic, and the resulting system consists of a discrete planner that plans, in the abstracted discrete domain, a set of transitions of the system to ensure the correct behaviors, and a continuous controller that continuously implements the plan. Extensions to this work allow the incorporation of optimization of an appropriate cost function, probabilistic execution, and on-the-fly (re-)synthesis. Application areas include autonomous driving, vehicle management systems, and distributed multi-agent systems.
|Mon., Nov-4||Miroslav Krstic, University of California, San Diego|
|Hyperbolic PDEs and Nonlinear Delay Systems: Control Designs and Applications|
Delay systems belong to a larger family of dynamical systems that incorporate ODEs and first-order hyperbolic PDEs. Such differential equation ingredients (ordinary and partial) give rise to numerous possible nonlinear system structures. Various novel applications in 3D printing, oil drilling, oil production, and other technologies incorporate hyperbolic PDEs and delays, as I will illustrate in the seminar, along with presenting nonlinear infinite-dimensional feedback designs for such systems.
Miroslav Krstic holds the Daniel L. Alspach endowed chair and is the founding director of the Cymer Center for Control Systems and Dynamics at UC San Diego. He also serves as Associate Vice Chancellor for Research at UCSD. Krstic has held the Springer Professorship at UC Berkeley, the Royal Academy of Engineering Distinguished Fellowship, and visiting professorships at Universities Denis Diderot and Pierre et Marie Curie in Paris. He is a Fellow of IEEE and IFAC and a recipient of the PECASE, NSF Career Award, ONR Young Investigator Award, the Axelby and Schuck Paper Prizes, and was the first recipient of the UCSD Research Award from engineering. Krstic serves as Senior Editor in IEEE Transactions on Automatic Control and Automatica and as editor in several book series with Springer-Verlag and Birkhauser. He has served as chair of the IEEE CSS Fellow Committee. Krstic has delivered nearly 20 keynote lectures and has coauthored 10 books on adaptive, nonlinear, and stochastic control, extremum seeking, and control of PDE and delay systems.
|Mon., Nov-11||No Seminar|
|Mon., Nov-18||Victor Preciado, University of Pennsylvania|
|Spectral Analysis, Modeling, and Control of Complex Dynamic Networks|
Research on complex networks has important applications in a wide variety of social and technological systems, such as the Internet, online social networks, the electric grid, and other cyber-physical systems. Since critical societal functions are increasingly dependent on these complex networks, it is important to ensure that their performance and reliability are not degraded as the structure of the network evolves over time. This requires us to acquire a deeper understanding of the relationship between the structure and dynamics of these complex systems. Understanding the dynamics of complex networks is a challenging problem due to: (1) the cumbersome size and complexity of the systems (e.g. the Internet), (2) the nonexistence of a vantage point with complete information about the structure of the system (e.g. social networks), and (3) the structure of the system itself may be changing over time. In this talk, I present a novel theoretical foundation to analyze massive networks of dynamical elements using tools and techniques at the intersection of dynamical systems and control, probability, optimization, and graph theory. Alongside, I use real experimental data from technological networks (e.g. electric transmission networks and the Internet), as well as online social networks (e.g. Facebook and Twitter graphs) to validate theoretical predictions and build computational tools of practical interest.
Victor M. Preciado is the Raj and Neera Singh Assistant Professor of Electrical and Systems Engineering at the University of Pennsylvania. He received the PhD degree in Electrical Engineering and Computer Science from the Massachusetts Institute of Technology in 2008. His main research interests lie in the modeling, analysis, control and optimization of dynamical processes and strategic interactions in large-scale complex networks, with applications in social networks, electric power distribution, multi-agent systems, and biological networks.
|Mon., Nov-25||Maryam Shanechi, Cornell University|
|Real-Time Brain-Machine Interface Architectures: Algorithmic Development and Experimental Implementation|
A brain-machine-interface (BMI) is a system that interacts with the brain to control its state or to allow it to control an external device. Examples include deep brain stimulators to treat depression and motor BMIs for restoring movement in paralyzed patients. Motor BMI research has largely focused on the problem of restoring the original motor function by using standard signal processing techniques. However, developing novel algorithmic solutions that are tailored to the neural system can significantly improve the performance of these BMIs and enable their clinical viability. Moreover, while developing high-performance BMIs with the goal of matching the original motor function is indeed valuable, a compelling goal is that of designing BMIs that can enhance original motor function. In this talk, I first develop a novel real-time BMI for restoration of natural motor function and then introduce a BMI architecture aimed at enhancing original motor function. I demonstrate the successful implementation of both these designs in rhesus monkeys. I will also briefly present an extension of our algorithmic developments to design BMIs that monitor and modify the state of the brain under anesthesia.
To facilitate the restoration of lost motor function, I develop a two-stage decoder to decode jointly the target and trajectory of a reaching movement and show that it outperforms target or trajectory decoding alone. The decoder incorporates an optimal feedback-control model of BMI and directly processes the spiking activity using point process modeling. Our ongoing work combines our optimal feedback control model with adaptive point process filtering to advance the state-of-the-art BMIs. To enable enhancement of the original motor function, I introduce a concurrent BMI architecture for performing complex tasks that involve a sequence of planned movements. In contrast to a traditional BMI, this BMI decodes all the elements of the sequential motor plan concurrently prior to movement, which enables it to find ways to perform the task more effectively than is possible by natural movement. I demonstrate that sequential motor plans can indeed be decoded simultaneously, accurately, robustly, and in advance of movement.
|Mon., Dec-2||Zico Kolter, Carnegie Mellon University|
|Sparsity in Learning and Control: Algorithms and Applications in Energy Systems|
Sparsity has received a great deal of attention in recent years in both the machine learning and control communities, though the specific focus in the two domains has typically been different; in machine learning, sparse approaches have often focused on convex methods and large-scale algorithms, whereas in control there has been a focus on decentralized approaches and restricted classes of systems that admit optimal sparse control laws. In this talk, I will present two recent applications of sparse methods to machine learning and control domains, exploiting a common optimization framework, and then apply these methods to tasks in energy forecasting and electrical power system control. In particular, on the machine learning side, I will present the sparse Gaussian conditional random field, a discriminative analogue of sparse inverse covariance estimation, and use this model for high-dimensional probabilistic forecasting of wind power. On the control side, I will present a sparse formulation of the linear quadratic regulator, and use this method to develop decentralized controllers for frequency regulation and voltage control in power grids. In both cases we develop a common algorithmic approach, based upon inexact proximal Newton methods, that can vastly outperform previous algorithms, thus enabling these methods to scale to previously-inaccessible domains.
J. Zico Kolter is an assistant professor in the Computer Science Department and the Computation, Organizations, and Society program in the School of Computer Science at Carnegie Mellon University. Previously, he was a postdoctoral fellow in the Computer Science and Artificial Intelligence Laboratory at MIT, supported by a Computing Innovation Fellowship. He received his his Ph.D. in Computer Science from Stanford University in 2010 and his B.S. from Georgetown University in 2005. His research focuses on sustainable energy domains, with a focus on core learning, inference, and control tasks within this space.
|Mon., Dec-9||No Seminar|