The Control Theory Seminar is at 4pm in Cory 540AB
|Mon., Jan-27||Ram Vasudevan, MIT|
|Computational Tools for Automating Preventive and Rehabilitative Physical Therapy|
The doubling of the elderly population within the next twenty years presents a formidable challenge to the healthcare system due to their higher likelihood of chronic disease and accidental injury. Although the care of chronic disease has improved, the CDC expects that hospitalization and rehabilitative costs due predominantly to accidental injuries will more than double within the next twenty years. Successfully reducing expenditures will require the development of methods to automatically predict the likelihood of injury and automatically prescribe and verify the delivery of physical therapy. Designing techniques to measure and reduce the likelihood of falling will be especially critical since it is the leading cause of fatal and nonfatal injury amongst the elderly.
In this talk, I illustrate how to determine the likelihood of falling by measuring the stability basin of a locomotor pattern. To compute this stability basin, I describe a switched system optimal control algorithm to identify a personalized dynamical systems description of an individual's locomotor pattern, and the application of this algorithm to identify models of walking for nine individuals. By taking advantage of this dynamical model, I detail how to compute a stability basin from the solution of a linear partial differential equation. This solution can be approximated by a semidefinite program that can concurrently perform feedback control synthesis to maximize the size of the stability basin. This optimally constructed control law can be used to automatically devise strategies to retrain an individual who has a high likelihood of falling.
|Mon., Feb-10||Joao Hespanha, University of California, Santa Barbara|
|Randomized Methods in Game Theory with Applications to Network Security|
This talk addresses the solution of large zero-sum matrix games using randomized methods. We formalize a procedure -- termed the Sampled Security Policy (SSP) algorithm -- by which a player can compute policies that, with high probability, guarantees a certain level of performance against an adversary engaged in a random exploration of the game's decision tree.
The SSP Algorithm has applications to numerous combinatorial games in which decision makers are faced with a number of possible options that increases exponentially with the size of the problem. In this talk we focus on an applications in the area of network security, where system administrators need to consider multi-stage, multi-host attacks that may consist of long sequences of actions by an attacker in their attempt to circumvent the system defenses. In practice, this leads to policy spaces that grow exponentially with the number of stages involved in an attack.
His current research interests include hybrid and switched systems; multi-agent control systems; distributed control over communication networks (also known as networked control systems); the use of vision in feedback control; stochastic modeling in biology; and network security.Dr. Hespanha is the recipient of the Yale University's Henry Prentiss Becton Graduate Prize for exceptional achievement in research in Engineering and Applied Science, a National Science Foundation CAREER Award, the 2005 best paper award at the 2nd Int. Conf. on Intelligent Sensing and Information Processing, the 2005 Automatica Theory/Methodology best paper prize, the 2006 George S. Axelby Outstanding Paper Award, and the 2009 Ruberti Young Researcher Prize. Dr. Hespanha is a Fellow of the IEEE and an IEEE distinguished lecturer since 2007.
|Thu., Feb-20||Luca Schenato, University of Padova|
|Control over finite capacity channels: the role of data losses, delays and SNR limitations|
In this talk we will discuss the problem of feedback control design in the present of a finite capacity communication channel, which gives rise to tightly coupled limitations in terms of quantization errors, decoding/computational delays and erasure probability affecting the closed loop control performance. After providing an overview of the most notable results available in the literature, we will restrict the analysis in the context of LQG control subject to SNR limitations, packet loss, and delay and derive their impact on optimal design for the controller parameters. In particular, we will show that the stability of the closed loop system depends on a tradeoff among quantization, packet loss probability and delay. This analysis recaptures, as special cases, several results already available in the literature that have treated packet loss, quantization error and delay separately. We also show that the separation principle does not hold even if the controller has full knowledge of the packet loss sequence and the control inputs entering the plants. In fact the optimal control gain, when accounting for the communication constraints is, in general, different from the optimal gain derived under the classical LQG scenario, which is recaptured when the SNR over the channel goes to infinity.
Luca Schenato was born in Treviso in 1974 and he is currently Associate Professor at the Department of Information Engineering at the University of Padova. He received the Laurea Degree in Electrical Engineering from the University of Padova in 1999, the Management of Technology Certificate from the Haas Business School at UC Berkeley, USA, in 2003, and the Ph.D. from the Electrical Engineering and Computer Science Department of UC Berkeley in 2003. In 1997 he was a visiting student at the Department of Computing Science at the University of Aberdeen, UK, and in 2004 he help a post-doctoral fellowship at the EECS Department of UC Berkeley. From 2004 to 2007 he was the recipient of the Professorship "Returning Brains (Rientro dei Cervelli)" sponsored by the Italian Ministry of Education and he served as Adjunct Professor at the Department of Information Engineering at the University of Padova. He won the Eli Jury Award from the EECS Department of UC Berkeley in 2006 for his "for outstanding achievement in the area of Systems, Communications, Control, or Signal Processing". His expertise includes distributed control, estiamation and optimization; networked control systems; wireless sensor networks; consensus algorithms; and biomimetic locomotion. He is the author of more than 100 publications in international peer-reviewed conferences and journals and is current h-index=30 (Google Scholar). He is one of the founders of the Conference on Robotics, Communication and Coordination (Robocomm) for which he served as General vice-Chair (2007) and Technical Program Chair (2009), and of the Workshop on Estimation and Control of Networked Systems (Necsys) for which he served as as General vice-Chair (2009) and in the Advisory Board (2010). He is also serving as Associate Editor for the IEEE Transactions of Automatic Control and was Associate Editor for the following international conferences: IEEE CDC, IEEE ACC, IEEE ECC, IEEE IROS, IEEE CASE, IEEE ICRA.
|ROOM CHANGE: 3110 Etcheverry|
|Mon., Feb-24||Ketan Savla, University of Southern California|
|Robust Routing for Dynamical Network Flows|
Resilience is becoming a key consideration in the design and operation of infrastructure networks such as transportation. Network flow provides a compelling framework to model such systems. The celebrated max-flow min-cut theorem yields a procedure to quantify resilience of a network in terms of the underlying graph topology and link properties. However, such a procedure assumes a static framework and centralized computation and control.
In this talk, we present a dynamical framework for network flows, consisting of mass balance equations driven by dynamic routing policies at the nodes and congestion properties of the links. The evolution of the resulting dynamical model under capacity constraints exhibits forward and backward cascading effects, which are qualitatively different than standard percolation models. We present tight results on the stability and resilience properties of such dynamical network flows under a variety of distributed routing architectures. We also discuss applications of these results to the analysis of well-known dynamical transportation models and to the design of distributed traffic signal control.
|Mon., Mar-3||Noah Cowan, Johns Hopkins University|
|Encoding 3D Spatial Orientation in the Brain|
The visual encoding of 3D object orientation is critical for artificial and natural systems. Where and how the brain visually encodes 3D object orientation remains unknown, but prior studies suggest the caudal intraparietal area (CIP) may be involved. Here, we develop rigorous analytical methods for quantifying 3D orientation tuning curves, and use these tools to the study the neural coding of surface orientation. Specifically, we show that single neurons in area CIP of the rhesus macaque jointly encode the slant and tilt of a planar surface, and that across the population, the distribution of preferred slant-tilts is not statistically different from uniform. This suggests that all slant-tilt combinations are equally represented in area CIP. Furthermore, some CIP neurons are found to also represent the third rotational degree of freedom that determines the orientation of the image pattern on the planar surface. Together, the present results suggest that CIP is a critical neural locus for the encoding of all three rotational degrees of freedom specifying an object's 3D spatial orientation. This is joint work with Ari Rosenberg and Dora Angelaki at the Baylor College of Medicine.
Noah J. Cowan received a B.S. degree from the Ohio State University, Columbus, in 1995, and M.S. and Ph.D. degrees from the University of Michigan, Ann Arbor, in 1997 and 2001 — all in electrical engineering. Following his Ph.D., he was a Postdoctoral Fellow in Integrative Biology at the University of California, Berkeley for 2 years. In 2003, he joined the mechanical engineering department at Johns Hopkins University, Baltimore, MD, where he is now an Associate Professor. Prof. Cowan's research interests include mechanics and multisensory control in animals and machines. Prof. Cowan received the NSF PECASE award in 2010, the James S. McDonnell Foundation Scholar Award in Complex Systems in 2012, and the William H. Huggins Award for excellence in teaching in 2004.
|(hosted by CiBER-IGERT, ROOM CHANGE: 306 Soda, HP Auditorium)|
|Mon., Mar-10||Alyson Fletcher, University of California, Santa Cruz|
|Scalable Identification for Structured Nonlinear Neural Systems|
A key challenge in many large data analysis problems is to provide tractable algorithms that can capture the complex structure and dynamics inherent to high-dimensional systems. In this talk, I focus on a framework for computationally efficient identification and estimation of a rich class of systems composed of interconnected, linear (possibly dynamical) blocks and memoryless nonlinearities. The methodology leverages message passing techniques to provide a scalable general approach with provable guarantees in consistency and convergence in a wide variety of settings. This work improves upon current estimation of large neural dynamical networks from state-of-the-art multi-neuron imaging.
Alyson Fletcher received a master's degree in mathematics and master's and PhD degrees in electrical engineering from UC-Berkeley. Her awards include the University of California President's Postdoctoral Fellowship, a Luce Foundation fellowship, and the NSF CAREER Award. She is currently an Assistant Professor of Electrical Engineering at the UCSC Baskin School of Engineering. Her research interests include signal processing, estimation, learning, control theory, and computational neuroscience.
|ROOM CHANGE: HP Auditorium|
|Mon., Mar-17||Ram Rajagopal, Stanford University|
|What is the Power of Groups?|
Aggregation of supply and demand-side resources at various scales has been proposed as a mechanism to manage the variability of power networks with significant penetration of renewables. In the supply-side, most studies argue for the formation of large groups of producers that compensate each other's shortfalls. In the demand-side, loads are managed as large aggregates at the unit of neighborhoods or cities. Yet, not much is known based on actual supply, demand and market data. In this talk we investigate the data-driven design of right-sized groups. In the supply-side, we formulate a Cournot competition game based on system operator market data. The resulting model demonstrates that efficient coalitions have an optimal size. In the demand-side we investigate optimal pricing for consumers utilizing hourly smart meter data from 500,000 households. We propose a simple and scalable revenue management mechanism that shows effective pricing divides customers into groups of optimal size. We conclude the talk by highlighting additional important features from data and outline some ongoing and future work. This presentation is based on joint work with Ramesh Johari, Jungsuk Kwac, Sid Patel, Raffi Sevlian and Baosen Zhang.
Ram Rajagopal is an Assistant Professor of Civil and Environmental Engineering and has a courtesy appointment in Electrical Engineering at Stanford University. He directs the Stanford Sustainable Systems Lab (S3L), focused on large scale monitoring, data analytics and stochastic control for infrastructure networks, in particular energy and transportation. His current research interests in power systems are in data-driven approaches for the integration of renewables, smart distribution systems and demand-side data analytics. Prior to his current position he was a DSP Research Engineer at National Instruments and a Visiting Research Scientist at IBM Research. He holds a Ph.D. in Electrical Engineering and Computer Sciences and an M.A. in Statistics, both from the University of California Berkeley, Masters in Electrical and Computer Engineering from University of Texas, Austin and Bachelors in Electrical Engineering from the Federal University of Rio de Janeiro. He is a recipient of the Powell Foundation Fellowship, Berkeley Regents Fellowship and the Makhoul Conjecture Challenge award. He holds more than 30 patents from his work, and has advised or founded companies in the fields of sensor networks, power systems and data analytics.
|Mon., Mar-24||No Seminar|
|Mon., Mar-31||Stefan Schaal, University of Southern California|
|Learning Motor Skills: From Movement Primitives to Associative Skill Memories|
Controlling a complex movement system requires making perceptual and control decisions at every moment of time, and learning and adaptation to improve the system's performance. High dimensional continuous state-action spaces still pose significant scaling problems for learning algorithms to find (approximately) optimal solutions, and appropriate task descriptions or cost functions require a large amount of human guidance. In order to address autonomous skillful movement generation in complex robot and task scenarios, we have been working on a variety of subproblems to facilitate robust task achievement. Among these topics are general representations for movement in form of movement primitives, trajectory-based reinforcement learning with path integral reinforcement learning, and inverse reinforcement learning to extract the "intent" of observed behavior. However, this "action centric" view of skill acquisition needs to be extended with a stronger perceptual component, as, in the end, it is the entire perception-action-learning loop that could be considered the key element to address, rather than isolated components of this loop. In some tentative initial research, we have been exploring Associative Skill Memories, i.e., the simple idea to start memorizing all sensory events and their statistics together with each movement skill. This concepts opens a wide spectrum of adding predictive, corrective, and switching behaviors in motor skills, and may create an interesting foundation to automatically generate the graphs underlying complex sequential motor skills. Our research results will be illustrated in various experiments with complex anthropomorphic robot systems and also some results from behavioral experiments.
|(hosted by CiBER-IGERT, ROOM CHANGE: 306 Soda, HP Auditorium)|
|Mon., Apr-7||Katherine Kuchenbecker, University of Pennsylvania|
|The Value of Tactile Sensations in Haptics and Robotics|
Although physical interaction with the world is at the core of human experience, few computer and machine interfaces provide the operator with high-fidelity touch feedback, limiting their usability. Similarly, autonomous robots rarely take advantage of touch perception and thus struggle to match the manipulation capabilities of humans. My long-term research goal is to leverage scientific knowledge about the sense of touch to engineer haptic interfaces and robotic systems that increase the range and quality of tasks humans can accomplish. This talk will describe my group's three main research thrusts: haptic texture rendering, touch feedback for robotic surgery, and touch perception for autonomous robots. Our work in all three of these areas has been principally enabled by a single insight: although less studied than kinesthetic cues, tactile sensations convey much of the richness of physical interactions.
Katherine J. Kuchenbecker is an Associate Professor in Mechanical Engineering and Applied Mechanics at the University of Pennsylvania. Her research centers on the design and control of haptic interfaces for applications such as robot-assisted surgery, medical simulation, stroke rehabilitation, and personal computing. She directs the Penn Haptics Group, which is part of the General Robotics, Automation, Sensing, and Perception (GRASP) Laboratory. She has won several awards for her research, including an NSF CAREER Award in 2009, Popular Science Brilliant 10 in 2010, and the IEEE Robotics and Automation Society Academic Early Career Award in 2012. Prior to becoming a professor, she completed a postdoctoral fellowship at the Johns Hopkins University, and she earned her Ph.D. in Mechanical Engineering at Stanford University in 2006.
|Mon., Apr-14||No Seminar|
|Mon., Apr-21||Maryam Fazel, University of Washington|
|Geodesic Distance Maximization Via Convex Optimization|
Given a graph with fixed edge weights, finding the shortest path, also known as the geodesic, between two nodes is a well-studied network flow problem. We introduce the Geodesic Distance Maximization Problem (GDMP), i.e., the problem of finding the edge weights that maximize the length of the geodesic, subject to convex constraints on the weights.
We show that GDMP is a convex optimization problem for a wide class of flow costs, and provide a physical interpretation using the dual. We present applications of the GDMP in various fields, including network interdiction, optical lens design, and resource allocation in the control of forest fires. GDMP can be generalized from graphs to continuous fields, where the Eikonal equation (a fundamental partial differential equation governing flow propagation) naturally arises as a constraint in the dual problem. For the case of propagation on a regular grid, the problem can be cast as a second-order cone program; however standard solvers fail to scale to the large grid sizes of interest. We develop a new Alternating Direction Method of Multipliers (ADMM) by exploiting specific problem structure to solve large-scale GDMP, and demonstrate its effectiveness in numerical examples.
This talk is based on joint work with De Meng, Pablo Parrilo, and Stephen Boyd.