GoGE Workshop on Optimization, Decision and AI

Abstract

For decades, the robotics community has been working on developing intelligent autonomous machines that can perform complex tasks in unstructured environments. We are now closer than ever to delivering on this promise. Robotic systems are being developed, tested and deployed for a wide range of applications. In this talk, I will present our work on building robots for agriculture and home automation which are two application domains with distinct sets of associated challenges. In agriculture, robots must be capable of operating on very large farms under rough conditions while maintaining precision to efficiently perform tasks such as yield mapping, fruit picking and weeding. In these applications, the state-of-the-art perception algorithms are capable of generating intermediate geometric representations of the environment. However, the resulting planning problems are often hard. I will present some of our work on tracking and mapping and give examples of field deployments. In home automation, the robots must be able to handle a large variety of objects and clutter. In such settings, generating precise geometric models as intermediate representations is not always possible. To address this challenge, I will present our recent and ongoing work on developing state representations for coupled perception and action planning for representative home automation applications such as decluttering.

Biography

Ibrahim Volkan Isler is Professor of Computer Science & Engineering at the University of Minnesota and the head of Samsung AI Center in NY.

Abstract

Designing a safety-ensured controller for autonomous driving has been a well known issue for robotics community. However, the mainstream of this method relies on rule-based methods, as machine learning methods has shown inadequate results in autonomous driving control. To compensate with this issue, we propose to enhance the performance of the machine learning method by using various qualities of driving data. We use a high quality learning data, which is called the expert data, and a low quality learning data, which is called the abnormal data. By exploiting both of these data, we produce a controller with higher performance than the controller exploiting only high quality data.

Biography

Gunmin Lee received the B.S. degree in electrical and computer engineering from Seoul National University, Seoul, Korea, in 2019. He is currently working toward the Ph.D. degree with the Department of Electrical and Computer Engineering, Seoul National University, Seoul, South Korea. His research interests include applications of contrastive learning and imitation learning on autonomous driving.

Abstract

By using cryptographic technique named verifiable computation, verification scheme for distributed system enables each controllers to verify output of their adjacent controllers. Previously in linear dynamic system, actuator of plant-side can verify output of controller by using input from sensor. However, in distributed system, controllers cannot receive information about input of adjacent controllers, which is essential for computing verification. To this end, we propose verification of control system with 3-linear graph. By generating proof of input to protect information about input, the verification scheme achieve security without certain condition. Moreover, by expanding this 3-linear graph, the verification scheme can be adapted to any distributed system with undirected, connected graph.

Biography

Seungbeom Lee received the B.S. degree in Mathematics from Seoul National University, Seoul, South Korea, in 2016. He is currently working toward the Ph.D. degree in control theory at the Seoul National University, Seoul, South Korea. His research interests are related to security of control system, including encryption and verification.

Abstract

The ability to automatically tune the control parameters is very important in real engineering applications. In industrial servo and robotic systems, the demand for high throughput with high precision has led to high control gains, which in turn have caused exciting many intrinsic resonant modes of the system. In this presentation, these resonant vibrations are controlled with automatically tuned filters. A parameter tuning strategy of resonance suppression filters using frequency response function is developed. Then, an optimization problem of resonance suppression filters is formulated, and a particle swarm optimization algorithm is utilized to tune control parameters. Experimental results on an industrial belt-drive system show that the developed method achieves stable tuning performance even in the presence of position-dependent and multiple resonances in the system.

Biography

Tae-Ho Oh received the B.S. degree in electrical and computer engineering from Seoul National University, Seoul, Korea, in 2017. He is currently working toward the Ph.D. degree with the Department of Electrical and Computer Engineering, Seoul National University, Seoul, South Korea. His research interests include applications of nonlinear control theory and machine learning algorithms to servo drive and sensors.

Abstract

Data-driven controls using Gaussian process regression have recently gained much attention. In such approaches, system identification by Gaussian process regression is mostly followed by model-based controller designs. However, the outcomes of Gaussian process regression are often too complicated to apply conventional control designs, which makes the numerical design such as model predictive control employed in many cases. To overcome the restriction, our idea is to perform Gaussian process regression to the inverse of the plant with the same input/output data for the conventional regression. With the inverse, one can design a model reference controller without resorting to numerical control methods. In this presentation, we consider single-input single-output (SISO) discrete-time nonlinear systems of minimum phase with relative degree one. It is highlighted that the model reference Gaussian process regression controller is designed directly from pre-collected input/output data without system identification.

Biography

Hyuntae Kim received the B.S. degree in Electrical and Computer Engineering from Seoul National University, Seoul, South Korea, in 2015. He is currently pursuing the Ph.D. degree in control theory at the Seoul National University, Seoul, South Korea. His research interests include data-driven control and disturbance observer.

Abstract

For a class of hybrid systems, where the jumps occur very frequently, we analyze the stability of system trajectories using the viewpoint of singularly perturbed dynamics. Our framework comprises an interconnection of two hybrid subsystems, a timer which triggers the jumps, and some discrete variables to determine the index of the jump maps. The flow equations of these variables are singularly perturbed differential equations, and in particular, smaller value of the singular perturbation parameter leads to increase in the frequency of the jump instants. For the limiting value of this parameter, we consider a decomposition which comprises a quasi steady-state system modeled by a differential equation without any jumps, and a boundary-layer system described by purely discrete dynamics. By imposing appropriate assumptions on the quasi steady-state system and the boundary-layer system, we derive results showing practical, as well as asymptotic, stability of a compact attractor when the jumps occur frequently often. As an application of our results, we discuss two network design problems. First one is the control design problem in a network of second-order continuous-time coupled oscillators, where each agent communicates the information about its position to the neighbors at discrete times. In particular, we show that if the information exchange between the agents and their neighbors is frequent enough, then the oscillators achieve practical consensus. Secondly, we consider the design of observers for nonlinear systems with time-sampled measurements, and provide a design criterion for asymptotic stability of estimation error under an appropriate detectability assumption.

Biography

Aneel Tanwani graduated from University of Illinois at Urbana-Champaign, USA in 2011 where he obtained the M.S. degree in applied mathematics, and the M.S. and Ph.D. degrees in electrical engineering. He held postdoctoral positions at INRIA and Gipsa-lab in Grenoble, France, and then in the Department of Mathematics, Technical University Kaiserslautern, Germany. Since 2015, he has been a CNRS Associate Researcher with the LAAS, Toulouse, France. His research interests include switched and hybrid dynamical systems, stochastic processes, nonlinear control, variational analysis, and control with limited information. Dr. Tanwani was a recipient of the Fulbright scholarship in 2006. He is serving as an associate editor for IFAC Journal of Automatica since 2018, and for IEEE Transactions on Automatic Control since 2020. He has been an International Program Committee member for ACM Conference on Hybrid Systems: Computation and Control (2021, 2022), an associate editor for Conference Editorial Board of IEEE Control Systems Society (2016–2020) and a technical associate editor for several other conferences.

Abstract

Model predictive control (MPC) calls for repeatedly solving an optimisation problem on-line and applying the “opening moves” of the optimal decision to the system in a receding horizon fashion. Though computationally demanding at first sight, with advances in embedded computation and optimisation MPC, has emerged as a powerful methodology for a range of applications, fast and slow. In many of these applications, however, obtaining a model of the system dynamics, the “M” in MPC, to include in the constraints of the optimisation problem can be challenging. The standard approach is to use data collected from the system in a two step process of system identification to get an “M”, followed by conventional “PC”. Here we explore an alternative one step approach, where the data is used directly in the constraints of the optimisation problem. We show that for deterministic linear systems this is equivalent to conventional MPC. The method is then extended to uncertain or nonlinear systems through regularisation; we discuss how this can be interpreted as robustifying the optimisation problem against uncertainty in the data. Finally, we demonstrate the applicability of the method through benchmark examples and problems in power systems.

Biography

John Lygeros received a B.Eng. degree in 1990 and an M.Sc. degree in 1991 from Imperial College, London, U.K. and a Ph.D. degree in 1996 at the University of California, Berkeley. After research appointments at M.I.T., U.C. Berkeley and SRI International, he joined the University of Cambridge in 2000 as a University Lecturer. Between March 2003 and July 2006 he was an Assistant Professor at the Department of Electrical and Computer Engineering, University of Patras, Greece. In July 2006 he joined the Automatic Control Laboratory at ETH Zurich where he is currently serving as the Professor for Computation and Control and the Head of the laboratory. His research interests include modelling, analysis, and control of large scale systems, with applications to biochemical networks, energy systems, transportation, and industrial processes. John Lygeros is a Fellow of IEEE, and a member of IET and the Technical Chamber of Greece. Since 2013 he is serving as the Vice-​President Finances and a Council Member of the International Federation of Automatic Control and since 2020 as the Director of the National Center of Competence in Research “Dependable Ubiquitous Automation” (NCCR Automation).

Abstract

This talk presents a data-based construction of inverse dynamics for LTI systems. Specifically, the problem addressed here is to find an input sequence from the corresponding output sequence based on pre-collected input and output data. The problem is a reverse of the recent use of behavioral approach, in which the output sequence is obtained for a given input sequence by solving an equation formed by pre-collected data. The condition under which the problem gives a solution along with the result in the form of an algorithm will be presented.

Biography

Yongsoon Eun received the B.A. degree in mathematics, and the B.S. and M.S.E. degrees in control and instrumentation engineering from Seoul National University, Seoul, Korea, in 1992, 1994, and 1997, respectively, and the Ph.D. degree in electrical engineering and computer science from the University of Michigan, Ann Arbor, MI, USA, in 2003. From 2003 to 2012, he was a Research Scientist with the Xerox Innovation Group, Webster, NY, USA. Currently he is a Professor with the Department of Electrical Engineering and Computer Science, DGIST, Korea. His research interests include data-driven control, resilient control, production systems, and cyber-physical systems.