研究会 (2025 年 12 月 20 日)

Abstract:
1. This talk addresses the problem of finite and fixed-time stabilization for
   linear time-invariant (LTI) systems in the presence of state quantization,
   using a homogeneous control framework. Quantized control has received
   significant attention due to practical limitations such as constrained
   communication bandwidth and memory resources. Quantization, which encodes
   continuous signals into discrete values, introduces non-negligible effects
   into feedback loops. Finite and fixed-time stability, which ensures
   convergence within a uniformly bounded time, is particularly desirable for
   time-critical systems. Homogeneous systems are popular for achieving this
   property because their convergence rates are determined by the degree of
   homogeneity. This talk explains how homogeneity can be systematically
   exploited in both controller design and quantizer design to guarantee
   finite and fixed-time stabilization of LTI systems.

   Bio:
   Yu Zhou is a Specially Appointed Assistant Professor at the Graduate School
   of Advanced Science and Engineering, Hiroshima University, Japan. He
   conducted his Ph.D research at INRIA Lille within the Valse team. He
   obtained his Ph.D. from Centrale Lille in January 2025. He received his
   B. Eng. degree in Flight Vehicle Power Engineering in 2017 and his M. Eng.
   degree in Aerospace Science and Technology in 2020 from the Beijing
   Institute of Technology, China. His research interests include control
   over networks and robotic control for high-speed and agile motion tasks.

2. This talk will firstly present some new theoretical results linked to the
   Stochastic Model Predictive framework, namely, the exact characterization of
   covariance dynamics for discrete-time linear systems affected by stochastic
   i.i.d. additive and parametric uncertainties. It will present how the use
   of some Kronecker product related transformation can help to derive a
   deterministic characterization of the error dynamics in the tube-based MPC
   framework, a key step for tractable chance-constraint tightening.
      In the case of parametric stochastic uncertainties, however, the derived
   exact characterization brings some challenges, particularly with respect to
   the control design, since such characterizations involve bilinear terms in
   the control gain. This problem will be addressed through the use of a
   Kronecker identity allowing to linearize the system, a descriptor
   representation is then used together with the S-variable approach in order
   to derive sufficient conditions with reduced conservatism.
      Then, this presentation will address some practical challenges related
   to biomedical systems (particularly anesthesia dynamics), such as handling
   varying equilibrium points and range tracking for some health indicators;
   and will illustrate how MPC for tracking based techniques can address such
   issues.
      Finally, the talk will outline future research directions, both theoretical
   and practical, related to MPC methods and their use in biomedical systems
   control.

   Bio:
   Kaouther Moussa is an associate professor in control at INSA Hauts de France
   (engineering school) and LAMIH (research laboratory) at Valenciennes, France.
   She completed her PhD in Grenoble Polytechnic Institute (Grenoble INP) and
   Gipsa-Lab, on optimization-based control and probabilistic certification of
   regions of attraction for cancer dynamics, followed by a postdoctoral position
   at the same laboratory within the TActical Multi-Objectives Swarming UAVs
   project. Her research focuses on predictive control and uncertainties handling
   through probabilistic and stochastic approaches, with an emphasis on biomedical
   applications. She is a recipient of a research grant for young researchers
   from the French National Research Agency (ANR) for the Clinical project
   (2024-2028), which lies at the intersection of learning and control approaches
   for the prediction and monitoring of anesthesia dynamics.

3. This talk introduces BADCONTROL, the first backdoor attack against low-level
   controllers that uses physical triggers. The attack poisons operational data
   to implant a vulnerability that can be activated by an exogenous signal from
   the environment, such as a specific driving maneuver or adversarial road patches
   within autonomous driving applications. BADCONTROL solves a constrained
   optimization problem by using a projected gradient ascent to modify the data,
   maximizing the frequency response of the controlled system at a target frequency.
   We evaluate BADCONTROL on Proportional-Integral-Derivative (PID) and
   Linear-Quadratic-Regulator (LQR) controllers through simulations and physical
   experiments. In the adaptive cruise control scenario, we achieve a 100% crash
   rate, while in lane keeping control, the backdoor causes the victim vehicle to
   steer 62% into the opposing lane, compared to 0% in both cases without a
   backdoor. By contrast, a state-of-the-art falsification framework for autonomous
   vehicles identifies only a single crash instance over 30 trials, underscoring
   its stealthiness.

   Bio:
   Hampei Sasahara is Lecturer with the Department of Information Physics and
   Computing, Graduate School of Information Science and Technology, the University
   of Tokyo, Tokyo, Japan. He received the Ph.D. degree in engineering from Tokyo
   Institute of Technology in 2019. From 2019 to 2021, he was a Postdoctoral Scholar
   with KTH Royal Institute of Technology, Stockholm, Sweden. From 2022 to 2024,
   he was an Assistant Professor with Tokyo Institute of Technology, Tokyo, Japan.
   From 2024 to 2025, he was an Assistant Professor with Institute of Science Tokyo,
   Tokyo, Japan. His main interests include secure control system design and control
   of large-scale systems.