Liren Yang, 5-6 Feb 2020

From MurrayWiki
Revision as of 17:00, 3 February 2020 by Murray (Talk | contribs)

Jump to: navigation, search

Liren Yang is a PhD student at U. Michigan who has recently defended his thesis on correct-by-construction fault-tolerant control synthesis. He will visit Caltech on 5-6 Feb. Sign up here if you would lik to meet with him (use your Caltech credentials).


5 Feb (Wed)

  • 2:00p: NCS group meeting
  • 3:00p: CDS tea
  • 3:30p: Chuchu
  • 4:15p: Karena
  • 5:00p: Informal seminar - 121 Annenberg
  • 6:00p: Dinner with Richard + students (add your name here if you want to go): Apurva

6 Feb (Thu)

  • 9:15a: Richard Murray, 109 Steele Lab
  • 10:00a: Apurva, Pick up from Richard’s office. Meeting location: CDS Library
  • 10:45a: Yuxiao
  • 11:30a: Sumanth
  • 12:15p: done for the day


Correct-by-construction fault-tolerant control of complex dynamical systems

Correct-by-construction control synthesis methods can be used to algorithmically design controllers that render dynamical systems to satisfy complex tasks specified by formal languages. However, there is a gap between such techniques and real applications, especially when the systems experience faults, such as physical component failures and extreme operating conditions. Moreover, it is conservative and computationally expensive to apply such techniques developed for pure discrete systems directly to continuous-state dynamical systems. In this talk, I will present our work tackling these challenges. First, I will briefly mention our work on guaranteed fault-detection with linear temporal logic (LTL) constraints, and a hierarchical abstraction-based fault-tolerant controller synthesis approach. Then, regarding the first issue, I will focus on how to construct abstractions for systems with complex but structured dynamics. In particular, we leverage a special system structural property called mixed monotonicity to ease abstraction computation, and to develop synthesis techniques to incorporate this property. The presented methodology will be illustrated on a fuel cell thermal-power management problem.