Risk-Aware Safe Optimal Control of Uncertain Linear Systems

EasyChair Preprint no. 8994

Abstract

This paper synthesizes a risk-aware safe optimal controller for partially unknown linear systems under additive Gaussian noises. The risk is assessed through the concept of Conditional Value-at-Risk (CVaR) to account for the extreme low probability events that occurred in a one-step cost function without being overly conservative. The safety of the CVaR optimization solution is also guaranteed with high probability by imposing a chance constraint. A state-feedback risk-aware controller is first obtained that provides an upper bound to the formulated safe CVaR optimization problem. Then, an online data-driven quadratic programming (QP) optimization problem is devised to simultaneously and safely learn the unknown dynamics and control the system with high probability. As more measurements are collected, the safety constraint is tightened due to increasing the confidence in estimating the dynamic model. In the end, a numerical example is given to elucidate the efficacy of the proposed method.

Keyphrases: Risk-aware Control, Safe Control, Uncertain systems

@Booklet{EasyChair:8994,
  author = {Pouria Tooranjipour and Bahare Kiumarsi and Hamidreza Modares},
  title = {Risk-Aware Safe Optimal Control of Uncertain Linear Systems},
  howpublished = {EasyChair Preprint no. 8994},

  year = {EasyChair, 2022}}