Skip to content

Regret Bounds for the Adaptive Control of Linear Quadratic Systems

Authors: Yasin Abbasi-Yadkori, Csaba Szepesvari

Published: 2011 (Conference Paper)

Source: Conference on Learning Theory

Algorithm: OFU-LQ

Summary

Derives O(sqrt(T)) regret bounds for adaptive control of unknown linear-quadratic systems without forced exploration, using optimistic control with a confidence set around the estimated model parameters. Establishes the first regret bound for the LQ control problem and connects online learning to adaptive control theory.

Abstract

Primary

Paper proceedings.mlr.press

Alternate

proceedings.mlr.press PDF proceedings.mlr.press

Tags

  • Reinforcement learning

  • Linear quadratic regulator

  • Adaptive control

  • Regret bounds

  • Exploration-exploitation

  • System identification