Continuous-time LQR policy optimization satisfies a saturated PŁI condition that yields input-to-state stability of perturbed gradient flows, and overparametrization can restore global exponential convergence.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Some remarks on gradient dominance and LQR policy optimization
Continuous-time LQR policy optimization satisfies a saturated PŁI condition that yields input-to-state stability of perturbed gradient flows, and overparametrization can restore global exponential convergence.