betaFeedback

WorksEngineering

First-Order Model-Free Policy Gradient for Undiscounted LQR with Global Optimality and Convergence Guarantees

Abstract

We provide a first-order model-free policy gradient (PG) framework for learning undiscounted infinite-horizon linear quadratic regulator (LQR) policies with global optimality and convergence guarantees. We first construct a stochastic wrapper policy class parameterized by the LQR gain to enable exploration around a deterministic linear control law. We analyze the additional cost induced by exploration over infinite-horizon, and establish a weak gradient dominance condition that characterizes global optimality under exploration. We then construct a model-free first-order PG algorithm using sampled trajectory information, without identifying the dynamics or approximating a value function. Finally, we establish finite-time convergence guarantees, providing the sufficient number of samples required for near-global optimality with our proposed PG algorithms.

Paper details

vlnc:W_56986zm5
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC-BY-NC-ND-4.0
Usage

views · 5

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Engineering

Not recorded for this paper

View statement
Looking for related papers…