<ul><li>Domain randomization (DR) enables sim-to-real transfer by training controllers on a distribution of simulated environments.</li><li>Simple policy gradient (PG) methods are often used to solve DR, but the theoretical guarantees are limited.</li><li>A convergence analysis of PG methods for domain-randomized linear quadratic regulation (LQR) is provided in this study.</li><li>The study shows that PG converges globally under suitable bounds on the heterogeneity of sampled systems, and proposes a discount-factor annealing algorithm.</li></ul>

Policy Gradient for LQR with Domain Randomization

Discover more