<ul><li>A new two-stage training framework is proposed for synthesizing neural controllers and Lyapunov function for continuous-time systems.</li><li>The framework utilizes a Zubov-inspired region of attraction characterization to estimate stability boundaries, reducing conservatism in training.</li><li>State-of-the-art neural network verifier is extended for automatic bound propagation and a novel verification scheme to avoid expensive bisection.</li><li>Experimental results show significant improvement in region of attractions and faster verification compared to traditional methods.</li></ul>

Two-Stage Learning of Stabilizing Neural Controllers via Zubov Sampling and Iterative Domain Expansion

Discover more