Paper accepted to ACC 2026: ReLU Barrier Functions for Nonlinear Systems with Constrained Control: A Union of Invariant Sets Approach

This paper led by Pouya Samanipour, Ph.D., present his work on finding valid barrier functions parametrized as ReLU neural networks using linear programming techniques. Many barrier functions are not valid, and represent best-efforts at safety, because valid barrier functions are thought to be too hard to find. Pouya’s work continues his effort to make valid barrier functions straightforward to find, even for nonlinear systems.

Pouya developed methods that reuse computations, and enlarge the forward invariant set, leading to improved estimates of the unknown forward invariant set for some given closed-loop dynamical system.

This work was supported by NSF grant 2330794.