- Starts: 11:15 am on Friday, October 2, 2026
- Ends: 12:00 pm on Friday, October 2, 2026
Title: Control Barrier Function-Based Safety Filters: Equilibria, Stability, and Policy Optimization
Abstract: Control barrier function (CBF)-based methods are widely used in feedback control to enforce constraints while preserving performance. When embedded in feedback, however, these methods induce closed-loop dynamics that can fundamentally affect equilibrium selection and stability, even when feasibility is maintained. We first analyze CBF-based safety filters and show that, for a fixed safe set, equilibrium and boundary behavior are structurally independent of the specific CBF parametrization, highlighting fundamental limitations of CBF tuning as a design strategy. Motivated by this limitation, we identify the design variables that shape the closed-loop behavior: the nominal controller and the safety-filter parameters. We formulate their joint design as a trajectory-based policy optimization problem with explicit stability constraints. To solve this problem, we develop a robust safe gradient flow method using rollout-based gradient estimates. By construction, nominal closed-loop stability is preserved at every policy iterate.
- Location:
- EMB 105
- Hosting Professor
- Emiliano Dall'Anese
