Synthesis of Controllers for Partially-Observable Systems: A Data-Driven Approach

This paper is concerned with the formal synthesis of safety controllers for partially-observable continuous-time polynomial-type systems with unknown dynamics. Given a continuous-time polynomial-type estimator with a partially-unknown dynamic and a known upper bound on the estimation accuracy, we propose a data-driven approach to compute a polynomial-type controller ensuring safety of the system. The proposed framework is based on a notion of so-called control barrier functions and only requires a single output trajectory collected from the system and a single state trajectory collected from its estimator. We show the application of our technique by synthesizing a safety controller for a partially-observable jet engine with unknown dynamics.