Formal Guarantees in Data-Driven Model Identification and Control Synthesis

Abstract

For many performance-critical control systems, an accurate (simple) model is not available in practice. Thus, designing controllers with formal performance guarantees is challenging. In this paper, we develop a framework to use input-output data from an unknown system to synthesize controllers from signal temporal logic (STL) specifications. First, by imposing mild assumptions on system continuity, we find a set-valued piecewise affine (PWA) model that contains all the possible behaviors of the concrete system. Next, we introduce a novel method for STL control of PWA systems with additive disturbances. By taking advantage of STL quantitative semantics, we provide lower-bound certificates on the degree of STL satisfaction of the closed-loop concrete system. Illustrative examples are presented.