Correct-By-Construction Control Synthesis in Model-Based Design of Autonomous Systems

speaker: Ufuk Topcu

abstract: How can we affordably build trustworthy autonomous, networked systems? Partly motivated by this question, I describe a shift from the traditional "design+verify" approach to "specify+synthesize" in model-based engineering. I then discuss our recent results on automated synthesis of correct-by-construction, hierarchical control protocols. These results account for hybrid dynamics that are subject to rich temporal logic specifications and heterogenous uncertainties, and that operate in adversarial environments. They combine ideas from control theory with those from computer science, and exploit underlying system-theoretic interpretations to suppress the inherent computational complexity. The expressivity of the resulting design methodology enables us to formally investigate a number of emerging issues in autonomous, networked systems. I conclude my talk with a brief overview of several such issues from my ongoing projects: (i) compositional synthesis for the so-called fractionated systems; (ii) effects of perception imperfections on protocol synthesis; (iii) interfaces between learning modules and reactive controllers with provable guarantees of correctness; and (iv) human-embedded autonomy.

bio: Ufuk Topcu is a Research Assistant Professor in the Department of Electrical and Systems Engineering at the University of Pennsylvania. He received his Ph.D. from the University of California, Berkeley and was a Postdoctoral Scholar at the California Institute of Technology until 2012. His research is on the analysis, design, and verification of autonomous, networked systems.

speaker: Alwyn Goodloe

Critical cyber-physical systems, such as avionics, typically have one or more components that control the behavior of dynamical physical systems. The design of such control systems is well understood with mature and sophisticated foundations, but control engineers typically only work on Matlab/Simulink models, ignoring the implementation all together. I will speak about an ongoing collaboration with Prof. Eric Feron of Georgia Tech aimed at narrowing this gap. I will briefly describe the design of a Matlab to C translator being written in Haskell and verified using the Frama-C tool and the Prototype Verification System (PVS). In addition, I will give a survey of our efforts in enhancing PVS’ capabilities in this area by building a Linear Algebra library targeted at the math used by control engineers.