Principal Investigator Domitilla Del Vecchio
Real life control of any system has to be robust to imperfect state information. We study the problem of safety control for hybrid automata with imperfect continuous/discrete state information. Imperfect state information arises from poor or missing sensory information or by the presence of decision agents (humans, for example) that are not controllable. We work on theory and algorithms for hybrid state estimation and dynamic feedback for hybrid automata. We focus on problems of complexity and device techniques that rely on partial order theory and interval abstraction to provide efficient solutions.
The leading application is collision avoidance at traffic intersections, mergings and roundabouts, which we first implement in our multi-agent decision and control lab, in which we re-create collision instances among multiple vehicles (autonomous and human driven) on circular geometries (see the lab wiki for more details and for movies of our experiments). Then, we transfer the algorithms on the software system of the SMART Lab at the TOYOTA Technical Center of Ann Arbor. Finally, we transport the system on full vehicles and test it on the several miles long test track of Ann Arbor.