Sim2Real of real-time NMPC for autonomous driving using Digital Twin

The two key aspects of Advanced Driver Assistance Systems (ADAS) namely controls development and testing & validation need to go hand-in-hand with a continuous loop back between the two activities. 

The presentation will cover the work and some results of my industrial PhD. I will first discuss the practical aspects of the real-time deployment of an NMPC control scheme (formulation, validation and verification..) for motion planning on road vehicles. Second, I will present our safety enhanced collocation schemes allowing real-time NMPC computation, and showcase it for safety-critical collision avoidance situations. In the last part, I will introduce the concept of transferring the performance from simulation to real world components (Sim2Real) with better safety guarantees than standard approaches, while saving testing time and cost. This will be demonstrated through a method for black-box controller adaptation/tuning in the case of automated valet parking.