Data-driven Learning and Control - DDLC

Abstract: Advancing autonomous systems requires not only improving the control of complex dynamical systems, but also achieving complex tasks in challenging environments. Learning has emerged as a promising means to practically address these challenges; however, the recovery of guarantees, particularly concerning safety, is often still lacking. This talk will highlight our results towards addressing this problem by building on an optimization-based control paradigm. I will begin by defining our notion of safety and how it can be effectively formulated as a planning problem. The talk will then address concepts of learning dynamics, objective functions, and constraint functions, which require a careful tradeoff between achievable guarantees and performance, while managing computational efficiency.  The results in this presentation will be illustrated with applications from autonomous racing and robotics.