Abstract 

Two key elements characterize today’s modern decision-making problems: an abundance of historical data and tasks that are entirely or partially repetitive. While the requirements are state and input constraint satisfaction, performance is assessed by evaluating the cost associated with the closed-loop trajectories. Improving the performance of decision-making algorithms by leveraging historical data has been an active theme of research in the past few decades. The key idea is to use recorded state-input pairs in order to compute at least one of the following three components: i) a model which describes the evolution of the system; ii) a safe set of states (and an associated control policy) from which the control task can be safely executed; and iii) a value function which represents the cumulative closed-loop cost from a given state of the safe set.

This presentation will first provide an overview of the theory of Learning Model Predictive Control, showing in particular, how to leverage data in the decision-making process while ensuring safety, exploration, and performance improvement. The second part of the presentation will show that the proposed methodology can be used to teach a full-size real-world autonomous vehicle how to race, and culminate in a discussion on preliminary work to extend this strategy to multi-agent racing.

Bio

Ugo Rosolia is a Senior Research Scientist and Science Manager at Amazon. He received the B.S. and M.S. degrees (cum laude) in mechanical engineering from Politecnico di Milano in 2012 and 2014, respectively. Moreover, he was a Visiting Scholar at the Tongji University in Shanghai in the context of the PoliTong Double Degree Program, (fall 2010 - spring 2011), as well as at the University of Illinois at Urbana-Champaign (fall 2013 - spring 2014), sponsored by a Global E3 Scholarship. He was also a Research Engineer with Siemens PLM Software in Belgium (spring and summer 2015). Subsequently, he achieved his Ph.D. degree in mechanical engineering at the University of California, Berkeley in 2019, followed by a stint as a postdoctoral scholar at the California Institute of Technology (2020-2021). His current research interests include approximate dynamic programming, system identification, decision making in mixed observable Markov decision processes, and predictive control.