About the speaker
Ruggero Carli received the Laurea degree in computer engineering and the Ph.D. degree in information engineering from the University of Padova, Padua, Italy, in 2004 and 2007, respectively. From 2008 to 2010, he was a Postdoctoral Fellow with the Department of Mechanical Engineering, University of California, Santa Barbara. He is currently an Associate Professor with the Department of Information Engineering, University of Padova. His research interests include distributed optimisation, estimation and control, nonparametric estimation, and learning-based control for robotic systems.
Abstract
In the last decades, Machine Learning (ML) and Deep Learning (DL) algorithms proved promising solutions to solve complex problems, ranging from modeling, classification, regression, and control. These algorithms are particularly effective in settings where a large amount of data is available, such as virtual environments. As an example, DL algorithms reach super-human performance in playing Chess, Shogi, and Go. On the contrary,when dealing with physical systems, the number of samples available is limited, possibly compromising the effectiveness of these technologies. This motivates the interest in data-efficient ML and DL algorithms for physical systems, also known as physics-informed models. These algorithms aim at limiting interaction time on the actual system by exploiting prior knowledge on the underlying dynamics. In this talk, we will present a class of physics-informed solutions based on Gaussian Process Regression (GPR). The first part of the talk will discuss the application of GPR to inverse dynamics identification. Instead, in the last part, we will present MC-PILCO, a data-efficient Reinforcement Learning algorithm based on GRP.