Seminar - Learning for Autonomy: From Policies to Operators to Distributions - Mar. 10
Adam Thorpe
Postdoctoral Researcher, University of Texas at Austin
Monday, Mar. 10 | 10 a.m. | AERO 114
Abstract: Autonomous systems must learn, adapt, and make decisions in novel, unpredictable environments. However, data-driven approaches often struggle to generalize and can be fragile in such environments. My research addresses this challenge by developing learning-based methods that explicitly incorporate mathematical structure, enabling autonomy to adapt, scale, and generalize.Ìý
Central to this approach are structured representations of policies, operators, and distributions that leverage Hilbert space theory, statistical learning, and neural network models. In this talk, I will present results demonstrating how autonomous systems can adapt and transfer to new scenarios within seconds using minimal online data—without the need for additional retraining.Ìý
I will highlight advances in neural operator learning, where efficient function-to-function mappings achieve orders-of-magnitude improvements in accuracy over state-of-the-art methods, with significant implications for autonomy. Finally, I will discuss efforts to design autonomous systems that operate safely around humans by tailoring responses to individual needs and preferences.
Bio: Adam Thorpe is a postdoctoral researcher at the Oden Institute for Computational Engineering and Sciences at The University of Texas at Austin. He received his PhD at the University of New Mexico in 2023. Adam's research interests are in the area of data-driven and learning-based control, with applications to humans and autonomy, space systems, and robotics.
Ìý