Koopman Operators in Robotics

A full-day workshop at Robotics: Science and Systems

Abstract

Recent advances in Koopman operators have opened the door to exciting opportunities within robotics that span modeling and control, human-robot interaction, and soft robotic systems. Koopman operators are linear representations of nonlinear dynamical systems in a larger, potentially infinite space. They offer a solution for modeling nonlinear dynamics as linear dynamical systems. As a result, Koopman operators enable a paradigm shift that impacts how we model complex robotic systems and perform learning and control tasks. With Koopman operators it becomes possible to leverage linear systems analysis and linear control theory on linear descriptions of robot dynamics, ultimately improving the capabilities of complex robot systems.

While Koopman operators are promising, the current theory is still in its infancy. The purpose of this workshop is to bring together experts in robotics, machine learning, and Koopman operator theory to discuss recent developments, challenges, and future research directions. Furthermore, the workshop will facilitate discussion on alternative formulations using Koopman-like structures, applicability of Koopman operators in unconventional areas in robotics, and identify promising directions of Koopman operators in robotics.

Topics of Discussion

What is the Koopman operator and how does it help with interpretability and data?
Learning Capabilities and Analytical Limits
Limitations and Control in Lifted Spaces
Future Open Areas: Lifted Dynamic Constraints