ESE Ph.D. Thesis Defense: “Learning Environmental Models with Multi-Robot Teams Using a Dynamical Systems Approach”

Robots have the capability to sense and track natural phenomena for environmental monitoring, deepening our understanding of the world. Robotic modeling of such phenomena is essential to operating in complex environments, allowing robots to perform in more realistic scenarios. Thus, representing complex environments is paramount to the success of multi-robot teams. While considerable efforts have been made for modeling with multi-robot teams, specifically in coordination and distributed methods, these techniques have limitations in spatiotemporal, complex environments. These environments can be vastly different, such as fluid flows, oceans, and space. Robots operating in these environments typically create representations of their surroundings using computationally expensive techniques or by leveraging human expert knowledge.

Interestingly, though these environmental processes may seem unrelated, they can all be analyzed with dynamical systems theory. This thesis presents methods for representing the environment as a dynamical system with machine learning techniques. We formulate machine learning methods that lend to data-driven modeling of the phenomena for robotic applications, specifically using dimensionality reduction techniques and kernel methods. The data-driven modeling explicitly leverages theoretical foundations of dynamical systems theory. Dynamical systems theory offers mathematical and physically interpretable intuitions about the environmental representation. The procedures presented include distributed algorithms, online adaptation, uncertainty quantification, and feature extraction to allow for the actualization of these techniques on-board robots. The environmental representations guide robot behavior in developing strategies such as optimal sensing and energy-efficient navigation. The methods and procedures provided in this thesis were verified across prototypical environments and on experimental robots.