PhD Dissertation Defense - David Millard

PhD Dissertation Defense - David Millard

Committee Members: Gaurav Sukhatme (chair), Lars Lindemann, Somil Bansal

Title: Augmented Simulation Techniques for Robotic Manipulation

Abstract: The development and application of capable robot manipulators require advances in world modeling and simulation. This thesis provides a comprehensive overview of our work in simulation methodologies, covering diverse physical phenomena and target applications, and with augmented structures that make associated computation tractable. First, we present work in rigid body dynamics with contact, augmented for fast computation of first and second order derivatives and probability distributions, and present applications to parameter estimation and control. We then present work in soft object modeling and control, first by using differentiable solid mechanics for constrained parameter estimation and then by machine learning based predictive control. Finally, we present work on GPU accelerated parallel Discrete Element Methods DEM and their applicability to the challenges of robotic sampling and excavation. To demonstrate the translational utility of this work, we present results from several methodologies from real data or on real hardware