Parameter Estimation for Deformable Objects

This lecture presented by David Millard focuses on parameter estimation for deformable objects in robotic manipulation tasks, particularly in the context of autonomous maintenance on the International Space Station. The lecture explores the challenges of dealing with complex dynamics of deformable objects, the use of finite element techniques for solving these dynamics, and the concept of interpretability and debugability in robotic systems. The main focus is on parameter estimation for deformable objects using a Neo-Hookian elasticity model, where the elasticity parameters lambda and mu are estimated based on sensor measurements and trajectories. The lecture introduces a novel differentiable finite element simulator framework for jointly estimating state and parameters through linear programs, leveraging sparsity and symmetry techniques for efficient computation.