Manipulation skill generalization across different object requires accounting for their variations, while still transferring relevant past experience. Even within a category, object shape can vary in complex, nonlinear ways, that are difficult to capture when adjusting an existing policy. We propose a new approach to skill transfer to novel objects that accounts for known categorical shape variation. A low-dimensional shape representation is learned from a set of deformations sampled between known objects within a category. This latent representation is mapped to the set of control parameters that result in successful execution of a category-level skill on that object. This method generalizes a learned manipulation policy to even complex unseen objects with few training examples.

My goal, through SuperUROP, is to make a positive contribution to the fields of robotics and embodied intelligence. I’ m excited to apply what I’ ve learned in my previous courses on AI, robotic manipulation, and machine learning towards this long-term research project.