Skip to contentSkip to filters

We use cookies for analytics.

Do we need diffusion in robotics?

Max Simchowitz

Carnegie Mellon University

Recording

49 min

Abstract

Generative control policies are often credited with representing multimodal robot actions. This talk tests that explanation through experiments and theory, finding benefits instead associated with iterative computation, training noise and on-manifold biases. The results motivate alternatives that obtain those algorithmic advantages without relying on distribution matching as the explanation.

Topics