Hwang, Chaewon
Diffusion-EDFs: Bi-equivariant Denoising Generative Modeling on SE(3) for Visual Robotic Manipulation
Ryu, Hyunwoo, Kim, Jiwoo, An, Hyunseok, Chang, Junwoo, Seo, Joohwan, Kim, Taehan, Kim, Yubin, Hwang, Chaewon, Choi, Jongeun, Horowitz, Roberto
Equivariant Descriptor Fields (EDFs) [61] achieve dataefficient end-to-end learning on 6-DoF visual robotic manipulation Diffusion generative modeling has become a promising tasks by employing SE(3) bi-equivariant [37, approach for learning robotic manipulation tasks 61] energy-based models. However, EDFs require more from stochastic human demonstrations. In this paper, than 10 hours to learn from only a few demonstrations due we present Diffusion-EDFs, a novel SE(3)-equivariant to the inefficient training of energy-based models.