Gaussian Splatting as SE(3) Equivariant Features for Imitation Learning
Research Intership @ SNURPM
Sanghyun Hahn, Taekyun Ha, Inhee Lee
Sanghyun Hahn, Taekyun Ha, Inhee Lee
In this project, we achieve one-shot imitation learning of robotic manipulations with Gaussian Splatting. The key idea is that the Gaussian Features obtained from Gaussian Splatting carries more information compared to point clouds, enabling SE(3) pose matching between objects. Check the pdf for more information.