The Fourier Transporter (FOURTRAN) is introduced as a method to improve sample efficiency in robotic manipulation tasks by leveraging bi-equivariant symmetries. The paper addresses the challenges of sample efficiency in complex robotic manipulation tasks, especially in 3D environments. By proposing an open-loop behavior cloning method trained using expert demonstrations, FOURTRAN aims to predict pick-place actions on new configurations efficiently. The method utilizes a fiber space Fourier transformation for memory-efficient computation and achieves state-of-the-art results across various tasks on the RLbench benchmark. The architecture of FOURTRAN involves modeling SE(3) bi-equivariance using 3D convolutions and a Fourier representation of rotations, allowing for high sample efficiency and angular resolution.
To Another Language
from source content
arxiv.org
Key Insights Distilled From
by Haojie Huang... at arxiv.org 03-19-2024
https://arxiv.org/pdf/2401.12046.pdfDeeper Inquiries