notesum.ai

Published at November 27

G3Flow: Generative 3D Semantic Flow for Pose-aware and Generalizable Object Manipulation

cs.RO

cs.AI

cs.CV

cs.SY

eess.SY

Released Date: November 27, 2024

Authors: Tianxing Chen¹, Yao Mu¹, Zhixuan Liang¹, Zanxin Chen², Shijia Peng², Qiangyu Chen², Mingkun Xu³, Ruizhen Hu², Hongyuan Zhang¹, Xuelong Li⁴, Ping Luo¹

Aff.: ¹The University of Hong Kong; ²Shenzhen University; ³Guangdong Institute of Intelligence Science and Technology; ⁴Institute of Artificial Intelligence (TeleAI), China Telecom

Arxiv: http://arxiv.org/abs/2411.18369v1

	Shoe Place (T)	Dual Shoes Place (T)	Tool Adjust (T)	Bottle Adjust (T)	Average
DP	$26.0\pm{11.4}$	$3.3\pm{1.2}$	$21.7\pm{5.1}$	$16.3\pm{4.6}$	16.8 ( $\downarrow$ 51.5)
3D DP	$54.0\pm{6.9}$	$13.0\pm{1.7}$	$43.3\pm{9.1}$	$74.3\pm{8.6}$	46.2 ( $\downarrow$ 22.1)
3D DP w/ color	$58.0\pm{3.0}$	$7.0\pm{1.0}$	$70.3\pm{9.3}$	$41.0\pm{12.8}$	44.1 ( $\downarrow$ 24.2)
DP w/ G3Flow	$\textbf{83.0}\pm{3.6}$	$\textbf{24.0}\pm{3.6}$	$\textbf{84.3}\pm{10.1}$	$\textbf{82.0}\pm{4.0}$	68.3