EgoFlow: Gradient-Guided Flow Matching for Egocentric 6DoF Object Motion Generation

Abhishek Saroha, Huajian Zeng, Xingxing Zuo, Daniel Cremers, Xi Wang · 2026

Read EgoFlow: Gradient-Guided Flow Matching for Egocentric 6DoF Object Motion Generation by Abhishek Saroha, Huajian Zeng, Xingxing Zuo, Daniel Cremers, Xi Wang (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2604-01421.txt.

Description

Saroha and colleagues present EgoFlow, a flow-matching system making plausible trajectories from first-person video. It uses a hybrid Mamba-Transformer-Perceiver design to model motion, scene shape and meaning. Gradient-guided steps enforce physics constraints such as crash dodging and motion smoothness. That yields smooth, steerable motion without post-hoc cleanup or extra labels. Tested on HD-EPIC, EgoExo4D and HOT3D, it beats diffusion and transformer rivals in accuracy, generalization and physical realism. It cuts crash rates by 79 percent.

Catalog record

CREATORAbhishek Saroha, Huajian Zeng, Xingxing Zuo, Daniel Cremers, Xi Wang
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE36361
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2604.01421

Browse the full library · About the Tower of Babel