From Noise to Intent: Anchoring Generative VLA Policies with Residual Bridges

Yiming Zhong, Yaoyu He, Zemin Yang, Pengfei Tian, Yifan Huang, Qingqiu Huang, Xinge Zhu, Yuexin Ma · 2026

Read From Noise to Intent: Anchoring Generative VLA Policies with Residual Bridges by Yiming Zhong, Yaoyu He, Zemin Yang, Pengfei Tian, Yifan Huang, Qingqiu Huang, Xinge Zhu, Yuexin Ma (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-2604-21391.txt.

Description

Yiming Zhong and seven colleagues reframe generative robot policies from noise-first to intent-first generation, bridging meaning and physical control. Noting that motion splits into global intent and local dynamics, their ResVLA design uses spectral analysis to split a fixed low-frequency anchor from a random high-frequency leftover. Starting from predicted intent lets the model refine only local dynamics through a residual bridge. Tests show competitive performance, toughness to language and robot-body changes, and faster convergence, with strong real-world robot results too.

Catalog record

CREATORYiming Zhong, Yaoyu He, Zemin Yang, Pengfei Tian, Yifan Huang, Qingqiu Huang, Xinge Zhu, Yuexin Ma
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE61859
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2604.21391

Browse the full library · About the Tower of Babel