Wall-OSS-0.5 Technical Report Pretrain Once, Act Anywhere

X Square Robot Team · 2026

Read Wall-OSS-0.5 Technical Report Pretrain Once, Act Anywhere by X Square Robot Team (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-2605-30877.txt.

Description

Yu and colleagues test whether pretraining can give a vision-language-action model robot skill, rather than a good start for tuning. Their open Wall-OSS-0.5 pairs a vision-language base with action parts and trains on over a million runs across twenty types. A three-part mix routes gradients into the base, keeps language skill, and sets flow matching as the action step. Before task tuning, it completes held-out tasks on robots, and after tuning it beats a rival on robot progress.

Catalog record

CREATORX Square Robot Team
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE83130
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.30877

Browse the full library · About the Tower of Babel