V-JEPA 2: Self-Supervised Video Models Enable Understanding, Prediction and Planning

Mahmoud Assran, Adrien Bardes, David Fan, Quentin Garrido, Russell Howes, Mojtaba Komeili, Matthew Muckley, Ammar Rizvi, Claire Roberts, Koustuv Sinha, Artem Zholus, Sergio Arnaud, Abha Gejji, Ada Martin, Francois Robert Hogan, Daniel Dugas, Piotr Bojanowski, Vasil Khalidov, Patrick Labatut, Francis · 2025

Read V-JEPA 2: Self-Supervised Video Models Enable Understanding, Prediction and Planning by Mahmoud Assran, Adrien Bardes, David Fan, Quentin Garrido, Russell Howes, Mojtaba Komeili, Matthew Muckley, Ammar Rizvi, Claire Roberts, Koustuv Sinha, Artem Zholus, Sergio Arnaud, Abha Gejji, Ada Martin, Francois Robert Hogan, Daniel Dugas, Piotr Bojanowski, Vasil Khalidov, Patrick Labatut, Francis (2025) 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-2506-09985.txt.

Description

Assran and a large team at Meta present V-JEPA 2, a video model that learns the physical world by watching. The goal is machines that understand, predict, and plan from observation, much as children do. Training used over a million hours of internet video plus images, with no action labels.

The model reads motion well, topping a standard action test, and it leads on guessing what a person will do next in kitchen videos. Linked to a language model, it also set records on video question tasks at its size. A robot-ready version, trained on under sixty-two hours of robot clips, plans movements in new settings. The authors frame video pretraining as a path to world models.

Catalog record

CREATORMahmoud Assran, Adrien Bardes, David Fan, Quentin Garrido, Russell Howes, Mojtaba Komeili, Matthew Muckley, Ammar Rizvi, Claire Roberts, Koustuv Sinha, Artem Zholus, Sergio Arnaud, Abha Gejji, Ada Martin, Francois Robert Hogan, Daniel Dugas, Piotr Bojanowski, Vasil Khalidov, Patrick Labatut, Francis
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE133 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2506-09985

Browse the full library · About the Tower of Babel