stable-worldmodel-v1: Reproducible World Modeling Research and Evaluation
Lucas Maes* Quentin Le Lidec* Dan Haramati Nassim Massaudi, [3pt] Damien Scieur Yann LeCun Randall Balestriero, [8pt] Mila & Université de Montréal New York University · 2026
Read stable-worldmodel-v1: Reproducible World Modeling Research and Evaluation by Lucas Maes* Quentin Le Lidec* Dan Haramati Nassim Massaudi, [3pt] Damien Scieur Yann LeCun Randall Balestriero, [8pt] Mila & Université de Montréal New York University (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-2602-08968.txt.
Description
World-model research suffers from scattered, one-off code that is hard to reuse or compare. Lucas Maes and six co-authors introduce stable-worldmodel, a modular, tested, and documented toolkit for learning compact predictive models of how environments change. It bundles data-collection tools, standard test worlds, planning methods, and baselines in one place, with tunable visual and physical traits in each world.
Those controls support robustness and continual-learning studies that one-off code makes awkward. As a demo, the team uses the toolkit to study zero-shot robustness in DINO-WM, a world model built on self-supervised vision features. The project aims to cut bugs and make world-model results genuinely comparable.
Catalog record
| CREATOR | Lucas Maes* Quentin Le Lidec* Dan Haramati Nassim Massaudi, [3pt] Damien Scieur Yann LeCun Randall Balestriero, [8pt] Mila & Université de Montréal New York University |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 23 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-08968