Exploration-Driven Generative Interactive Environments
Nedko Savov, Naser Kazemi, Mohammad Mahdi, Danda Pani Paudel, Xi Wang, Luc Van Gool · 2025
Read Exploration-Driven Generative Interactive Environments by Nedko Savov, Naser Kazemi, Mohammad Mahdi, Danda Pani Paudel, Xi Wang, Luc Van Gool (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-2504-02515.txt.
Description
Nedko Savov and co-authors train world models without a single human demonstration, in a 2025 paper. World models that predict video from actions usually eat costly recorded play. Their shortcut: let a random agent blunder through many cheap virtual worlds and learn from that. The catch is that random wandering visits boring corners. Their AutoExplore agent instead chases the model's own uncertainty, harvesting the scenes it understands least.
The explorer needs no game-specific rewards, so it ports anywhere. To feed pretraining, the team labeled the controls of 974 retro environments into one dataset. Their open rebuild of a leading world model, further enhanced, adapted to new worlds with better fidelity and control after the uncertainty diet.
Catalog record
| CREATOR | Nedko Savov, Naser Kazemi, Mohammad Mahdi, Danda Pani Paudel, Xi Wang, Luc Van Gool |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 57 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2504-02515