World-In-World/world-in-world

Code implementation of the paper "World-in-World: World Models in a Closed-Loop World" (ICLR'26 Oral)

37
/ 100
Emerging

This project offers a standardized way to test how well visual world models help intelligent agents perform real-world tasks like navigating environments, answering questions about what they see, or manipulating objects. It takes a trained visual world model and task data as input, then provides clear metrics on how effectively the model improves the agent's ability to act and perceive within a simulated environment. This is for researchers and engineers developing embodied AI agents and advanced robotics.

139 stars.

Use this if you need a reliable benchmark to measure the practical utility of visual world models for embodied agents, beyond just how realistic their generated images look.

Not ideal if you are looking for a simple, off-the-shelf solution for a specific robotics task, as this is primarily an evaluation framework for advanced AI models.

embodied-AI robotics-research agent-simulation model-evaluation AI-benchmarking
No Package No Dependents
Maintenance 10 / 25
Adoption 10 / 25
Maturity 13 / 25
Community 4 / 25

How are scores calculated?

Stars

139

Forks

2

Language

Python

License

MIT

Last pushed

Feb 15, 2026

Commits (30d)

0

Get this data via API

curl "https://pt-edge.onrender.com/api/v1/quality/generative-ai/World-In-World/world-in-world"

Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.