Understand the world.
Imagine what’s next.
Exploring the representations, predictions, and actions behind world models.
03Beyond the pixels
Our field of viewUnderstanding what is.
Imagining what comes next.
Learn the structure.
Explore the future.
World models learn how an environment changes. By predicting in a learned representation, they can imagine outcomes and help plan the next action.
Our first toolkit brings together joint-embedding architectures, action-conditioned models, and the environments to explore them.
Explore our researchResearch areas
Three connected areas, one shared toolkit.
Representation learning
Learning useful representations from images, video, and language through joint-embedding prediction.
Predictive world models
Predicting how the world changes in response to actions, with shared interfaces for building and comparing models.
Planning & environments
Turning predictions into actions through trajectory planning, pixel environments, and simulated robotics tasks.
Build with us.
Follow the work. Bring your ideas.
Find us on GitHub