World-model data glossary

World model

Definition

A learned representation of how an environment changes over time. It can predict future observations or latent states and support planning through simulated rollouts.

Why it matters

Dataset requirements follow the transition the model must learn. A world model trained without actions may learn visual continuity but lack a reliable control interface.

Example

Given recent game frames and the current control input, a model predicts the next visual state and selected engine variables.