Interactive lesson~20 minAdvanced
World Models
World models learn compact simulations of how environments evolve. They let agents imagine outcomes before acting.
PlanningLatent stateSpatial AI
Mental model
Planning gets cheaper when the agent can rehearse internally.
Robotics, games, autonomous systems, and spatial AI need models that predict consequences, not just labels.
Planning quality
balanced69% modeled signal
Compute cost
balanced66% modeled signal
Grounding
balanced63% modeled signal
Concept pipeline
Build the idea in four moves
Interactive lab
Tune an agent that plans with imagined rollouts.
Observe
Encode state from pixels, sensors, or text.
Focus lens
The part that usually clicks late
Latent state
A compressed state can hold what matters for prediction.
Planning quality
69
Compute cost
66
Grounding
63
Knowledge check
What does a world model enable?
Next horizon
Where this topic is headed
Video world models
Spatial intelligence
Model-based RL