World models.
One research library.

Compares predictive world-model representations with human behavior and macaque cortex, identifying strengths in appearance-invariant motion coding and a remaining temporal-integration gap.

Origin of the concept. First articulation that organisms carry a "small-scale model" of external reality in their heads, enabling them to try out alternatives and react to future situations before they arise.

Classic behavioral evidence that animals learn map-like internal representations of the environment rather than mere stimulus-response chains — the ancestral "world model" in psychology.

Landmark neuroscience synthesis identifying the hippocampus (place cells) as the neural substrate of Tolman's cognitive map; foundation for spatial world-model research.

Cognitive-science theory that reasoning operates over constructed internal models of situations rather than formal logic rules — a direct intellectual ancestor of "mental world modeling."

First computational formalization of a spatial world model for a physical agent.

Foundational representation: encode the future occupancy of states rather than immediate reward.

Canonical evidence that the brain uses forward models to predict sensory consequences of motor commands — the neuroscience blueprint for action-conditioned prediction.

The concrete computational predictive-coding model (predating Friston's free-energy generalization): higher cortical areas predict lower-level activity and feed back prediction errors.

Neuro-scientific grounding: the brain as a hierarchical Bayesian inference machine minimizing prediction error.

Humans use fast approximate physics simulators as a world model for intuitive physics.

Seminal work. Learn a compressed (V) perception model + recurrent (M) world model, train a small controller (C) entirely inside imagination. Introduced MDN-RNN for stochastic world model.