Fei-Fei Li & World Labs  ·  June 2026

A Functional
Taxonomy of
World Models

Renderers, Simulators, Planners — and the loop that connects them all.

Spatial Intelligence · AI Frontier
01   The overloaded term problem 02   The POMDP loop 03   Three functional kinds 04   Why simulation is the linchpin 05   The convergence thesis
The Problem

"World model" is one of the most important
— and most overloaded — terms in AI today.

Computer vision, robotics, reinforcement learning, and generative AI all claim to be building world models. Each means something quite different.

"A video model that produces gorgeous but physically impossible flames, a language model improvising a playable game, and a physics engine that faithfully simulates combustion all go by the same name."

— Fei-Fei Li
The Loop Beneath the Taxonomy

Every world model is a projection of one ancient diagram

Agent
A person, robot, or software system that takes actions based on observations.
World State
Actions → State
Agent actions affect world state
State → Observations
Agent perceives partial view only
The Three Outputs
Renderer → pixels Simulator → state Planner → actions

The POMDP loop — formalised by Sutton & Barto — is where the term "world model" got its technical meaning. Traced to Kenneth Craik's 1943 proposal that minds reason by running "small-scale models" of reality.

The Taxonomy

Three functional kinds of world model

🎬 Renderer

Outputs observations — pixels for human eyes. A video model turning a prompt into a drone shot is a renderer. Quality that matters: visual fidelity. Constraint: no explicit 3D understanding; beautiful output can be physically impossible.

Examples: text-to-video, Genie 3, World Labs RTFM

🤖 Planner

Outputs actions. Given an observation and a goal, a planner answers "what should the agent do next?" — the inverse of the renderer. VLA models and World Action Models are planners closing the perception-action loop.

Examples: Vision-Language-Action models, model-based RL systems

Maturity Landscape

Where each category actually stands today

The three functions are at wildly different stages of commercial and technical maturity.

🎬 Renderer — Most Mature

Image- and text-to-video products are expanding rapidly in consumer and enterprise markets. Google's Nano Banana model has put renderer-quality generation in the hands of hundreds of millions. The ceiling: visual plausibility, not physical accuracy.

🤖 Planner — Most Nascent

The most intriguing and most nascent. Robotic demos in the last two years look impressive — but almost all are confined to heavily constrained lab setups. The gap between a demo reel and a robot that reliably works in a kitchen remains vast.

🌉 Simulation Is the Bridge

A model mastering simulation can project into pixels (for renderers) and into action predictions (for planners). A model mastering only rendering, or only planning, cannot do either. The bridge is the point of leverage.

Why Simulation Is the Linchpin

The hardest open problems in the field all live in simulation

01
Data scarcity is acute

3D data with explicit geometry, material properties, and physical annotations is orders of magnitude scarcer than internet video. Renderers are data-rich; simulators and planners are not.

02
The sim-to-real gap persists

The difference between how things behave in simulation versus reality remains a core, unsolved problem. AI-generated geometry can look correct while containing self-intersections or wrong scale that produce nonsensical physics.

03
Multi-physics is brutally expensive

Rigid bodies, deformable objects, fluids, and cloth interacting simultaneously remains orders of magnitude more expensive than single-domain simulation.

04
The market surface area is enormous

Robotics training, AV testing, architectural visualisation, engineering simulation, and drug discovery all depend on something simulation-shaped. NVIDIA Omniverse alone targets >$1T addressable market.

World Labs · Marble

The first concrete move into the simulation territory

What Marble Does

Multimodal → Explorable 3D

Takes multimodal prompts — text, image, video, or spatial sketch — and generates explorable 3D environments. Outputs Gaussian splats for visual exploration alongside collision meshes a physics engine can operate on.

The lines between rendering, simulation, and planning are beginning to collapse — Marble is an early signal of that convergence.

The Convergence Thesis

Three separate research threads are becoming one

The shared insight: the knowledge required to render a world, simulate it, and act in it is largely the same. A model that truly understands a cup on a table should render it, simulate a push, and plan a hand to pick it up.

1
Passive Output
Each type produces one kind of output independently
2
Action-Conditioned
Renderers become interactive; conditioned on user input
3
Shared Backbone
Pretrained renderer used as backbone for joint world & action prediction
4
Unified Architecture
One model, multiple output modalities: pixels, state, actions
5
Spatial Intelligence
Machines that understand, imagine, reason, and act in the physical world
"Every level is moving from passive output to interactive system." — Fei-Fei Li
The Long Arc

A bet the field has been making
since the late 1980s.

A sufficiently rich model of the world is all that any agent needs — to see worlds, build them, and act in them. What gives this "big bet" weight now is that three threads, each already driving multi-billion-dollar industries on its own, are starting to behave like one.

"Language gave machines a way to talk about the world. World models are how machines will finally come to understand, imagine, reason and interact with it."

— Fei-Fei Li
The Open Questions · What Comes Next

The defining open problem: unifying rendering, simulation, and planning inside one architecture.

The data picture is uneven — renderers are awash in internet video while simulators face 3D asset shortages. Optimising for visual beauty can sacrifice the precision a robot needs. Reconciling these tensions is the central research challenge.

Source: A Functional Taxonomy of World Models, Fei-Fei Li & World Labs, June 2026

01
Read the full essay
drfeifei.substack.com — Fei-Fei Li's original post with deeper technical detail and animations of Marble.
02
Explore World Labs
worldlabs.ai — see Marble in action and follow the research programme building toward unified world models.
03
Watch the renderer → planner bridge
Several robotics labs have shown that a pretrained video renderer can be used as the backbone for joint world-and-action prediction — the nascent convergence is already underway.
1 / 10
← → to navigate · click left or right