Delete README.md

README.md

@@ -1,606 +0,0 @@
-# World-Models
-
-<p align="center">
-  <strong>Build systems that do not only observe the world — but model how it changes.</strong>
-</p>
-
-<p align="center">
-  <img src="https://img.shields.io/badge/Predictive-Worlds-2563EB?style=for-the-badge" alt="Predictive Worlds">
-  <img src="https://img.shields.io/badge/Simulation-7C3AED?style=for-the-badge" alt="Simulation">
-  <img src="https://img.shields.io/badge/Embodied-AI-14B8A6?style=for-the-badge" alt="Embodied AI">
-  <img src="https://img.shields.io/badge/Planning-F59E0B?style=for-the-badge" alt="Planning">
-</p>
-
----
-
-## A research home for learned worlds
-
-**World-Models** is an independent Hugging Face organization for models, datasets, benchmarks, and experiments that explore one central idea:
-
-> **Intelligence becomes more useful when it can represent the world, imagine alternatives, and predict consequences before acting.**
-
-This organization is not about one architecture.
-
-It is about a family of systems that attempt to learn:
-
-- what exists
-- what changes
-- what causes change
-- what may happen next
-- what an action is likely to produce
-
----
-
-# The question
-
-Most AI systems answer:
-
-> **What is this?**
-
-World models aim to answer:
-
-> **What happens next?**
-
-And eventually:
-
-> **What happens if I act?**
-
-That shift changes the role of the model.
-
-It becomes less like a static predictor and more like an internal simulator.
-
----
-
-# A minimal world model
-
-```text
-observation
-    ↓
-internal state
-    ↓
-dynamics model
-    ↓
-possible future
-```
-
-An action-aware world model adds one more element:
-
-```text
-current state
-    +
-possible action
-    ↓
-predicted next state
-```
-
-Repeated over time:
-
-```text
-STATE₀
-  ↓ action
-STATE₁
-  ↓ action
-STATE₂
-  ↓ action
-STATE₃
-```
-
-That sequence can become the basis for planning.
-
----
-
-# Three things matter most
-
-## 1. Representation
-
-The system needs a useful internal description of the environment.
-
-That representation may contain:
-
-- objects
-- locations
-- motion
-- geometry
-- agents
-- memory
-- uncertainty
-- goals
-- interactions
-- physical constraints
-
-A good representation should preserve what matters for prediction.
-
----
-
-## 2. Dynamics
-
-A world is not a snapshot.
-
-It changes.
-
-A dynamics model tries to learn:
-
-```text
-state + action + time → next state
-```
-
-Possible domains include:
-
-- video
-- robotics
-- games
-- physical simulation
-- traffic
-- industrial systems
-- spatial environments
-- agent workflows
-
----
-
-## 3. Imagination
-
-Prediction becomes powerful when a model can compare multiple possible futures.
-
-```text
-             ┌─ action A → future A
-current state├─ action B → future B
-             └─ action C → future C
-```
-
-Planning can then ask:
-
-> Which future best satisfies the objective?
-
----
-
-# Research Paths
-
-World-Models may explore several distinct research paths.
-
-### Latent world models
-
-Compress observations into a learned internal state.
-
-Possible topics:
-
-- latent dynamics
-- compact state spaces
-- representation learning
-- predictive embeddings
-- recurrent state models
-
----
-
-### Video world models
-
-Use temporal visual data to learn how scenes evolve.
-
-Possible tasks:
-
-- future-frame prediction
-- video continuation
-- action-conditioned video
-- motion forecasting
-- object persistence
-- scene consistency
-
----
-
-### Spatial world models
-
-Build representations of space and geometry.
-
-Possible topics:
-
-- 3D scenes
-- depth
-- object relations
-- navigation
-- spatial memory
-- scene reconstruction
-- viewpoint changes
-
----
-
-### Embodied world models
-
-Connect perception to action.
-
-Possible environments:
-
-- robots
-- vehicles
-- drones
-- simulated agents
-- interactive environments
-
-The core loop:
-
-```text
-perceive
-  ↓
-model
-  ↓
-predict
-  ↓
-plan
-  ↓
-act
-  ↓
-perceive again
-```
-
----
-
-### Agent world models
-
-An agent may also need a model of a digital environment.
-
-For example:
-
-- available tools
-- system state
-- task progress
-- permissions
-- external services
-- likely tool outcomes
-
-This creates a digital world model:
-
-```text
-task state
-  ↓
-possible action
-  ↓
-predicted system response
-```
-
----
-
-# What We Want to Measure
-
-World models need evaluation beyond visual realism.
-
-A convincing future is not necessarily a correct future.
-
-Useful evaluation dimensions may include:
-
-| Dimension | Question |
-|---|---|
-| **State accuracy** | Is the internal world state meaningful? |
-| **Future prediction** | Is the next state plausible and correct? |
-| **Temporal consistency** | Does the world remain coherent over time? |
-| **Action sensitivity** | Do different actions produce different futures? |
-| **Spatial consistency** | Are geometry and positions preserved? |
-| **Object persistence** | Do entities remain stable across time? |
-| **Long-horizon stability** | Does quality degrade over many steps? |
-| **Planning utility** | Does the model help choose better actions? |
-| **Uncertainty** | Does the model represent multiple plausible futures? |
-
----
-
-# The Prediction Horizon Problem
-
-Short predictions are easier.
-
-Long predictions accumulate error.
-
-```text
-t+1   ✓
-t+2   ✓
-t+3   ~
-t+4   ?
-t+5   ??
-```
-
-One of the most interesting world-model problems is therefore:
-
-> **How long can a model remain coherent when predicting into the future?**
-
-This matters for:
-
-- robotics
-- long-horizon agents
-- simulation
-- autonomous systems
-- planning
-
----
-
-# Reality vs. Plausibility
-
-A world model can fail in two different ways.
-
-### Unrealistic prediction
-
-The generated future is obviously impossible.
-
-### Plausible but wrong prediction
-
-The future looks convincing, but does not match the true dynamics.
-
-The second failure is more dangerous because it can be harder to detect.
-
-World-model evaluation should therefore separate:
-
-**realism** from **predictive correctness**.
-
----
-
-# Possible Spaces
-
-World-Models can host practical research tools such as:
-
-### Future State Explorer
-Enter a state and candidate actions, then compare possible next states.
-
-### World Model Capability Map
-Document which environments, modalities, horizons, and actions a model supports.
-
-### Trajectory Comparator
-Compare predicted and real object or agent trajectories.
-
-### Long-Horizon Stability Lab
-Measure how prediction quality changes as rollout length increases.
-
-### Spatial Consistency Check
-Evaluate geometry, relative position, and object persistence.
-
-### Action-Conditioned Simulator
-Explore how different actions affect predicted futures.
-
-### World Model Eval Builder
-Create structured evaluation scenarios.
-
-### Counterfactual Playground
-Compare “what happened” with “what might have happened.”
-
-### Robot Rollout Viewer
-Inspect predicted action sequences before execution.
-
-### Latent State Explorer
-Visualize compact internal world representations.
-
----
-
-# Possible Datasets
-
-Potential datasets may include:
-
-```text
-state-transition-pairs
-action-conditioned-trajectories
-robot-rollouts
-spatial-scene-transitions
-synthetic-physical-worlds
-video-future-sequences
-counterfactual-transitions
-agent-environment-traces
-```
-
-A useful world-model dataset might include:
-
-```json
-{
-  "observation": "...",
-  "state": "...",
-  "action": "...",
-  "next_state": "...",
-  "timestamp": "...",
-  "reward": "...",
-  "uncertainty": "..."
-}
-```
-
----
-
-# Possible Models
-
-Projects may include models for:
-
-- next-state prediction
-- video dynamics
-- action-conditioned generation
-- trajectory prediction
-- latent dynamics
-- scene reconstruction
-- spatial reasoning
-- robot planning
-- environment simulation
-- predictive control
-
----
-
-# World Models Are Not Just Video Generators
-
-Video generation can be part of a world model.
-
-But visual realism alone is not enough.
-
-A useful world model may need to preserve:
-
-- causality
-- geometry
-- identity
-- temporal state
-- physical constraints
-- action consequences
-- uncertainty
-
-The goal is not simply:
-
-> generate the next frame
-
-The stronger goal is:
-
-> **model what the world is doing**
-
----
-
-# World Models + Omnimodal Intelligence
-
-A world model may receive information from many modalities:
-
-```text
-vision
-audio
-language
-3D
-sensors
-tools
-actions
-```
-
-These observations can update a shared internal state.
-
-That makes world models a natural bridge between:
-
-**omnimodal perception** and **intelligent action**.
-
----
-
-# World Models + Spatial Intelligence
-
-Spatial intelligence helps represent:
-
-- where objects are
-- how they relate
-- how they move
-- what is reachable
-- what is occluded
-- what changes under action
-
-World models add the time dimension:
-
-> not only **where things are**, but **where they may be next**.
-
----
-
-# World Models + Agents
-
-A capable agent may need to simulate consequences before acting.
-
-```text
-goal
- ↓
-possible plan
- ↓
-world model
- ↓
-predicted outcome
- ↓
-compare alternatives
- ↓
-choose action
-```
-
-That may become especially important for:
-
-- long-horizon tasks
-- autonomous agents
-- robotics
-- tool use
-- planning under uncertainty
-
----
-
-# Principles
-
-### Predict consequences
-A world model should connect action and outcome.
-
-### Preserve state
-Important entities should remain coherent over time.
-
-### Model uncertainty
-The future is rarely singular.
-
-### Evaluate rollouts
-One-step accuracy is not enough.
-
-### Separate realism from correctness
-A plausible prediction can still be wrong.
-
-### Connect perception to planning
-Prediction becomes more useful when it improves decisions.
-
-### Prefer measurable environments
-World-model claims should be tested against observable outcomes where possible.
-
----
-
-# Who Is This For?
-
-World-Models may be useful for:
-
-- world-model researchers
-- robotics teams
-- embodied-AI developers
-- reinforcement-learning researchers
-- video-model teams
-- agent developers
-- spatial-intelligence researchers
-- simulation engineers
-- autonomous-systems teams
-- open-source contributors
-
----
-
-# The Long View
-
-Language models taught machines to model language.
-
-World models aim at something broader:
-
-> **model environments, dynamics, actions, and possible futures.**
-
-If future AI systems are expected to operate in the world rather than only describe it, they may need internal predictive models that answer:
-
-```text
-Where am I?
-What is happening?
-What could happen next?
-What if I act?
-Which action should I choose?
-```
-
-That is the territory of World-Models.
-
----
-
-# Research, not certainty
-
-Projects published here are intended primarily for:
-
-- research
-- development
-- evaluation
-- simulation
-- education
-- prototyping
-
-Predicted futures are not guaranteed futures.
-
-World-model outputs should not be treated as reliable control instructions in safety-critical environments without extensive validation.
-
----
-
-# Independent Organization
-
-**World-Models is an independent Hugging Face community organization.**
-
-It is not an official model provider, research institute, robotics company, standards organization, or Hugging Face organization.
-
-The organization exists to explore a simple but ambitious idea:
-
-> **AI systems that can represent a world, predict its evolution, and reason about possible futures.**
-
----
-
-<p align="center">
-
-# WORLD-MODELS
-
-### **Represent the world. Imagine futures. Plan before acting.**
-
-</p>