Introduces textual belief states and factorized GRPO to enforce strict latent state mediation in text-based world models, yielding preserved prediction accuracy with large gains in representation quality and rollout performance on TextWorld and ScienceWorld.
super hub Canonical reference
World Models
Canonical reference. 88% of citing Pith papers cite this work as background.
270
Pith papers citing it
424
external citations · Pith
Background
88% of classified citations
abstract
We explore building generative neural network models of popular reinforcement learning environments. Our world model can be trained quickly in an unsupervised manner to learn a compressed spatial and temporal representation of the environment. By using features extracted from the world model as inputs to an agent, we can train a very compact and simple policy that can solve the required task. We can even train our agent entirely inside of its own hallucinated dream generated by its world model, and transfer this policy back into the actual environment. An interactive version of this paper is available at https://worldmodels.github.io/
hub tools
citation-role summary
background 36
method 3
other 1
citation-polarity summary
claims ledger
- abstract We explore building generative neural network models of popular reinforcement learning environments. Our world model can be trained quickly in an unsupervised manner to learn a compressed spatial and temporal representation of the environment. By using features extracted from the world model as inputs to an agent, we can train a very compact and simple policy that can solve the required task. We can even train our agent entirely inside of its own hallucinated dream generated by its world model, and transfer this policy back into the actual environment. An interactive version of this paper is
authors
co-cited works
representative citing papers
LeJEPA achieves linear identifiability of latent variables uniquely when the latents are Gaussian in worlds with stationary additive-noise transitions.
Task structure is identifiable across time steps and task-relevant representations are identifiable within steps in a nonparametric setting under sparsity regularization.
AIQI, the first model-free agent proven asymptotically ε-optimal in general reinforcement learning, achieves this by universal induction over action-value distributions under a grain-of-truth condition.
A 5B-parameter latent diffusion model generates real-time four-player Rocket League matches conditioned on all players' actions, staying stable far beyond its training horizon.
SWAM jointly generates intermediate RGB-D sequences and action trajectories from monocular RGB start/goal observations for embodied navigation.
None of ten tested video-generation models reliably remembers objects after occlusion in dynamic scenes; static-camera videos inflate consistency scores.
Equilibrium World Models are a deep-learning solver that enforces exact equilibrium conditions on broad model-generated state distributions to globally solve dynamic stochastic models featuring rare disasters, binding constraints, and counterfactual states.
Hypernetworks distill modular reservoir connectivity via a genomic bottleneck to generate sparse recurrent networks solving difficult temporal tasks with minimal training and maintained robustness.
VLWMs learn variable-length action-conditioned dynamics in latent space with curriculum training, yielding 13% average gains over prior latent world models on long-horizon tasks.
PreAct compiles successful agent executions into verifiable state-machine programs for 8.5-13x faster replay on repeated tasks, with an independent evaluator check before storing each program.
Self-distillation from a caption-conditioned video diffusion model to an image-and-prompt-conditioned executor, enhanced by RL from VLM feedback, enables task solving in world models.
Dream.exe evaluates 8 video generation models on 101 manipulation tasks by converting generated videos into executable robot trajectories in a simulator, finding measurable success rates that visual metrics do not predict.
SVI-Bench provides 35K hours of sports video with 9 tasks across four cognitive levels, revealing models drop from ~74% on action QA to 5% on agentic evidence integration.
YoCausal benchmark shows video diffusion models detect the arrow of time but lack genuine causal understanding relative to humans.
FlatSounds benchmark shows state-of-the-art V2A models rely more on text captions than visual input for physical and semantic accuracy, with captions improving correctness but degrading temporal alignment.
A sleep mechanism with N offline recurrent passes consolidates context into fast weights, improving performance on reasoning tasks where standard transformers fail.
WBench is a benchmark with 289 test cases and 1,058 turns for evaluating interactive world models using 22 automated metrics validated against human judgments.
UWM-JEPA uses a density-matrix latent and unitary predictor in JEPA to preserve joint-state spectrum during blind rollouts, achieving 0.77 accuracy on a five-step hidden-velocity task versus 0.53 for an LSTM baseline.
JEPA guidance steers diffusion models toward low-density regions under an implicit density from a world model, producing minority samples with improved fidelity and semantic validity over generator-centric baselines.
SliceWorld introduces a world-state model for CT report generation that uses predictive and factor-aware objectives on axial slice sequences.
CRONOS benchmark shows recent open-source video generators fail to preserve physical consistency under controlled changes to viewpoint, scene, object category, and appearance.
MemGym unifies agent gyms into a memory benchmark with isolated scoring across tool-use, research, coding, and computer-use regimes plus a lightweight reward model for tractable coding evaluation.
Demo-JEPA enables one-shot cross-embodiment imitation by mapping visual demonstrations to shared latent future trajectories that serve as subgoals for the target agent's own forward dynamics planning.
citing papers explorer
-
Textual Belief States for World Models: Identifiable Representation Learning Under Strict Mediation
Introduces textual belief states and factorized GRPO to enforce strict latent state mediation in text-based world models, yielding preserved prediction accuracy with large gains in representation quality and rollout performance on TextWorld and ScienceWorld.
-
When Does LeJEPA Learn a World Model?
LeJEPA achieves linear identifiability of latent variables uniquely when the latents are Gaussian in worlds with stationary additive-noise transitions.
-
From Generalist to Specialist Representation
Task structure is identifiable across time steps and task-relevant representations are identifiable within steps in a nonparametric setting under sparsity regularization.
-
A Model-Free Universal AI
AIQI, the first model-free agent proven asymptotically ε-optimal in general reinforcement learning, achieves this by universal induction over action-value distributions under a grain-of-truth condition.
-
Multiplayer Interactive World Models with Representation Autoencoders
A 5B-parameter latent diffusion model generates real-time four-player Rocket League matches conditioned on all players' actions, staying stable far beyond its training horizon.
-
Pondering the Way: Spatial-perceiving World Action Model for Embodied Navigation
SWAM jointly generates intermediate RGB-D sequences and action trajectories from monocular RGB start/goal observations for embodied navigation.
-
MemoBench: Benchmarking World Modeling in Dynamically Changing Environments
None of ten tested video-generation models reliably remembers objects after occlusion in dynamic scenes; static-camera videos inflate consistency scores.
-
Equilibrium World Models
Equilibrium World Models are a deep-learning solver that enforces exact equilibrium conditions on broad model-generated state distributions to globally solve dynamic stochastic models featuring rare disasters, binding constraints, and counterfactual states.
-
Distilling a Modular Reservoir Through a Genomic Bottleneck
Hypernetworks distill modular reservoir connectivity via a genomic bottleneck to generate sparse recurrent networks solving difficult temporal tasks with minimal training and maintained robustness.
-
Beyond the Next Step: Variable-Length Latent World Models for Long-Horizon Planning
VLWMs learn variable-length action-conditioned dynamics in latent space with curriculum training, yielding 13% average gains over prior latent world models on long-horizon tasks.
-
PreAct: Computer-Using Agents that Get Faster on Repeated Tasks
PreAct compiles successful agent executions into verifiable state-machine programs for 8.5-13x faster replay on repeated tasks, with an independent evaluator check before storing each program.
-
World Model Self-Distillation: Training World Models to Solve General Tasks
Self-distillation from a caption-conditioned video diffusion model to an image-and-prompt-conditioned executor, enhanced by RL from VLM feedback, enables task solving in world models.
-
Dream.exe: Can Video Generation Models Dream Executable Robot Manipulation?
Dream.exe evaluates 8 video generation models on 101 manipulation tasks by converting generated videos into executable robot trajectories in a simulator, finding measurable success rates that visual metrics do not predict.
-
SVI-Bench: A Dynamic Microworld for Strategic Video Intelligence
SVI-Bench provides 35K hours of sports video with 9 tasks across four cognitive levels, revealing models drop from ~74% on action QA to 5% on agentic evidence integration.
-
YoCausal: How Far is Video Generation from World Model? A Causality Perspective
YoCausal benchmark shows video diffusion models detect the arrow of time but lack genuine causal understanding relative to humans.
-
Benchmarking Single-Factor Physical Video-to-Audio Generation
FlatSounds benchmark shows state-of-the-art V2A models rely more on text captions than visual input for physical and semantic accuracy, with captions improving correctness but degrading temporal alignment.
-
Do Language Models Need Sleep? Offline Recurrence for Improved Online Inference
A sleep mechanism with N offline recurrent passes consolidates context into fast weights, improving performance on reasoning tasks where standard transformers fail.
-
WBench: A Comprehensive Multi-turn Benchmark for Interactive Video World Model Evaluation
WBench is a benchmark with 289 test cases and 1,058 turns for evaluating interactive world models using 22 automated metrics validated against human judgments.
-
UWM-JEPA: Predictive World Models That Imagine in Belief Space
UWM-JEPA uses a density-matrix latent and unitary predictor in JEPA to preserve joint-state spectrum during blind rollouts, achieving 0.77 accuracy on a five-step hidden-velocity task versus 0.53 for an LSTM baseline.
-
Beyond Generative Priors: Minority Sampling with JEPA-Guided Diffusion
JEPA guidance steers diffusion models toward low-density regions under an implicit density from a world model, producing minority samples with improved fidelity and semantic validity over generator-centric baselines.
-
SliceWorld: A Predictive and Controllable World-State Model for CT Report Generation
SliceWorld introduces a world-state model for CT report generation that uses predictive and factor-aware objectives on axial slice sequences.
-
CRONOS: Benchmarking Counterfactual Physical Consistency in Video Models
CRONOS benchmark shows recent open-source video generators fail to preserve physical consistency under controlled changes to viewpoint, scene, object category, and appearance.
-
MemGym: a Long-Horizon Memory Environment for LLM Agents
MemGym unifies agent gyms into a memory benchmark with isolated scoring across tool-use, research, coding, and computer-use regimes plus a lightweight reward model for tractable coding evaluation.
-
Demo-JEPA: Joint-Embedding Predictive Architecture for One-shot Cross-Embodiment Imitation
Demo-JEPA enables one-shot cross-embodiment imitation by mapping visual demonstrations to shared latent future trajectories that serve as subgoals for the target agent's own forward dynamics planning.
-
Baba in Wonderland: Online Self-Supervised Dynamics Discovery for Executable World Models
Alice uses preservation conflicts from failed candidate updates to create class-stratified hypotheses and guide exploration, improving executable world-model learning under prior misalignment.
-
Learning POMDP World Models from Observations with Language-Model Priors
Pinductor leverages language-model priors to learn POMDP world models from limited trajectories, matching privileged-access methods in performance and exceeding tabular baselines in sample efficiency.
-
JEDI: Joint Embedding Diffusion World Model for Online Model-Based Reinforcement Learning
JEDI is the first online end-to-end latent diffusion world model that trains latents from denoising loss rather than reconstruction, achieving competitive Atari100k results with 43% less VRAM and over 3x faster sampling than pixel diffusion baselines.
-
Runtime Monitoring of Perception-Based Autonomous Systems via Embedding Temporal Logic
Embedding Temporal Logic (ETL) performs runtime monitoring directly in learned embedding spaces using distance-based predicates composed with temporal operators, supported by conformal calibration for reliable predicate evaluation.
-
Support-Safe Variational Hybrid Filtering for Contact-Mode and Sparse-Law Recovery
VHYDRO is a support-safe variational hybrid filter that jointly recovers continuous latent states, discrete contact modes, and sparse port-Hamiltonian laws per regime while preventing loss of feasible transitions.
-
The Gordian Knot for VLMs: Diagrammatic Knot Reasoning as a Hard Benchmark
KnotBench benchmark shows state-of-the-art VLMs perform near random on diagrammatic knot reasoning tasks and lack ability to simulate structural moves.
-
ACWM-Phys: Investigating Generalized Physical Interaction in Action-Conditioned Video World Models
ACWM-Phys is a controllable simulator benchmark with in- and out-of-distribution protocols for evaluating action-conditioned world models across rigid, kinematic, deformable, and particle dynamics.
-
SYNCR: A Cross-Video Reasoning Benchmark with Synthetic Grounding
SYNCR benchmark shows leading MLLMs reach only 52.5% average accuracy on cross-video reasoning tasks against an 89.5% human baseline, with major weaknesses in physical and spatial reasoning.
-
Learning Visual Feature-Based World Models via Residual Latent Action
RLA-WM predicts residual latent actions via flow matching to create visual feature world models that outperform prior feature-based and diffusion approaches while enabling offline video-based robot RL.
-
Operator-Guided Invariance Learning for Continuous Reinforcement Learning
VPSD-RL discovers exact and approximate value-preserving Lie-group operators in continuous RL to stabilize learning via transition augmentation and consistency regularization.
-
Render, Don't Decode: Weight-Space World Models with Latent Structural Disentanglement
NOVA represents world states as INR weights for decoder-free rendering, compactness, and unsupervised disentanglement of background, foreground, and motion in video world models.
-
Counterfactual identifiability beyond global monotonicity: non-monotone triangular structural causal models
Non-monotone triangular SCMs with mechanism-wise invertibility and context-independent inverse transport are equivalent to exogenous isomorphism and achieve complete counterfactual identifiability, with supporting experiments on synthetic data and MuJoCo tasks.
-
Learning to Theorize the World from Observation
NEO is a probabilistic neural model that induces compositional programs as a learned Language of Thought from non-textual observations and executes them via a shared transition model to enable explanation-driven generalization.
-
Divide and Conquer: Decoupled Representation Alignment for Multimodal World Models
M²-REPA decouples modality-specific features from diffusion intermediates and aligns them to complementary expert foundation models via a multi-modal alignment loss and modality-specific decoupling regularization for improved multimodal video generation.
-
Latent State Design for World Models under Sufficiency Constraints
World models succeed when their latent states are built to meet task-specific sufficiency constraints rather than preserving the maximum amount of information.
-
Graph World Models: Concepts, Taxonomy, and Future Directions
The paper unifies emerging graph-based world models under a new paradigm and proposes a taxonomy organized by spatial, physical, and logical relational inductive biases.
-
Exploring Spatial Intelligence from a Generative Perspective
Fine-tuning multimodal models on a new synthetic spatial benchmark improves generative spatial compliance on real and synthetic tasks and transfers to better spatial understanding.
-
EgoTL: Egocentric Think-Aloud Chains for Long-Horizon Tasks
EgoTL provides a new egocentric dataset with think-aloud chains and metric labels that benchmarks VLMs on long-horizon tasks and improves their planning, reasoning, and spatial grounding after finetuning.
-
MotionScape: A Large-Scale Real-World Highly Dynamic UAV Video Dataset for World Models
MotionScape is a large-scale UAV video dataset with highly dynamic 6-DoF motions, geometric trajectories, and semantic annotations to train world models that better simulate complex 3D dynamics under large viewpoint changes.
-
SCP: Spatial Causal Prediction in Video
SCP defines a new benchmark task for predicting spatial causal outcomes beyond direct observation and shows that 23 leading models lag far behind humans on it.
-
Joint Embedding Variational Bayes
VJE is a new variational non-contrastive SSL method that models target embeddings with a directional-radial Student-t distribution to enable structured uncertainty estimation directly in the learned representation space.
-
Neural Neural Scaling Laws
NeuNeu, a neural network trained on HuggingFace checkpoints, predicts language model accuracy on 66 downstream tasks at 1.99% MAE by extrapolating trajectories, outperforming logistic scaling laws by 44% and generalizing zero-shot to new models and tasks.
-
Latent Chain-of-Thought World Modeling for End-to-End Driving
LCDrive unifies chain-of-thought reasoning and action selection for end-to-end driving by interleaving action-proposal tokens and latent world-model tokens that predict action outcomes, yielding faster inference and better trajectories than text-based or non-reasoning baselines.
-
Training Agents Inside of Scalable World Models
Dreamer 4 is the first agent to obtain diamonds in Minecraft from only offline data by reinforcement learning inside a scalable world model that accurately predicts game mechanics.
-
Mastering Diverse Domains through World Models
DreamerV3 uses world models and robustness techniques to solve over 150 tasks across domains with a single configuration, including Minecraft diamond collection from scratch.
-
Mastering Atari with Discrete World Models
DreamerV2 reaches human-level performance on 55 Atari games by learning behaviors inside a separately trained discrete-latent world model.