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.
hub Canonical reference
Training Agents Inside of Scalable World Models
Canonical reference. 76% of citing Pith papers cite this work as background.
53
Pith papers citing it
Background
76% of classified citations
abstract
World models learn general knowledge from videos and simulate experience for training behaviors in imagination, offering a path towards intelligent agents. However, previous world models have been unable to accurately predict object interactions in complex environments. We introduce Dreamer 4, a scalable agent that learns to solve control tasks by reinforcement learning inside of a fast and accurate world model. In the complex video game Minecraft, the world model accurately predicts object interactions and game mechanics, outperforming previous world models by a large margin. The world model achieves real-time interactive inference on a single GPU through a shortcut forcing objective and an efficient transformer architecture. Moreover, the world model learns general action conditioning from only a small amount of data, allowing it to extract the majority of its knowledge from diverse unlabeled videos. We propose the challenge of obtaining diamonds in Minecraft from only offline data, aligning with practical applications such as robotics where learning from environment interaction can be unsafe and slow. This task requires choosing sequences of over 20,000 mouse and keyboard actions from raw pixels. By learning behaviors in imagination, Dreamer 4 is the first agent to obtain diamonds in Minecraft purely from offline data, without environment interaction. Our work provides a scalable recipe for imagination training, marking a step towards intelligent agents.
hub tools
citation-role summary
background 14
method 3
citation-polarity summary
representative citing papers
SUNTA uses surprise-driven chunk boundaries and decoupled training in hierarchical state-space models to sustain accurate video predictions over 250 timesteps where baselines fail after 10.
World models introduce a stealthy poisoning vector into robot learning pipelines where malicious prompts or dynamics in teleoperated data activate only during synthetic trajectory generation, enabling backdoors in downstream policies.
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.
3D-Belief maintains and updates explicit 3D beliefs about partially observed environments to enable multi-hypothesis imagination and improved performance on embodied tasks.
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.
Reducing visual input to one token per frame in VLA world models maintains or improves long-horizon performance on MetaWorld, LIBERO, and real-robot tasks.
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.
AGWM improves world model accuracy in compositional environments by learning an explicit DAG of action affordance prerequisites to handle dynamic executability.
Dream-Cubed releases a billion-scale voxel dataset and 3D diffusion models that generate controllable Minecraft worlds by operating directly on blocks.
Mask World Model predicts semantic mask dynamics with video diffusion and integrates it with a diffusion policy head, outperforming RGB world models on LIBERO and RLBench while showing better real-world generalization and texture robustness.
An autoregressive diffusion model on sparse point trajectories predicts multi-modal future scene dynamics from single images with orders-of-magnitude faster sampling than dense video simulators while matching accuracy.
DreamDojo is a foundation world model pretrained on the largest human video dataset to date that uses continuous latent actions to transfer interaction knowledge and achieves controllable physics simulation after robot post-training.
Hallucination in world models is a data coverage issue predictable by three signals and preventable through targeted training sampling and online data collection.
Qwen-AgentWorld are language world models that simulate multi-domain agent environments and boost general agent capabilities via decoupled RL simulation and unified foundation model training.
WEAVER is a multi-view world model using flow-matching that jointly satisfies fidelity, consistency, and efficiency for robotic manipulation, yielding 0.87 correlation with real success and policy gains on hardware.
StressDream optimizes initial noise in diffusion video world models using VLM semantic and plausibility objectives to steer generations toward specified high-impact outcomes for improved policy evaluation.
Sensor2Sensor uses 4D Gaussian Splatting to create synthetic training pairs and a diffusion model to convert monocular dashcam videos into high-fidelity multi-modal AV sensor data.
PH-Dreamer integrates a port-Hamiltonian framework into generative world models to enforce physical priors, yielding tighter imagined-real reward alignment and reduced latent space volume on visual control benchmarks.
Latent prediction video models exhibit a distinct robustness profile across corruption, occlusion, fine-grained discrimination, and temporal sensitivity compared to other self-supervised video models when used as world models.
SCAR proposes a joint inverse-forward dynamics framework to learn transferable continuous action representations across embodiments from visual data using regularization and adversarial invariance.
The work derives the optimal ratio of dynamics-to-reward samples that minimizes a bound on return error and characterizes the tradeoff between noisy but cheap rewards versus accurate but expensive ones in imagination-based policy optimization.
Fisher Decorator refines flow policies in offline RL via a local transport map and Fisher-matrix quadratic approximation of the KL constraint, yielding controllable error near the optimum and SOTA benchmark results.
A vision-language-aligned world model turns visuomotor MPC into a language-following planner that reaches 87% success on 288 unseen semantic tasks where standard VLAs drop to 22%.
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.
-
SUNTA: Hierarchical Video Prediction with Surprise-based Chunking
SUNTA uses surprise-driven chunk boundaries and decoupled training in hierarchical state-space models to sustain accurate video predictions over 250 timesteps where baselines fail after 10.
-
Targeting World Models to Compromise Robot Learning Pipelines
World models introduce a stealthy poisoning vector into robot learning pipelines where malicious prompts or dynamics in teleoperated data activate only during synthetic trajectory generation, enabling backdoors in downstream policies.
-
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.
-
3D-Belief: Embodied Belief Inference via Generative 3D World Modeling
3D-Belief maintains and updates explicit 3D beliefs about partially observed environments to enable multi-hypothesis imagination and improved performance on embodied tasks.
-
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.
-
One Token Per Frame: Reconsidering Visual Bandwidth in World Models for VLA Policy
Reducing visual input to one token per frame in VLA world models maintains or improves long-horizon performance on MetaWorld, LIBERO, and real-robot tasks.
-
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.
-
AGWM: Affordance-Grounded World Models for Environments with Compositional Prerequisites
AGWM improves world model accuracy in compositional environments by learning an explicit DAG of action affordance prerequisites to handle dynamic executability.
-
Dream-Cubed: Controllable Generative Modeling in Minecraft by Training on Billions of Cubes
Dream-Cubed releases a billion-scale voxel dataset and 3D diffusion models that generate controllable Minecraft worlds by operating directly on blocks.
-
Mask World Model: Predicting What Matters for Robust Robot Policy Learning
Mask World Model predicts semantic mask dynamics with video diffusion and integrates it with a diffusion policy head, outperforming RGB world models on LIBERO and RLBench while showing better real-world generalization and texture robustness.
-
Envisioning the Future, One Step at a Time
An autoregressive diffusion model on sparse point trajectories predicts multi-modal future scene dynamics from single images with orders-of-magnitude faster sampling than dense video simulators while matching accuracy.
-
DreamDojo: A Generalist Robot World Model from Large-Scale Human Videos
DreamDojo is a foundation world model pretrained on the largest human video dataset to date that uses continuous latent actions to transfer interaction knowledge and achieves controllable physics simulation after robot post-training.
-
Hallucination in World Models is Predictable and Preventable
Hallucination in world models is a data coverage issue predictable by three signals and preventable through targeted training sampling and online data collection.
-
Qwen-AgentWorld: Language World Models for General Agents
Qwen-AgentWorld are language world models that simulate multi-domain agent environments and boost general agent capabilities via decoupled RL simulation and unified foundation model training.
-
WEAVER, Better, Faster, Longer: An Effective World Model for Robotic Manipulation
WEAVER is a multi-view world model using flow-matching that jointly satisfies fidelity, consistency, and efficiency for robotic manipulation, yielding 0.87 correlation with real success and policy gains on hardware.
-
StressDream: Steering Video World Models for Robust Policy Evaluation and Improvement
StressDream optimizes initial noise in diffusion video world models using VLM semantic and plausibility objectives to steer generations toward specified high-impact outcomes for improved policy evaluation.
-
Sensor2Sensor: Cross-Embodiment Sensor Conversion for Autonomous Driving
Sensor2Sensor uses 4D Gaussian Splatting to create synthetic training pairs and a diffusion model to convert monocular dashcam videos into high-fidelity multi-modal AV sensor data.
-
PH-Dreamer: A Physics-Driven World Model via Port-Hamiltonian Generative Dynamics
PH-Dreamer integrates a port-Hamiltonian framework into generative world models to enforce physical priors, yielding tighter imagined-real reward alignment and reduced latent space volume on visual control benchmarks.
-
Latent Video Prediction Learns Better World Models
Latent prediction video models exhibit a distinct robustness profile across corruption, occlusion, fine-grained discrimination, and temporal sensitivity compared to other self-supervised video models when used as world models.
-
SCAR: Self-Supervised Continuous Action Representation Learning
SCAR proposes a joint inverse-forward dynamics framework to learn transferable continuous action representations across embodiments from visual data using regularization and adversarial invariance.
-
On Training in Imagination
The work derives the optimal ratio of dynamics-to-reward samples that minimizes a bound on return error and characterizes the tradeoff between noisy but cheap rewards versus accurate but expensive ones in imagination-based policy optimization.
-
Fisher Decorator: Refining Flow Policy via a Local Transport Map
Fisher Decorator refines flow policies in offline RL via a local transport map and Fisher-matrix quadratic approximation of the KL constraint, yielding controllable error near the optimum and SOTA benchmark results.
-
Grounded World Model for Semantically Generalizable Planning
A vision-language-aligned world model turns visuomotor MPC into a language-following planner that reaches 87% success on 288 unseen semantic tasks where standard VLAs drop to 22%.
-
VAG: Dual-Stream Video-Action Generation for Embodied Data Synthesis
VAG is a synchronized dual-stream flow-matching framework that generates aligned video-action pairs for synthetic embodied data synthesis and policy pretraining.
-
Video Generation Models as World Models: Efficient Paradigms, Architectures and Algorithms
Video generation models can function as world simulators if efficiency gaps in spatiotemporal modeling are bridged via organized paradigms, architectures, and algorithms.
-
Flow Map Language Models: One-step Language Modeling via Continuous Denoising
Continuous flows on token embeddings with flow-map distillation produce one-step language models whose quality exceeds recent 8-step discrete diffusion baselines on LM1B and OpenWebText.
-
World Action Models are Zero-shot Policies
DreamZero uses a 14B video diffusion model as a World Action Model to achieve over 2x better zero-shot generalization on real robots than state-of-the-art VLAs, real-time 7Hz closed-loop control, and cross-embodiment transfer with 10-30 minutes of data.
-
RISE: Self-Improving Robot Policy with Compositional World Model
RISE combines a controllable dynamics model and progress value model into a closed-loop self-improving pipeline that updates robot policies entirely in imagination, reporting over 35% absolute gains on three real-world tasks.
-
Multimodal Reinforcement Learning with Adaptive Verifier for AI Agents
Argos is an agentic verifier that adaptively picks scoring functions to evaluate accuracy, localization, and reasoning quality, enabling stronger multimodal RL training for AI agents.
-
Back to Basics: Let Denoising Generative Models Denoise
Directly predicting clean data with large-patch pixel Transformers enables strong generative performance in diffusion models where noise prediction fails at high dimensions.
-
Ctrl-World: A Controllable Generative World Model for Robot Manipulation
A controllable world model trained on the DROID dataset generates consistent multi-view robot trajectories for over 20 seconds and improves generalist policy success rates by 44.7% via imagined trajectory fine-tuning.
-
Valdi: Value Diffusion World Models
Valdi pairs a latent diffusion dynamics model with end-to-end MPC training and reports that one diffusion step matches an MLP baseline on CarRacing while exposing a multimodality-control trade-off.
-
MilliVid: Hierarchical Latents for Long-Range Consistency in Video Generation
MilliVid compresses video frames into multi-scale token hierarchies and uses coarse-to-fine rollout in a diffusion model to maintain long-range geometric and object consistency on Minecraft videos.
-
What Makes Video World Model Latents Action-Relevant: Prediction over Reconstruction
Temporal video pretraining induces stronger action-relevant structure in video world model latents than pixel reconstruction, as shown by inverse-dynamics probing across encoder families.
-
Representation Learning Enables Scalable Multitask Deep Reinforcement Learning
MR.Q combines predictive auxiliary tasks with high-capacity value functions in a model-free architecture to achieve strong multitask RL performance without planning.
-
Hybrid Neural World Models
Hybrid neural world models train one network with horizon conditioning to predict multi-horizon physical states and extract a per-trajectory error map from forward passes alone for hybrid accuracy-speed operation across PDE and rigid-body domains.
-
Scaling World-Model Reinforcement Learning Through Diffusion Policy Optimization
MBDPO reformulates policy optimization as a diffusion process over searched trajectories in latent world models to reduce misalignment between search and value learning.
-
stable-worldmodel: A Platform for Reproducible World Modeling Research and Evaluation
The paper presents stable-worldmodel (swm), a platform with high-performance data layer, modern world model baselines, planning solvers, and extended environments for reproducible research and generalization evaluation.
-
Nano World Models: A Minimalist Implementation of Future Video Prediction
Nano World Models supplies a unified minimalist codebase and evaluation framework for studying diffusion forcing in video prediction across control, games, and robot domains.
-
ASH: Agents that Self-Hone via Embodied Learning
ASH learns long-horizon embodied policies from unlabeled internet video via a self-improvement loop that trains an IDM on its own trajectories and extracts supervision plus key-moment memory from video.
-
Nautilus: From One Prompt to Plug-and-Play Robot Learning
NAUTILUS is a prompt-driven harness that automates plug-and-play adapters, typed contracts, and validation for policies, benchmarks, and robots in learning research.
-
Probing the Impact of Scale on Data-Efficient, Generalist Transformer World Models for Atari
Transformer world models on Atari exhibit game-specific scaling regimes, but joint training on 26 environments produces consistent monotonic gains that improve downstream control policies to a median normalized score of 0.770.
-
Event-Driven Reinforcement Learning Enables Long-Horizon Control in Semiconductor Fabrication
Event-driven RL framework for semiconductor manufacturing control shows throughput and utilization gains in high-fidelity simulations under offline and online training.
-
Towards World Models in Biomedical Research
Proposes biomedical world models that learn latent states and intervention-conditioned dynamics to enable simulation of future biological trajectories for discovery in virtual cells, organoids, patients, and surgery.
-
Position: Good Embodied Reward Models Need Bad Behavior Data
Embodied reward models systematically over-reward unsafe, suboptimal, and shortcut robot behaviors due to training on successful data only, and modest inclusion of bad behavior data improves alignment with human preferences.
-
Behavior-Invariant Task Representation Learning with Transformer-based World Models for Offline Meta-Reinforcement Learning
The work introduces behavior-invariant latent task representations via information-theoretic learning in a Transformer world model plus conservative penalties on imagined rollouts to improve generalization in offline meta-RL.
-
World Action Models: The Next Frontier in Embodied AI
The paper introduces World Action Models as a new paradigm unifying predictive world modeling with action generation in embodied foundation models and provides a taxonomy of existing approaches.
-
Matrix-Game 3.0: Real-Time and Streaming Interactive World Model with Long-Horizon Memory
Matrix-Game 3.0 delivers 720p real-time video generation at 40 FPS with minute-scale memory consistency by combining residual self-correction training, camera-aware memory injection, and DMD-based autoregressive distillation on a 5B model.
-
World Action Models: A Survey
A survey that clarifies boundaries and organizes World Action Models by generation requirements and predictive substrates, identifying a trend toward generating less of the future.