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
Mastering Diverse Domains through World Models
Canonical reference. 79% of citing Pith papers cite this work as background.
233
Pith papers citing it
7
external citations · Pith
Background
79% of classified citations
abstract
Developing a general algorithm that learns to solve tasks across a wide range of applications has been a fundamental challenge in artificial intelligence. Although current reinforcement learning algorithms can be readily applied to tasks similar to what they have been developed for, configuring them for new application domains requires significant human expertise and experimentation. We present DreamerV3, a general algorithm that outperforms specialized methods across over 150 diverse tasks, with a single configuration. Dreamer learns a model of the environment and improves its behavior by imagining future scenarios. Robustness techniques based on normalization, balancing, and transformations enable stable learning across domains. Applied out of the box, Dreamer is the first algorithm to collect diamonds in Minecraft from scratch without human data or curricula. This achievement has been posed as a significant challenge in artificial intelligence that requires exploring farsighted strategies from pixels and sparse rewards in an open world. Our work allows solving challenging control problems without extensive experimentation, making reinforcement learning broadly applicable.
hub tools
citation-role summary
background 31
baseline 4
method 4
citation-polarity summary
claims ledger
- abstract Developing a general algorithm that learns to solve tasks across a wide range of applications has been a fundamental challenge in artificial intelligence. Although current reinforcement learning algorithms can be readily applied to tasks similar to what they have been developed for, configuring them for new application domains requires significant human expertise and experimentation. We present DreamerV3, a general algorithm that outperforms specialized methods across over 150 diverse tasks, with a single configuration. Dreamer learns a model of the environment and improves its behavior by ima
authors
co-cited works
representative citing papers
Promptbreeder evolves both task prompts and the mutation prompts that improve them using LLMs, outperforming Chain-of-Thought and Plan-and-Solve on arithmetic and commonsense reasoning benchmarks.
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.
ScratchWorld benchmark finds that language models achieve at most 13.8% value-aware changed-field F1 on replay-verified Scratch state transitions and frequently ignore executable rules.
None of ten tested video-generation models reliably remembers objects after occlusion in dynamic scenes; static-camera videos inflate consistency scores.
REGEN uses recurrent generative replays from World Action Models to cut catastrophic forgetting by up to 50% in continual imitation learning compared to sequential fine-tuning.
Prepending a few self-generated random interaction clips as context lets VLA policies identify novel camera viewpoints and morphologies at test time and outperform multi-view baselines without parameter updates.
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.
SWAAP is the first two-stage poisoning framework that identifies a harmful target world model via bilevel optimization and realizes it through stealth-constrained gradient matching on a limited fraction of fine-tuning transitions.
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.
M* introduces the Walk Graph abstraction to serve arbitrary compositions of multimodal model components and reports latency and throughput gains over vLLM-Omni and other baselines on text-to-image, text-to-speech, and robotic planning workloads.
MCPS adapts a trajectory generator from autonomous driving to simulate counterfactual 3D pass outcomes in football and produces distribution-aware execution-surplus scores from value model rollouts.
MBench is a new benchmark that quantifies long-term memory in video world models via three hierarchical consistency dimensions evaluated on curated real videos.
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.
MiraBench defines action-conditioned reliability via three levels (physics adherence, action-following fidelity, optimism bias detection) and applies it to 12 model configurations using a 16,000-judgment human corpus, finding visual fidelity a poor proxy for action fidelity, no reliable scale benefi
What-If World is a new paired-prompt benchmark showing that nine state-of-the-art video generation models achieve at most 52% on causal intervention tests and cluster near 28% for open-source systems.
TPS-Drive uses an agent-centric tokenizer supervised by a frozen 3D detection head to purify VLM spatial representations, enabling better scene forecasting and lower collision rates on nuScenes and NAVSIM benchmarks.
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.
Formalizes video world models as group actions on states and uses latent regularization with synthesized supervision to enforce consistency, introducing GAC and GAR metrics that improve structural correctness in SOTA models.
SliceWorld introduces a world-state model for CT report generation that uses predictive and factor-aware objectives on axial slice sequences.
WMAttack automates finite-budget attack search for world-model agents via SCAS and RGAR, reporting higher normalized reward drops than baselines on Atari and DMC tasks.
AffectVerse improves multimodal emotion recognition by at least 2.57% on nine benchmarks through an Emotion World Module that performs short-horizon latent affective prediction via cross-modal temporal imagination and belief aggregation.
Aero-World adapts a pretrained latent diffusion transformer for action-conditioned aerial video generation by injecting inertial action tokens and using a frozen latent-space Physics Probe for inertial consistency supervision during LoRA finetuning, with a new AeroBench benchmark showing improved AA
Alice uses preservation conflicts from failed candidate updates to create class-stratified hypotheses and guide exploration, improving executable world-model learning under prior misalignment.
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.
-
Promptbreeder: Self-Referential Self-Improvement Via Prompt Evolution
Promptbreeder evolves both task prompts and the mutation prompts that improve them using LLMs, outperforming Chain-of-Thought and Plan-and-Solve on arithmetic and commonsense reasoning benchmarks.
-
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.
-
ScratchWorld: Evaluating If World Models Compute Executable Consequences
ScratchWorld benchmark finds that language models achieve at most 13.8% value-aware changed-field F1 on replay-verified Scratch state transitions and frequently ignore executable rules.
-
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.
-
World Action Models Enable Continual Imitation Learning with Recurrent Generative Replays
REGEN uses recurrent generative replays from World Action Models to cut catastrophic forgetting by up to 50% in continual imitation learning compared to sequential fine-tuning.
-
In-Context World Modeling for Robotic Control
Prepending a few self-generated random interaction clips as context lets VLA policies identify novel camera viewpoints and morphologies at test time and outperform multi-view baselines without parameter updates.
-
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.
-
Stealthy World Model Manipulation via Data Poisoning
SWAAP is the first two-stage poisoning framework that identifies a harmful target world model via bilevel optimization and realizes it through stealth-constrained gradient matching on a limited fraction of fine-tuning transitions.
-
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.
-
M*: A Modular, Extensible, Serving System for Multimodal Models
M* introduces the Walk Graph abstraction to serve arbitrary compositions of multimodal model components and reports latency and throughput gains over vLLM-Omni and other baselines on text-to-image, text-to-speech, and robotic planning workloads.
-
Monte Carlo Pass Search: Using Trajectory Generation for 3D Counterfactual Pass Evaluation in Football
MCPS adapts a trajectory generator from autonomous driving to simulate counterfactual 3D pass outcomes in football and produces distribution-aware execution-surplus scores from value model rollouts.
-
MBench: A Comprehensive Benchmark on Memory Capability for Video World Models
MBench is a new benchmark that quantifies long-term memory in video world models via three hierarchical consistency dimensions evaluated on curated real videos.
-
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.
-
MiraBench: Evaluating Action-Conditioned Reliability in Robotic World Models
MiraBench defines action-conditioned reliability via three levels (physics adherence, action-following fidelity, optimism bias detection) and applies it to 12 model configurations using a 16,000-judgment human corpus, finding visual fidelity a poor proxy for action fidelity, no reliable scale benefi
-
What-If World: A Causal Benchmark for General World Models in Embodied Scenarios
What-If World is a new paired-prompt benchmark showing that nine state-of-the-art video generation models achieve at most 52% on causal intervention tests and cluster near 28% for open-source systems.
-
TPS-Drive: Task-Guided Representation Purification for VLM-based Autonomous Driving
TPS-Drive uses an agent-centric tokenizer supervised by a frozen 3D detection head to purify VLM spatial representations, enabling better scene forecasting and lower collision rates on nuScenes and NAVSIM benchmarks.
-
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.
-
World Models as Group Actions
Formalizes video world models as group actions on states and uses latent regularization with synthesized supervision to enforce consistency, introducing GAC and GAR metrics that improve structural correctness in SOTA models.
-
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.
-
WMAttack: Automated Attack Search for Adversarial Evaluation of World-Model Agents
WMAttack automates finite-budget attack search for world-model agents via SCAS and RGAR, reporting higher normalized reward drops than baselines on Atari and DMC tasks.
-
AffectVerse: Emotional World Models for Multimodal Affective Computing
AffectVerse improves multimodal emotion recognition by at least 2.57% on nine benchmarks through an Emotion World Module that performs short-horizon latent affective prediction via cross-modal temporal imagination and belief aggregation.
-
Aero-World: Action-Conditioned Aerial Video Generation from Inertial Controls
Aero-World adapts a pretrained latent diffusion transformer for action-conditioned aerial video generation by injecting inertial action tokens and using a frozen latent-space Physics Probe for inertial consistency supervision during LoRA finetuning, with a new AeroBench benchmark showing improved AA
-
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.
-
Deterministic Event-Graph Substrates as World Models for Counterfactual Reasoning
Event-graph substrates represent states as RDF triple logs, prove a duality reducing explanatory and counterfactual queries to causal-ancestor traversal, and outperform symbolic and parametric baselines on CLEVRER and a new Smallville benchmark.
-
WorldVLN: Autoregressive World Action Model for Aerial Vision-Language Navigation
WorldVLN proposes the first autoregressive world action model for aerial vision-language navigation that predicts short-horizon latent world states, decodes them to waypoints in closed loop, and uses two-stage training with Action-aware GRPO to achieve over 12% success-rate gains on benchmarks plus零
-
WorldParticle: Unified World Simulation of Lagrangian Particle Dynamics via Transformer
A transformer with prediction-correction and hierarchical super-token merging unifies simulation of six physical dynamics categories on Lagrangian particles and generalizes to unseen conditions.
-
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.
-
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.
-
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.
-
OA-WAM: Object-Addressable World Action Model for Robust Robot Manipulation
OA-WAM uses persistent address vectors and dynamic content vectors in object slots to enable addressable world-action prediction, improving robustness on manipulation benchmarks under scene changes.
-
The Predictive-Causal Gap: An Impossibility Theorem and Large-Scale Neural Evidence
Predictive representation learning structurally favors encoding slower or less noisy environment modes over causal system modes, as shown by an impossibility theorem for linear-Gaussian dynamics and large-scale neural experiments.
-
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.
-
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.
-
Privileged Foresight Distillation: Zero-Cost Future Correction for World Action Models
Privileged Foresight Distillation distills the residual difference in action predictions with versus without future context into a current-only adapter, yielding consistent gains on LIBERO and RoboTwin benchmarks.
-
Agent-Centric Observation Adaptation for Robust Visual Control under Dynamic Perturbations
ACO-MoE recovers 95.3% of clean-input performance in visual control tasks under Markov-switching corruptions by routing restoration experts and anchoring representations to clean foreground masks.
-
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.
-
3D-Anchored Lookahead Planning for Persistent Robotic Scene Memory via World-Model-Based MCTS
3D-ALP achieves 0.65 success on memory-dependent 5-step robotic reach tasks versus near-zero for reactive baselines by anchoring MCTS planning to a persistent 3D camera-to-world frame.
-
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.
-
Advantage-Guided Diffusion for Model-Based Reinforcement Learning
Advantage-guided diffusion (SAG and EAG) steers sampling in diffusion world models to higher-advantage trajectories, enabling policy improvement and better sample efficiency on MuJoCo tasks.
-
MoRight: Motion Control Done Right
MoRight disentangles object and camera motion via canonical-view specification and temporal cross-view attention, while decomposing motion into active user-driven and passive consequence components to learn and apply causality in video generation.
-
PlayWorld: Learning Robot World Models from Autonomous Play
PlayWorld learns high-fidelity robot world models from unsupervised self-play, producing physically consistent video predictions that outperform models trained on human data and enabling 65% better real-world policy performance via model-based RL.
-
PhysMem: Scaling Test-Time Memory for Embodied Physical Reasoning
PhysMem enables VLM-based robot planners to learn and verify physical properties through test-time interaction and hypothesis testing, raising success on a brick insertion task from 23% to 76%.
-
ReCogDrive: A Reinforced Cognitive Framework for End-to-End Autonomous Driving
ReCogDrive unifies VLM scene understanding with a diffusion planner reinforced by DiffGRPO to reach state-of-the-art results on NAVSIM and Bench2Drive benchmarks.
-
BiTrajDiff: Bidirectional Trajectory Generation with Diffusion Models for Offline Reinforcement Learning
BiTrajDiff augments offline RL datasets by running independent forward and backward diffusion processes from intermediate states, yielding higher performance than prior one-directional data-augmentation baselines on D4RL.
-
Zero-Shot Robotic Manipulation with Pretrained Image-Editing Diffusion Models
SuSIE uses a finetuned InstructPix2Pix diffusion model to propose subgoal images that guide a low-level goal-conditioned policy, achieving SOTA zero-shot performance on CALVIN and real-world manipulation.
-
Learning Interactive Real-World Simulators
UniSim learns a universal real-world simulator from orchestrated diverse datasets, enabling zero-shot deployment of policies trained purely in simulation.
-
Language Agent Tree Search Unifies Reasoning Acting and Planning in Language Models
LATS integrates Monte Carlo Tree Search with language models using in-context learning, value functions, and self-reflection to achieve 92.7% pass@1 on HumanEval and competitive web navigation performance.
-
Voyager: An Open-Ended Embodied Agent with Large Language Models
Voyager achieves superior lifelong learning in Minecraft by combining an automatic exploration curriculum, a library of executable skills, and iterative LLM prompting with environment feedback, yielding 3.3x more unique items and 15.3x faster milestone unlocks than prior methods while generalizing技能
-
SUNTA: Hierarchical Video Prediction with Surprise-based Chunking
Decoupled surprise-based hierarchical state-space models with top-down inconsistency chunking sustain accurate 250-step open-loop video prediction where prior HSSMs collapse within 10 steps.