AffectWorld introduces three progressive evaluation tasks and a counterfactual test that expose current world models' inability to treat emotion as a causal variable in future prediction.
super hub Mixed citations
Mastering Diverse Domains through World Models
Mixed citation behavior. Most common role is background (66%).
550
Pith papers citing it
14
external citations · Crossref
Background
66% 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 81
method 6
baseline 5
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
CivWorldBench evaluates language models on forecasting, hidden-state inference, and intervention prediction in a multi-agent game environment, finding that models are competent trend predictors but struggle with causal reasoning.
A decision-centered taxonomy that categorizes world models by whether they preserve state (Plausible), respond correctly to interventions (Controllable), or measurably improve behavior (Actionable).
A family of deadline-aware congestion metrics (Effective Congestion) and a demonstration-driven training protocol (MGA-RL) together reduce online interaction costs of deep RL network controllers by 7x while matching or exceeding reliability of from-scratch training.
Planning-time intervention fidelity is a distinct property of world models that is not captured by reward fit or task-anchored training; it can fail severely even when current-state capture and feature alignment pass.
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 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.
EvolvingBrain enables multimodal agents to self-evolve by extracting and reusing hierarchical skills from their own trajectories, outperforming direct agents on two open-world benchmarks.
COPE adds a multi-step action forecasting loss on successful trajectories to GRPO, improving aggregate performance and sample efficiency for language agents across five benchmarks.
CGWM grounds interpretable concepts from images using fuzzy temporal logic and plans with symbolic transitions, outperforming neural baselines in robustness and data efficiency.
Sparse episodic memory with native-feature addressing and native replay recovers long-context retrieval in pretrained linear attention models without additional training.
Explanatory World Models use LLM-generated explanations to redistribute reward to causally relevant earlier events, improving credit assignment in long-horizon RL and recommendation systems.
Scaling egocentric video world models improves modeling of the agent but leaves object-interaction fidelity far lower, with data and model size offering no practical route to closure.
The 4-tensor attention model predicts next semantic states by jointly attending over spatial and temporal positions using separate semantic and temporal-context fibers, achieving lower perplexity than a standard transformer on ROCStories.
A benchmark that scores LLM agents by speedrunning ability shows frontier agents approach human records only on short platformer levels and remain far behind on longer games, while offering a saturation-resistant evaluation signal.
Training world models on paired same-state rollouts, rather than on observational next-frame prediction, isolates the causal effect of an action, locates where it enters and propagates, and lowers effect prediction error from pixels.
LEMCA automatically synthesizes mode-switching control architectures via LLM-guided evolutionary search, consistently outperforming monolithic designs in resource efficiency across diverse robotic benchmarks.
CAST replaces the action-value critic with a state-value critic trained on alternating planner-and-policy transitions, achieving better sample efficiency in model-based RL.
Counterfactual Latent World Models (CLWM) augment standard world models with a contrastive objective that separates action-conditioned latent futures by their downstream consequences, not their visual similarity, improving planning under partial observability.
REFACTOR-VLA learns a library of typed motor programs via wake-sleep optimization, using a behavioral equivalence kernel from a latent world model, and reports improved clustering on LIBERO benchmarks.
Large-scale egocentric video pre-training (120K hours) improves zero-shot robot manipulation success from 36.1% to 77.8% via a three-stage training pipeline and a slow-fast dual-system architecture.
World-model and policy training in LLM agents share input-feature subspaces but write to orthogonal output subspaces, and sequential training induces redundant input pathways.
A sequential hypothesis test is introduced that adaptively consumes real test data and Monte Carlo sign-flip rounds to determine whether a given synthetic dataset improves real-world model performance, with statistical guarantees.
Cross-shadow prediction on appearance-resampled video pairs yields a unified latent dynamics interface that transfers demonstrated actions across environments better than prior latent-action and interactive world models.
citing papers explorer
-
AffectWorld: Benchmarking Emotion-Driven Future World Prediction
AffectWorld introduces three progressive evaluation tasks and a counterfactual test that expose current world models' inability to treat emotion as a causal variable in future prediction.
-
CivWorldBench: Evaluating Language World Models in Structured Multi-Agent Sandbox Worlds
CivWorldBench evaluates language models on forecasting, hidden-state inference, and intervention prediction in a multi-agent game environment, finding that models are competent trend predictors but struggle with causal reasoning.
-
World Models for Embodied Intelligence: From Plausible to Controllable to Actionable
A decision-centered taxonomy that categorizes world models by whether they preserve state (Plausible), respond correctly to interventions (Controllable), or measurably improve behavior (Actionable).
-
From Prior-Guided Heuristics to Deployable Agents: Accelerating Demonstration-Driven Reinforcement Learning for Deadline-Constrained Network Control
A family of deadline-aware congestion metrics (Effective Congestion) and a demonstration-driven training protocol (MGA-RL) together reduce online interaction costs of deep RL network controllers by 7x while matching or exceeding reliability of from-scratch training.
-
The Intervention Gap in Latent World Models
Planning-time intervention fidelity is a distinct property of world models that is not captured by reward fit or task-anchored training; it can fail severely even when current-state capture and feature alignment pass.
-
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.
-
EvolvingBrain: An Adaptive Skill System for Multi-modal Agents Self-Evolution in Open Worlds
EvolvingBrain enables multimodal agents to self-evolve by extracting and reusing hierarchical skills from their own trajectories, outperforming direct agents on two open-world benchmarks.
-
Co-Evolving Acting and Forecasting for Language Agents
COPE adds a multi-step action forecasting loss on successful trajectories to GRPO, improving aggregate performance and sample efficiency for language agents across five benchmarks.
-
Neurosymbolic Learning of Interpretable World Models
CGWM grounds interpretable concepts from images using fuzzy temporal logic and plans with symbolic transitions, outperforming neural baselines in robustness and data efficiency.
-
History Replays Itself: Unleashing Long-Context Capability in Linear Attention
Sparse episodic memory with native-feature addressing and native replay recovers long-context retrieval in pretrained linear attention models without additional training.
-
Beyond Temporal Proximity: Explanatory World Models for Credit Assignment in Reinforcement Learning
Explanatory World Models use LLM-generated explanations to redistribute reward to causally relevant earlier events, improving credit assignment in long-horizon RL and recommendation systems.
-
What 30,000 Hours of Ego-centric Video Does Not Teach
Scaling egocentric video world models improves modeling of the agent but leaves object-interaction fidelity far lower, with data and model size offering no practical route to closure.
-
4-Tensor Attention Model for Semantic Physical Reality
The 4-tensor attention model predicts next semantic states by jointly attending over spatial and temporal positions using separate semantic and temporal-context fibers, achieving lower perplexity than a standard transformer on ROCStories.
-
SpeedrunBench: Challenging LLM Agents with Video Game Speedrunning
A benchmark that scores LLM agents by speedrunning ability shows frontier agents approach human records only on short platformer levels and remain far behind on longer games, while offering a saturation-resistant evaluation signal.
-
Do-JEPA: From Masking to Intervention in Latent World Models
Training world models on paired same-state rollouts, rather than on observational next-frame prediction, isolates the causal effect of an action, locates where it enters and propagates, and lowers effect prediction error from pixels.
-
LEMCA: LLM-Guided Synthesis of Efficient Mode-Switching Control Architectures
LEMCA automatically synthesizes mode-switching control architectures via LLM-guided evolutionary search, consistently outperforming monolithic designs in resource efficiency across diverse robotic benchmarks.
-
CAST: Alternating State-Value Targets and Expanded Policy Gradients for Model-Based Reinforcement Learning
CAST replaces the action-value critic with a state-value critic trained on alternating planner-and-policy transitions, achieving better sample efficiency in model-based RL.
-
Learning Counterfactual World Models for Embodied Reasoning under Partial Observability
Counterfactual Latent World Models (CLWM) augment standard world models with a contrastive objective that separates action-conditioned latent futures by their downstream consequences, not their visual similarity, improving planning under partial observability.
-
REFACTOR-VLA: Unsupervised Library Learning of Typed Motor Programs
REFACTOR-VLA learns a library of typed motor programs via wake-sleep optimization, using a behavioral equivalence kernel from a latent world model, and reports improved clustering on LIBERO benchmarks.
-
ZimaBlue: Evolving Generalizable World Action Models through Scalable Video Pre-training
Large-scale egocentric video pre-training (120K hours) improves zero-shot robot manipulation success from 36.1% to 77.8% via a three-stage training pipeline and a slow-fast dual-system architecture.
-
How do World Models and Policies Compose in LLM Agents? A Joint Spectral and Behavioral Account
World-model and policy training in LLM agents share input-feature subspaces but write to orthogonal output subspaces, and sequential training induces redundant input pathways.
-
Should I Use This Synthetic Dataset for Training? How to Test with Minimal Real Data
A sequential hypothesis test is introduced that adaptively consumes real test data and Monte Carlo sign-flip rounds to determine whether a given synthetic dataset improves real-world model performance, with statistical guarantees.
-
ShadowDancer: Teaching Video World Models Any Action by Learning Unified Dynamics Representations from a Video and Its Shadow
Cross-shadow prediction on appearance-resampled video pairs yields a unified latent dynamics interface that transfers demonstrated actions across environments better than prior latent-action and interactive world models.
-
The World Model Remembers, the Actor Forgets: Dream Rehearsal for Continual Model-Based RL
Under replay, DreamerV3's world model remembers old tasks but its actor forgets; supervised self-imitation on graded dreams retains behavior where RL-in-imagination fails.
-
Deform360: A Massive Multi-view Visuotactile Dataset for Deformable World Models
Deform360 supplies 215+ hours of synchronized multi-view video and tactile data plus markerless 3D tracks, revealing that 3D particle models win in low data while 2D video models generalize better at scale.
-
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.
-
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.