mlsanityData to 7 Oct 2026

Embodied and robot world models

575 papers18 this weekHigh confidence in this grouping: the papers form one clear topicHigh confidence in this grouping: the papers form one clear topic · About

Keywords: action, manipulation, robot, visual, video

Key facts. Embodied and robot world models: 575 papers on arXiv since 2014 (data through 7 Oct 2026). 187 of its 575 mapped papers (33%) cite “Diffusion Policy: Visuomotor Policy Learning via Action Diffusion” (2023, arXiv 2303.04137). 163 of its 575 mapped papers (28%) cite “World Action Models are Zero-shot Policies” (2026, arXiv 2602.15922). 114 new papers 17 Sep–7 Oct.

What’s new

17 Sep – 7 Oct 2026

114 new papers, 17 Sep – 7 Oct; 18 this week.

  1. This week · 18
  2. Cross-Embodiment Robot Foundation World Models with Latent Actions
  3. Long-WAM: Scaling the Context of World-Action Models
  4. RoboJEPA: Scaling Robotic Latent World Models also fits: Latent predictive world models
  5. Video Prediction Policy 2: Predict Better, Act Better
  6. RealtimeWAM: How Fast Can I Run My World Action Model?
  7. DeltaWAM: Distilling Action Tangent Fields into World Action Models
  8. Controllable Crowd Generation through World-Model Planning also fits: World models for autonomous driving
  9. Event-Aligned Visual Action Reasoning for World Action Models
  10. Predicted Futures Are Not Enough: Learning Executable Goals for Robot Manipulation
  11. DepthWorld: 3D World Model for Robot Manipulation
All 114
  1. AutodidactWAM: Cross-Modal Self-Distillation from Generated Video to Robot Actions
  2. OpenWAM: An Open Framework for Composable World-Action Models
  3. TAPDreamer: Transferable Adversarial Patches for World Action Models
  4. H-JEPA: End-to-End Learning of Hierarchical World Models for Visual Planning also fits: Latent predictive world models
  5. RealtimeWAM: One-Step Asynchronous World Action Models also fits: Video-generation world models and simulators
  6. SimForcing: Distilling Simulation Motion Priors into Real-Domain Robot World Models
  7. KineWorld: Action-Induced Transport Fields for Embodied World Modeling
  8. Future Anchored Verification and Online Recovery for World Action Models
  9. Earlier · 96
  10. FLEX-WAM: Flexible Block-Causal World-Action Models for Long-Horizon Imagination and Planning also fits: Latent predictive world models
  11. R2-WAM: Repair-and-Reject Post-Training for World Action Models
  12. ACG-WAM: World-Action Modeling via Action-Conditioned Geometric Latent Prediction
  13. Flow Policies as Actions of Skill-Level World Models: Learned and Symbolic Abstractions for Long-Horizon Planning also fits: Latent dynamics models and model-based RL
  14. RoboIRS: Inference-Time Internal Representation Steering for Generalist Robot Policies
  15. ForeAct3D: Policy-Grounded Future World Modeling for VLA Policies
  16. DreamFormer: Dream Imitation with a Transformer World Model for Language-Conditioned Robotic Manipulation
  17. SUAVE: Unified Video-Action Models via Masked Diffusion
  18. XGenAct: Geometry-Enhanced World Action Models through Cross-Task Generation
  19. Native Action-Prior Learning from Videos for World Action Models
  20. PointWAM: 3D World Action Modeling for Dexterous Robotic Manipulation
  21. WAMJET: A Harness for World Action Model Acceleration
  22. World Action Modeling with Progressive Visual Planning
  23. Rethinking World-Action Model for Compositional and In-Context Robotic Manipulation
  24. SkeleWAM: Skeleton World-Action Modeling for Efficient Robotic Manipulation
  25. UniWAM: Unified World-Action Model
  26. ActiveWAM: Evidence-Aware Active Vision for World-Action Models
  27. Completion Aware Guidance for World Action Models also fits: World models for navigation
  28. PhysicsLENS: Diagnosing Physical Property Blindness in Video Generation Models also fits: Latent predictive world models
  29. FutureWorlds: Learning Robotic World Models from Alternative Futures also fits: Latent dynamics models and model-based RL
  30. CtrlWAM: Controllable World Action Models with Aligned Intent and Foresight
  31. One from Infinity: Actualizing Futures from Pretrained World Models into Robot Actions
  32. Devol-ONE: One Autoregressive Mixture of Transformers to Unify Vision-Language-Action and Latent World Modeling also fits: Video-generation world models and simulators
  33. TacDyn-WAM: Learning Implicit Tactile Dynamics in a Heterogeneous Visuo-Tactile World Action Model
  34. Token-World: World Modeling in Vision-Language Model Token Space for Robot Manipulation
  35. Learning Skills from Historical Action Trajectories: Action Experience Dictionary for World Action Models
  36. EWAM: Emergent Depth-Wise Specialization in a Unified Embodied Model -- From Semantic Understanding through Visual Foresight to Action also fits: World models for navigation
  37. SplineWAM: Adaptive Action Horizons for World Action Models via B-Spline Representations
  38. Magic-W0: A Structured World-Action Foundation Model for Physical Intelligence
  39. RoboCoach: World Models as Active Coaches for Compositional Robot Skills
  40. UniWAM Technical Report: Unified Mobile Manipulation via Mixed-Stream World-Action Modeling and Manipulation Anchor Pose Supervision
  41. MotionWeave: Learning Motion-Centered Future Dynamics for Vision-Language-Action Policies
  42. The Planning Limits of Latent World Models also fits: Latent predictive world models
  43. Sparse-WAM: Accelerating World Action Models via Action-Guided Sparse Imagination
  44. AnyStep-WAM: Budget-Aligned Distillation and Adaptive Inference for World Action Models
  45. Rethinking Representations for World-Action Modeling
  46. EVO-WAM: Evolving World Action Models through Video-Action Verification
  47. Explore, Execute, Evolve: A Skill Acquisition and Reuse Loop for Embodied Agents also fits: World models for navigation
  48. MVG-WAM: Multiple View Geometry-Aware World-Action Modeling for Robotic Manipulation
  49. CogWAM: Aligning Semantic Cognition with World Action Modeling via Event-Driven Interfaces
  50. Direct Experience World-Model Optimization: Learning the World Beyond Action Imitation
  51. V-JEPA Policy: Building Effective World-Action Models on Predictive Visual Latents
  52. Staircase Policy: Streaming Inference for World-Action Models with Large Action Chunks
  53. Revision, Not Restart: Revisable Visual Plans for Closed-Loop World-Action Models
  54. RoGSW4RLD: Feed-Forward 4D Gaussian Lifting for Robot World Model Rollouts also fits: Video-generation world models and simulators
  55. Spatial Grafting: Grounding 3D Features for Flow-Matching Robot Policies
  56. From World Models to World Action Models: Rethinking Next-State Prediction
  57. FutureDuet: Decoupling Observation Access from Future Supervision in World Action Models
  58. Dexterous Tactile World Model
  59. WAM-OPD: Sharpening World Action Models via On-Policy Distillation
  60. WorldGuide: Learning Success-Failure Boundaries in Latent World Models for Vision-Language-Action Policies also fits: Latent dynamics models and model-based RL
  61. Achieve What You Imagined: Learning to Align Actions with Visual Plans
  62. VIDEAS: Distilling Explicit Action Semantics from Demonstration Videos for World Models via Prior-Guided Simulation also fits: Latent predictive world models
  63. Q-WAM: 4-Bit Quantization of World Action Models with Action-Subspace Protection
  64. Scope-WM: Scoped Computation for Efficient Visual World Models also fits: Latent predictive world models
  65. DeltaWAM: Change-Centric Visual Foresight via Delta Tokens for an Efficient World-Action Model
  66. Dynamic Manipulation with World-Action Models via Counterfactual Planning
  67. VPTwin: Real-Sim-Real Video Prediction for Robotic Manipulation Planning
  68. Copper-Policy: Focus on the Representation for Robust Robot Manipulation
  69. InternW0-Δ: A World Action Model Bridging Predictive Dynamics and Actions with 20K+ Hours of Open Data
  70. DyMD: Preserving Interaction Dynamics through Distribution Matching Distillation in Few-Step Video World Models also fits: Video-generation world models and simulators
  71. Towards VLA-Dreamer: Refining VLA Behavior Using World Models
  72. Action Forcing: Training World Models on Unsupervised Video by Recovering Underlying Egomotion Bases
  73. Rolling-WAM: World Action Models with Rolling Imagination
  74. Representation World Model: Learning States, Transition and Executable Plans in Representation also fits: Latent predictive world models
  75. Streaming-WAM: Action-Conditioned World-Action Model for Asynchronous Robot Manipulation
  76. DeltaWAM: Delta World Action Models for Bimanual Manipulation
  77. PointCast: One World Model for Rigid, Articulated, and Deformable Object Manipulation
  78. Generalizable Robotic Insertion with World Models
  79. InternW0: A Foundational Physical World Model for Efficient Real-World Interactions
  80. Latent evolving World Action Model also fits: Video-generation world models and simulators
  81. CoRe-WAM: Correspondence-Aligned Temporal Residuals for World Action Models
  82. TriWorldBench: A Tri-View Consistency Perspective on Embodied World Models
  83. An Action Is Worth One Patch: Unified World-Action Modeling with PatchWAM
  84. IndustrialVLA-Bench: A Traceable Multi-Axis Evaluation of Open Robot Policy Models
  85. DexTacWAM: A Visuo-Tactile World-Action Model for Dexterous Manipulation
  86. DualWAM: Dual-System World Action Models for Asynchronous Global Planning and Local Refinement
  87. Think Like a World Model, Act Like a VLA: Distilling World-Model Representations into Compact Robot Policies also fits: World models for navigation
  88. What Matters in Designing World Action Models: An Empirical Study
  89. Imagine-RL: Residual-Confidence-Guided Cross-Attention for World-Model-Augmented VLA Reinforcement Learning
  90. Grounded Action Model: 3D Grounding as a Foundation for Robotics
  91. AR-WAM: A Visual-Conditioned Agent-Ready World Action Model for Robotic Manipulation
  92. The Right Future for Action: Learning Action-Relevant Predictive States in World Action Models
  93. Robot World Models Are Not Invariant to How the Actions Are Written also fits: Latent dynamics models and model-based RL
  94. FIRM-WM: State-factorized factual-interventional recurrent modeling for reward-free visual planning also fits: Latent predictive world models
  95. Benchmarking World Models for Continual Learning on Compositional Tasks also fits: World models for language agents
  96. SkelWAM: A Skeleton-Guided World-Action Model for Zero-Shot Cross-Embodiment Manipulation
  97. Skel-WAM: A Hand-Skeleton-Conditioned World Action Model for Human-to-Robot Manipulation Transfer
  98. ME-Dex 1.0: Bringing Heterogeneous Tactile Sensing into World Action Modeling
  99. FOCAL-VLA: Subtask-Guided Geometry Distillation and Implicit World Modeling for Vision-Language-Action Models
  100. WM-VS: Progress-Aligned World Models for Closed-Loop Visual Servoing
  101. Agile-WAM: An Agile Tactile World Action Model for Contact-Rich Robot Control
  102. MoWAM: Explicit Future Motion Prediction for Efficient World Action Models
  103. DexTouch-WM: Learning Action-Conditioned Tactile World Models from Human Touch for Dexterous Robot Manipulation
  104. TacSushi: Tactile-Grounded World-Action Modeling for Dexterous Sushi Manipulation
  105. WorldContact: A Contact-Centric World Model for Scalable Robot Learning

Foundations

Most cited by papers here

  1. Diffusion Policy: Visuomotor Policy Learning via Action Diffusion2023cited by 187 here
  2. World Action Models are Zero-shot Policies2026cited by 163 here
  3. Causal World Modeling for Robot Control2026cited by 137 here
  4. Fast-WAM: Do World Action Models Need Test-time Future Imagination?2026cited by 135 here
  5. Motus: A Unified Latent Action World Model2025cited by 119 here

Lineage

The papers this direction builds on, oldest first.

Each paper on the line cites the one before.

  1. 2018World Modelscited by 113 here
  2. 2018Learning Latent Dynamics for Planning from Pixelscited by 94 here
  3. 2019Dream to Control: Learning Behaviors by Latent Imaginationcited by 148 here
  4. 2020Mastering Atari with Discrete World Modelscited by 102 here
  5. 2023Mastering Diverse Domains through World Modelscited by 128 here
  6. 2023Learning Interactive Real-World Simulatorscited by 94 here
  7. 2024Genie: Generative Interactive Environmentscited by 145 here
  8. 2025Cosmos World Foundation Model Platform for Physical AIcited by 141 here
  9. 2025Unified World Models: Coupling Video and Action Diffusion for Pretraining on Large Robotic Datasetscited by 117 here
  10. 2025Motus: A Unified Latent Action World Modelcited by 119 here
  11. 2026Causal World Modeling for Robot Controlcited by 137 here
  12. 2026World Action Models are Zero-shot Policiescited by 163 here
12 more it builds on, not in this chain
  1. 2021Learning Transferable Visual Models From Natural Language Supervisioncited by 78 here
  2. 2022DayDreamer: World Models for Physical Robot Learningcited by 81 here
  3. 2022Flow Matching for Generative Modelingcited by 105 here
  4. 2022Scalable Diffusion Models with Transformerscited by 102 here
  5. 2023Diffusion Policy: Visuomotor Policy Learning via Action Diffusioncited by 187 here
  6. 2024RoboDreamer: Learning Compositional World Models for Robot Imaginationcited by 79 here
  7. 2024DINO-WM: World Models on Pre-trained Visual Features enable Zero-shot Planningcited by 80 here
  8. 2025Wan: Open and Advanced Large-Scale Video Generative Modelscited by 182 here
  9. 2025V-JEPA 2: Self-Supervised Video Models Enable Understanding, Prediction and Planningcited by 105 here
  10. 2025WorldVLA: Towards Autoregressive Action World Modelcited by 94 here
  11. 2025Ctrl-World: A Controllable Generative World Model for Robot Manipulationcited by 82 here
  12. 2026Fast-WAM: Do World Action Models Need Test-time Future Imagination?cited by 135 here

The 24 most cited of 786.

Browse

30 of 575, ranked 6–35

  1. 6Unified World Models: Coupling Video and Action Diffusion for Pretraining on Large Robotic Datasets2025117
  2. 7V-JEPA 2: Self-Supervised Video Models Enable Understanding, Prediction and Planning2025105
  3. 8WorldVLA: Towards Autoregressive Action World Model202594
  4. 9Ctrl-World: A Controllable Generative World Model for Robot Manipulation202582
  5. 10RoboDreamer: Learning Compositional World Models for Robot Imagination202479
  6. 11DreamGen: Unlocking Generalization in Robot Learning through Video World Models202567
  7. 12IRASim: A Fine-Grained World Model for Robot Manipulation202467
  8. 13GigaWorld-Policy: An Efficient Action-Centered World--Action Model202660
  9. 14iVideoGPT: Interactive VideoGPTs are Scalable World Models202456
  10. 15DreamDojo: A Generalist Robot World Model from Large-Scale Human Videos202654
All 30
  1. 16FLARE: Robot Learning with Implicit World Modeling202549
  2. 17AdaWorld: Learning Adaptable World Models with Latent Actions202545
  3. 18Zero-Shot Robotic Manipulation with Pretrained Image-Editing Diffusion Models202342
  4. 19TesserAct: Learning 4D Embodied World Models202538
  5. 20WorldGym: World Model as An Environment for Policy Evaluation202538
  6. 21VLA-JEPA: Enhancing Vision-Language-Action Model with Latent World Model202636
  7. 22WorldEval: World Model as Real-World Robot Policies Evaluator202535
  8. 23Motubrain: An Advanced World Action Model for Robot Control202633
  9. 243D-VLA: A 3D Vision-Language-Action Generative World Model202432
  10. 25WMPO: World Model-based Policy Optimization for Vision-Language-Action Models202531
  11. 26BC-Z: Zero-Shot Task Generalization with Robotic Imitation Learning202231
  12. 27EnerVerse: Envisioning Embodied Future Space for Robotics Manipulation202530
  13. 28Structured World Models from Human Videos202330
  14. 29PointWorld: Scaling 3D World Models for In-The-Wild Robotic Manipulation202629
  15. 30Being-H0.7: A Latent World-Action Model from Egocentric Videos202628
  16. 31GigaWorld-0: World Models as Data Engine to Empower Embodied AI202527
  17. 32Unified 4D World Action Modeling from Video Priors with Asynchronous Denoising202626
  18. 33RynnVLA-002: A Unified Vision-Language-Action and World Model202526
  19. 34World Action Models: The Next Frontier in Embodied AI202623
  20. 35Masked World Models for Visual Control202222