mlsanityData to 7 Oct 2026

Fast-WAM: Do World Action Models Need Test-time Future Imagination?

arXiv 2603.16666 ↗First posted 17 Mar 2026cs.CV cs.AIcited by 311

On the map: World models → Embodied and robot world models

also in 1 more
Abstract (arXiv)

World Action Models (WAMs) have emerged as a promising alternative to Vision-Language-Action (VLA) models for embodied control because they explicitly model how visual observations may evolve under action. Most existing WAMs follow an imagine-then-execute paradigm, incurring substantial test-time latency from iterative video denoising, yet it remains unclear whether explicit future imagination is actually necessary for strong action performance. In this paper, we ask whether WAMs need explicit future imagination at test time, or whether their benefit comes primarily from video modeling during training. We disentangle the role of video modeling during training from explicit future generation during inference by proposing Fast-WAM, a WAM architecture that retains video co-training during training but skips future prediction at test time. We further instantiate several Fast-WAM variants to enable a controlled comparison of these two factors. Across these variants, we find that Fast-WAM remains competitive with imagine-then-execute variants, while removing video co-training causes a much larger performance drop. Empirically, Fast-WAM achieves competitive results with state-of-the-art methods both on simulation benchmarks (LIBERO and RoboTwin) and real-world tasks, without embodied pretraining. It runs in real time with 190ms latency, over 4× faster than existing imagine-then-execute WAMs. These results suggest that the main value of video prediction in WAMs may lie in improving world representations during training rather than generating future observations at test time. Project page: https://yuantianyuan01.github.io/FastWAM/

Key facts. Cited by 311 papers in the map (data through 7 Oct 2026). Placed in Embodied and robot world models, one of 11 directions in World models. Builds on 31 of its 32 references that are in the map. First posted on arXiv 17 Mar 2026.

Builds on

31 of 32 references.

Closest ancestors · 7

Same direction, newest first

  1. 2026Causal World Modeling for Robot Control29 Jan
  2. 2025Act2Goal: From World Model To General Goal-conditioned Policy29 Dec
  3. 2025RynnVLA-002: A Unified Vision-Language-Action and World Model21 Nov
  4. 2025Dual-Stream Diffusion for World-Model Augmented Vision-Language-Action Model31 Oct
  5. 2025WorldVLA: Towards Autoregressive Action World Model26 Jun
  6. 2025Unified World Models: Coupling Video and Action Diffusion for Pretraining on Large Robotic Datasets3 Apr
  7. 2024RoboDreamer: Learning Compositional World Models for Robot Imagination18 Apr

Other references · 24

  1. 2023RT-2: Vision-Language-Action Models Transfer Web Knowledge to Robotic Controlcited by 2,993
  2. 2024OpenVLA: An Open-Source Vision-Language-Action Modelcited by 2,777
  3. 2024π0: A Vision-Language-Action Flow Model for General Robot Controlcited by 2,297
  4. 2025π0.5: a Vision-Language-Action Model with Open-World Generalizationcited by 1,653
  5. 2023LIBERO: Benchmarking Knowledge Transfer for Lifelong Robot Learningcited by 1,338
  6. 2025GR00T N1: An Open Foundation Model for Generalist Humanoid Robotscited by 1,167
  7. 2024RDT-1B: a Diffusion Foundation Model for Bimanual Manipulationcited by 670
  8. 2025SmolVLA: A Vision-Language-Action Model for Affordable and Efficient Roboticscited by 507
  9. 2023Learning Universal Policies via Text-Guided Video Generationcited by 506
  10. 2025RoboTwin 2.0: A Scalable Data Generator and Benchmark with Strong Domain Randomization for Robust Bimanual Robotic Manipulationcited by 477
  11. 2025CoT-VLA: Visual Chain-of-Thought Reasoning for Vision-Language-Action Modelscited by 440
  12. 2025AgiBot World Colosseo: A Large-scale Manipulation Platform for Scalable and Intelligent Embodied Systemscited by 390
12 more
  1. 2023Unleashing Large-Scale Video Generative Pre-training for Visual Robot Manipulationcited by 353
  2. 2024GR-2: A Generative Video-Language-Action Model with Web-Scale Knowledge for Robot Manipulationcited by 290
  3. 2024Video Prediction Policy: A Generalist Robot Policy with Predictive Visual Representationscited by 279
  4. 2026Cosmos Policy: Fine-Tuning Video Models for Visuomotor Control and Planningcited by 274
  5. 2025DreamVLA: A Vision-Language-Action Model Dreamed with Comprehensive World Knowledgecited by 203
  6. 2025Unified Video Action Modelcited by 195
  7. 2025DexVLA: Vision-Language Model with Plug-In Diffusion Expert for General Robot Controlcited by 191
  8. 2024Gen2Act: Human Video Generation in Novel Scenarios enables Generalizable Robot Manipulationcited by 142
  9. 2025mimic-video: Video-Action Models for Generalizable Robot Control Beyond VLAscited by 133
  10. 2025Genie Envisioner: A Unified World Foundation Platform for Robotic Manipulationcited by 129
  11. 2025Video Generators are Robot Policiescited by 86
  12. 2025Vidar: Embodied Video Diffusion Model for Generalist Manipulationcited by 58

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2026: 311.

311 papers.

Sort:

By area:

  1. 2026π0.7: a Steerable Generalist Robotic Foundation Model with Emergent Capabilitiescited by 127
  2. 2026StarVLA: A Lego-like Codebase for Vision-Language-Action Model Developingcited by 103
  3. 2026Being-H0.7: A Latent World-Action Model from Egocentric Videoscited by 68
  4. 2026Unified 4D World Action Modeling from Video Priors with Asynchronous Denoisingcited by 65
  5. 2026Motubrain: An Advanced World Action Model for Robot Controlcited by 54
  6. 2026World Action Models: The Next Frontier in Embodied AIcited by 46
  7. 2026LaWAM: Latent World Action Models for Efficient Dynamics-Aware Robot Policiescited by 46
  8. 2026World Model for Robot Learning: A Comprehensive Surveycited by 42
  9. 2026AHA-WAM:Asynchronous Horizon-Adaptive World-Action Modeling with Observation-Guided Context Routingcited by 41
  10. 2026DSpark: Confidence-Scheduled Speculative Decoding with Semi-Autoregressive Generationcited by 36
  11. 2026Do World Action Models Generalize Better than VLAs? A Robustness Studycited by 35
  12. 2026RoboDojo: A Unified Sim-and-Real Benchmark for Comprehensive Evaluation of Generalist Robot Manipulation Policiescited by 35
38 more
  1. 2026ImageWAM: Do World Action Models Really Need Video Generation, or Just Image Editing?cited by 33
  2. 2026Native Video-Action Pretraining for Generalizable Robot Controlcited by 30
  3. 2026ABot-M0.5: Unified Mobility-and-Manipulation World Action Modelcited by 28
  4. 2026OA-WAM: Object-Addressable World Action Model for Robust Robot Manipulationcited by 25
  5. 2026Xiaomi-Robotics-1: Scaling Vision-Language-Action Models with over 100K Hours of Real-World Trajectoriescited by 24
  6. 2026τ0-WM: A Unified Video-Action World Model for Robotic Manipulationcited by 21
  7. 2026VGGRPO: Towards World-Consistent Video Generation with 4D Latent Rewardcited by 19
  8. 2026Light-WAM: Efficient World Action Models with State-Fusion Action Decodingcited by 19
  9. 2026When to Trust Imagination: Adaptive Action Execution for World Action Modelscited by 18
  10. 2026Efficient-WAM: A 1B-Parameter World-Action Model with Low-Cost Future Imaginationcited by 17
  11. 2026EgoWAM: World Action Models Beyond Pixels with In-the-Wild Egocentric Human Datacited by 17
  12. 2026G0.5: One Autoregressive Stream for Robot Reasoning and Actioncited by 17
  13. 2026Flash-WAM: Modality-Aware Distillation for World Action Modelscited by 16
  14. 2026MaskWAM: Unifying Mask Prompting and Prediction for World-Action Modelscited by 15
  15. 2026WALL-WM: Carving World Action Modeling at the Event Jointscited by 14
  16. 2026Next Forcing: Causal World Modeling with Multi-Chunk Predictioncited by 14
  17. 2026Learning 4D Geometric Priors for Inference-Efficient World Action Modelscited by 14
  18. 2026Faster-WAM: Efficient Inference-Time Future Conditioning for Robust World Action Modelscited by 14
  19. 2026World Action Verifier: Self-Improving World Models via Forward-Inverse Asymmetrycited by 13
  20. 2026RoboTTT: Context Scaling for Robot Policiescited by 13
  21. 2026AIM: Intent-Aware Unified world action Modeling with Spatial Value Mapscited by 12
  22. 2026DriveWAM: Video Generative Priors Enable Scalable World-Action Modeling for Autonomous Drivingcited by 12
  23. 2026WAM4D: Fast 4D World Action Model via Spatial Register Tokenscited by 12
  24. 2026MemoryWAM: Efficient World Action Modeling with Persistent Memorycited by 12
  25. 2026Is the Future Compatible? Diagnosing Dynamic Consistency in World Action Modelscited by 10
  26. 2026HarmoWAM: Harmonizing Generalizable and Precise Manipulation via Adaptive World Action Modelscited by 10
  27. 2026World-Language-Action Model for Unified World Modeling, Language Reasoning, and Action Synthesiscited by 10
  28. 2026Metis: A Generalizable and Efficient World-Action Model for Autonomous Driving and Urban Navigationcited by 10
  29. 2026Geometric Action Model for Robot Policy Learningcited by 10
  30. 2026World Action Models: A Surveycited by 10
  31. 2026VT-WAM: Visual-Tactile World Action Model for Contact-Rich Manipulationcited by 10
  32. 2026GigaWorld-1: A Roadmap to Build World Models for Robot Policy Evaluationcited by 10
  33. 2026FlowWAM: Optical Flow as a Unified Action Representation for World Action Modelscited by 10
  34. 2026GigaWorld-Policy-0.5: A Faster and Stronger WAM Empowered by AutoResearchcited by 10
  35. 2026ST-WAM: Semantic-Temporal World Action Model for Robust Manipulation under Visual Distribution Shiftscited by 10
  36. 2026Feedback World Model Enables Precise Guidance of Diffusion Policycited by 9
  37. 2026MemoryVLA++: Temporal Modeling via Memory and Imagination in Vision-Language-Action Modelscited by 9
  38. 2026RepWAM: World Action Modeling with Representation Visual-Action Tokenizerscited by 9

50 most cited of 311 All 311 (JSON)

Similar, not cited

Close in wording, no citation either way

  1. 2026OpenWAM: An Open, Modular Exploration Towards Systematic World-Action Model Pretraining
  2. 2026RealtimeWAM: How Fast Can I Run My World Action Model?
  3. 2026FLEX-WAM: Flexible Block-Causal World-Action Models for Long-Horizon Imagination and Planning
  4. 2026RealtimeWAM: One-Step Asynchronous World Action Models
  5. 2026World Action Models are Zero-shot Policies
  6. 2026OpenWAM: An Open Framework for Composable World-Action Models
  7. 2026WAM-OPD: On-Policy Distillation for World Action Models
  8. 2026World Action Modeling with Progressive Visual Planning