World Action Model Atlas

Interactive atlas and generated architecture diagrams for world action models.

How can worldmodels best serveas VLA backbones? How can imaginedfutures becomeaction signals? How should worldand action learningstay coupled? How do we keepforesight fastenough for control? How do unlabeledvideos becomeexecutable skills? How can modelsstay robust underphysical contact? Can inversedynamics inferactions frompredicted pixels?(1) Can inversedynamics inferactions frompredicted pixels?(2) Can inversedynamics inferactions frompredicted pixels?(3) Can latent futurestate replacefull visualrollout? Can predictivelatents exposethe action thatcaused change? Can…

saved by

• Dhruv Sheth

related reading

• What Matters for Latent Actions in Robot Learningcarldegio.github.io
• Pretrained to Imagine, Fine-Tuned to Act: The Rise of World-Action Models | NVIDIA Technical Blogdeveloper.nvidia.com
• [2602.10098] VLA-JEPA: Enhancing Vision-Language-Action Model with Latent World Modelarxiv.org
• Can Video World Models Track Unobserved World States?joonghyuk.com
• Moritz Reuss — Robotics & VLA Researchmbreuss.github.io
• State of Vision-Language-Action (VLA) Research at ICLR 2026 – Moritz Reussmbreuss.github.io
• Video models are zero-shot learners and reasonersarxiv.org
• World Models | Rohit Bandarurohitbandaru.github.io
• [2603.16666] Fast-WAM: Do World Action Models Need Test-time Future Imagination?arxiv.org
• pdfopenreview.net
• Explore | alphaXivalphaxiv.org
• Emergence of Human to Robot Transfer in Vision-Language-Action Modelspi.website