RESEARCH PAPERYear 2025

AdaWorld: Learning Adaptable World Models with Latent Actions

Shenyuan Gao; Siyuan Zhou; Yilun Du; Jun Zhang; Chuang Gan

Classification

Major category: Foundational work
Quadrant: Not applicable
Architecture: Not applicable
Prediction paradigm: Not applicable
Subcategories: Latent action pretraining Neural world simulators
Source review status: Verified from primary sources

Category review. AdaWorld supplies a pre-2026 generic latent-action pretraining/adaptation route: infer reusable action codes from frame transitions, condition a separate forward video model on them, and adapt to downstream controls. This underlies the manuscript latent-action thread rather than one task-specific WAM policy. Reading evidence

AT A GLANCE

Contribution

A contribution summary has not been added yet.

Abstract

An abstract has not been added yet.

Affiliations

Not listed in the collection.

BibTeX

@inproceedings{adaworld,
  author       = {Shenyuan Gao and
                  Siyuan Zhou and
                  Yilun Du and
                  Jun Zhang and
                  Chuang Gan},
  title        = {AdaWorld: Learning Adaptable World Models with Latent Actions},
  booktitle    = {ICML},
  year         = {2025},
 
}

Classification

Contribution

Abstract

Affiliations

BibTeX

More in this category

Critique of World Model

Precise and Dexterous Robotic Manipulation via Human-in-the-Loop Reinforcement Learning

Mean Flows for One-step Generative Modeling