Rocket - Think it. Type it. Launch it.
Mini toolsDexWM (Dexterous Interaction World Model)
DexWM is a research world model focused on dexterous manipulation, where subtle finger motion and contact dynamics matter. The project represents actions through finger keypoints extracted from egocentric video, allowing it to train on more than 900 hours of human and non-dexterous robot data without needing dense manual action labels. It also adds an auxiliary hand consistency loss so the model preserves accurate hand configuration, improving its ability to predict future states in hand-object interaction scenarios.
New
Multimodal
Gen 3
Overview
DexWM is a dexterous interaction world model for learning hand-object interactions from human videos. It predicts future latent states from past states and finger-level actions, aiming to model fine-grained manipulation more accurately than coarse-action world models.
About Meta Platforms
We're connecting people to what they care about, powering new, meaningful experiences, and advancing the state-of-the-art through open research and accessible tooling.
Industry:
Technology, Information and Media
Company Size:
78.000-79.000
Location:
Menlo Park, California, US
Website:
ai.meta.com
Tools using DexWM (Dexterous Interaction World Model)
No tools found for this model yet.
Last updated: March 27, 2026
