N_0-TWAM: Scaling Tactile-Native World-Action Model for Contact-Rich Manipulation

Researchers released N₀-TWAM, a tactile-native world-action model for contact-rich manipulation.

The model predicts future touch and vision together to derive actions, using visuo-tactile pretraining at scale.

It was trained on tactile-rich demonstrations across 6 embodiments and 450 tasks.

Code and pretrained checkpoints are now available, making the system easier to adopt and build on.

The work could advance tactile sensing, vision-language-action systems, and world models for robotics.