AW-0 is a joint robotics research model from UC Berkeley, NVIDIA and the University of Pennsylvania. Its Mixture-of-Transformers jointly predicts reusable skill code, future video and robot actions. The team evaluated it in RoboTwin 2.0 and on physical bimanual manipulation tasks.

Overview

Model type
World-action model
Architecture
WAM
Action generation
Diffusion
Listed versions
AW-0

Papers & reports (1)

Agentic World Action Models

Paper available

arXiv: 2610.30136DOI: 10.48550/arxiv.2610.30136
Authors & affiliations (18)

Author order: Original order

  1. Zhuoyang Liu

    University of California, Berkeley

  2. Dantong Niu

    University of California, Berkeley · NVIDIA

  3. Botong Zhang

    University of Pennsylvania

  4. Boning Shao

    University of California, Berkeley

  5. Jing Wang

    NVIDIA

  6. Ruijie Zheng

    NVIDIA

  7. Yuvan Sharma

    NVIDIA

  8. Runyu Lu

    NVIDIA

  9. Jiaming Liu

    NVIDIA

First published
2026-10-22
Latest paper revision
v1 · 2026-10-22
Venue
CoRL 2026 Workshop on Agentic Robotics

Covered model versions

Paper ↗PDF ↗

Official resources

Submit a source or correction