Berkeley and NVIDIA's AW-0 couples code, video, and action streams for robot manipulation
Zhuoyang Liu · hf · 2026-10-22
UC Berkeley, NVIDIA, and UPenn introduce AW-0, an agentic world action model framework that combines code-based planning with joint video and action prediction.
- Idea: Existing world action models lack compositional skill reuse, while agentic robotics lacks predictive grounding in physical dynamics. AW-0 models code, future video, and robot actions as three coupled generative streams in a unified mixture-of-transformers architecture.
- Dual role of code: A code expert generates parameterized skill routines and control logic, both structuring multi-step prediction inside the world model and representing skills as reusable programs for agentic execution.
- Data pipeline: An automated curation pipeline labels robot and human video trajectories with aligned programs and consolidates recurring routines into reusable skill libraries.
- Results: Outperforms VLA and WAM baselines on RoboTwin 2.0 and physical bimanual tasks, with gains in language following, semantic reasoning, and robustness.
More from Embodied
- HKU-led team launches Embodied Playground, 410-task benchmark for model physical intelligence — YiMaTweets · 2026-10-22
- AgenticRoboRL lifts a 9B robot planner from 18% to 71% on LIBERO-PRO — UCF-Ai-LLMs · 2026-10-22
- Yao Ming takes the stage to talk robotics at a tech event — micoolcho · 2026-10-22
- Matt Wolfe demos 'personal bodyguard robot' standing in a junkyard — Matt Wolfe · 2026-10-22
- Unitree founder says he's betting on recursive self-improvement — micoolcho · 2026-10-22
- Dyna Robotics' humanoid Taku shown in person: 'it works! It's real!' — chris_j_paxton · 2026-10-22