Fast-WAM
$ 0
Fast-WAM is an open world action model from Tsinghua University and Galaxea AI. It learns from video during training but skips imagining the future when acting, so it reacts faster.
Available on backorder
Brain Score
2
Specifications and details:
| Nationality | China |
|---|---|
| Website | https://iiis.tsinghua.edu.cn |
| Model type | Foundation Model |
| Manufacturer | Tsinghua University, IIIS |
| Release date | March 17, 2026 |
Description
Fast-WAM is a research robot model from Tsinghua University’s Institute for Interdisciplinary Information Sciences and the startup Galaxea AI. It belongs to a family called world action models, which learn by predicting future video. Most of these models imagine the future frame by frame before every move, so they react slowly. Fast-WAM asks whether robots truly need that step. Its answer is no: video prediction helps most during training, not while the robot acts. As a result, the model keeps the benefits of video learning but responds much faster.
In practice, Fast-WAM picks its next actions in about 190 milliseconds. That is more than four times faster than similar models that imagine first. It also matched those models on the LIBERO and RoboTwin simulation tests, even without large robot pre-training. On a real Galaxea R1 Lite robot, it folded towels, a hard task that needs long, careful handling. In addition, the team shared its code on GitHub and released trained checkpoints on Hugging Face.
New! 2026 Humanoid
Robot Market Report
198 pages of exclusive insight from global robotics experts — uncover funding trends, technology challenges, leading manufacturers, supply chain shifts, and surveys and forecasts on future humanoid applications.
now Google DeepMind
Contact Humanoid.guide
