Zhejiang University · robot AI model
Atomic-WAM
China · 2026
Sources retrieved: 19 November 2026 (A, B, C)
- Model type
- World action modelSource and note: Model typeA arxiv.org/abs/2610.20157 · Retrieved 19 November 2026Note. “To this end, we introduce Atomic-WAM, an understand-then-execute framework that connects task reasoning to action generation through atomic-motion descriptions.”
- What it controls
- Arms and handsSource and note: What it controlsA arxiv.org/abs/2610.20157 · Retrieved 19 November 2026Note. “Specifically, a vision-language reasoner describes the next local motion from the task instruction and visual context, guiding a WAM to predict future video and an action chunk.” Real-world evaluation: “We train on DROID and deploy on our Franka setup without further fine-tuning.”
- Inputs
- Camera Language Robot stateSource and note: InputsA arxiv.org/abs/2610.20157 · Retrieved 19 November 2026Note. “Specifically, a vision-language reasoner describes the next local motion from the task instruction and visual context, guiding a WAM to predict future video and an action chunk.” The WAM also receives current observation “and robot state”.
- Access
- Open weights (download)Source and note: AccessB github.com/ssj9596/Atomic-WAM · Retrieved 19 November 2026Note. “Paper, inference code, and model weights released.”
- Base model
- Cosmos3-NanoSource and note: Base modelC arxiv.org/html/2610.20157v1 · Retrieved 19 November 2026Note. Built as an “understand-then-execute framework with Cosmos3-Nano”.
- Licence document
- OpenMDW-1.1Source and note: Licence documentB github.com/ssj9596/Atomic-WAM · Retrieved 19 November 2026Note. Cosmos files are covered by OpenMDW-1.1; RoboLab files are covered by Apache-2.0.

Specifications
7 of 14 stated 7 not stated
4 fields hidden 3 not-stated fields with a note stay ·
How to read the figuresShow guideClose
- not statedNo source
- 0Reported as zero
- ASource
Select a source marker for source details and notes.
Release1/1 disclosed
Input and output3/4 disclosed
Size and speed0/2 disclosed
Where it runs and what it is built on1/2 disclosed
Openness2/3 disclosed
The developer’s own claims0/2 disclosed
Notes
Developed by researchers from Zhejiang University, University of Hong Kong, ACE Robotics, and Jiangnan University; first-author work was completed during an internship at ACE Robotics.
Described by the authors as “To this end, we introduce Atomic-WAM, an understand-then-execute framework that connects task reasoning to action generation through atomic-motion descriptions.”
WAM weights, Reasoner LoRA, and inference code are publicly released.
Sources
One line per source, not per figure: a source has one retrieval date.
- Aarxiv.org/abs/2610.20157Main source. Retrieved 19 November 2026
- Bgithub.com/ssj9596/Atomic-WAMRetrieved 19 November 2026
- Carxiv.org/html/2610.20157v1Retrieved 19 November 2026



