Autoregressive VLA based on Qwen2.5-VL with residual VQ action tokens for bimanual manipulation.
Access and availability
- Model files
- Public access
Model files
| File | Access | Weights license |
|---|---|---|
| RDT2-VQ | Access | Weights license |
Release information
SourceVerified on · 2026-10-30
At a glance
- Model type
- Vision-language-action model
- Backbone
- Qwen2.5-VL-7B-Instruct
- Weight parameters
- 8.2922 B
- Input
- Language instruction / RGB images
- Output
- 24-step, 20-dimensional relative bimanual action chunks
- Released
- September 2025
Technical details
Functional purpose
Action and control
Before you use this version: Available weights or code do not establish plug-and-play robot compatibility. Check the required configuration, data and interfaces in the source documentation.
Submit a source or correction