Compare

CUHK / StarVLA · robot AI model

VLAct

China · 2026

Sources retrieved: 19 November 2026 (A, B)

Model type
Vision-language-action (VLA)⁠
What it controls
Arms and hands⁠
Inputs
Camera Language Robot state⁠
Access
Open weights (download)⁠
Base model
Qwen3-VL-4B-Instruct-Action⁠
Licence type
Free, commercial use⁠

Two white Franka robot arms with grippers at a black table holding a yellow banana, a metal bowl and an egg

Specifications

10 of 14 stated 4 not stated

How to read the figuresShow guideClose
  • not statedNo source
  • yesStated by a source
  • 0Reported as zero
  • ASource

Select a source marker for source details and notes.

Release1/1 disclosed

Released27 August 2026

Input and output3/4 disclosed

Model typeVision-language-action (VLA)
InputsCamera Language Robot state
What it controlsArms and hands
Degrees of freedom controlledNot stated

Size and speed0/2 disclosed

ParametersNot stated
Control rateNot stated

Where it runs and what it is built on1/2 disclosed

On robot or cloudNot stated
Base modelQwen3-VL-4B-Instruct-Action

Openness3/3 disclosed

Licence documentApache-2.0
Licence typeFree, commercial use

The developer’s own claims2/2 disclosed

Learns new tasksNeeds retraining
Works across robotsYes

Notes

Developers describe VLAct: “We propose VLAct, a VLA-oriented VLM backbone trained on broad, heterogeneous, multi-embodiment robot data before task-specific fine-tuning.”

Sources

One line per source, not per figure: a source has one retrieval date.

  1. Ahuggingface.co/StarVLA/VLAct_Qwen3_PretrainMain source. Retrieved 19 November 2026
  2. Barxiv.org/abs/2608.27550Retrieved 19 November 2026

Wrong or missing data? Tell us

More models

Same model type