Galbot · robot AI model
GraspVLA
China · 2025
Sources retrieved: 18 November 2026 (A, B, C, D, E)
- Model type
- Vision-language-action (VLA)Source and note: Model typeC arxiv.org/abs/2505.03233 · Retrieved 18 November 2026Note. Paper abstract: “GraspVLA, a VLA model pretrained on large-scale synthetic action data as a foundational model for grasping tasks”.
- What it controls
- Arms and handsSource and note: What it controlsB pku-epic.github.io/GraspVLA-web/ · Retrieved 18 November 2026Note. Project page: “the model first predicts 2D bounding boxes of the target object for both synthetic data and web data, and additionally generates grasp pose and chunked actions for synthetic data.” Code repository: “`--action_steps` — Number of future actions predicted per sample.”
- Inputs
- Camera Language Robot stateSource and note: InputsA github.com/PKU-EPIC/GraspVLA · Retrieved 18 November 2026Note. Code repository, training options: “`--image_keys left,right` — Model camera order.” “`--proprio_steps` — Number of historical/current proprioception steps.” Paper: “GraspVLA supports natural language instructions”.
- Access
- Open weights (download)Source and note: AccessA github.com/PKU-EPIC/GraspVLA · Retrieved 18 November 2026Note. Code repository: "hf download vegebirrd/GraspVLA". The Hugging Face repository metadata mark the checkpoint repository as not gated (""gated":false").
- Base model
- InternLM2 1.8BSource and note: Base modelA github.com/PKU-EPIC/GraspVLA · Retrieved 18 November 2026Note. Code repository: “Serving also needs the InternLM backbone (`internlm/internlm2-1_8b`).”
- Licence type
- Free, research onlySource and note: Licence typeD huggingface.co/api/models/vegebirrd/GraspVLA · Retrieved 18 November 2026Note. Checkpoint licence: “license:cc-by-nc-4.0”.

Specifications
9 of 14 stated 5 not stated
3 fields hidden 2 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
Openness3/3 disclosed
The developer’s own claims1/2 disclosed
Notes
The project page and the paper list Galbot, Peking University, The University of Hong Kong and Beijing Academy of Artificial Intelligence as institutions, with Galbot first. The project page and the code repository are hosted under the PKU-EPIC organisation. The paper abstract says: “GraspVLA, a VLA model pretrained on large-scale synthetic action data as a foundational model for grasping tasks”.
Sources
One line per source, not per figure: a source has one retrieval date.
- Agithub.com/PKU-EPIC/GraspVLARetrieved 18 November 2026
- Bpku-epic.github.io/GraspVLA-web/Main source. Retrieved 18 November 2026
- Carxiv.org/abs/2505.03233Retrieved 18 November 2026
- Dhuggingface.co/api/models/vegebirrd/GraspVLARetrieved 18 November 2026
- Earxiv.org/html/2505.03233Retrieved 18 November 2026





