DEPLOYDatabase

Brain

GraspVLA (Galbot)

Galbot's grasping foundation model, GraspVLA (launched Jan 2025): an end-to-end Vision-Language-Action model pre-trained on SynGrasp-1B, a billion-frame synthetic grasping dataset with photorealistic rendering and domain randomization. Enables zero-shot generalization to new grasping tasks without additional training, powering dexterous manipulation on Galbot's robots.

Foundation model · Maturity: Research · Closed · Powers 1 robot

Machine-readable surfaces

Key facts

Model type
Vision-Language-Action model
Maturity stage
research
Open source
no
Training data
SynGrasp-1B, a billion-frame synthetic grasping dataset with photorealistic rendering and domain randomization
Architecture
end-to-end Vision-Language-Action model
SynGrasp-1B synthetic pretrain (primary arXiv)
The GraspVLA arXiv paper (abs/2505.03233) from Galbot/PKU describes a grasping VLA pretrained entirely on SynGrasp-1B, a billion-frame synthetic grasping dataset with about 10,000 objects across 240 categories, photorealistic rendering, and domain randomization, plus Progressive Action Generation co-training with Internet semantics. Treat dataset scale as paper-stated (arXiv)
Zero-shot real-world grasp results (primary arXiv)
The same paper reports real-world zero-shot grasping success around 90%+ across lighting, background, distractor, and height variants on both synthetic-category and web-category object sets, outperforming fine-tuned pi0 and OpenVLA baselines in the authors' tables. Treat rates as paper-stated experimental results (arXiv)
SynGrasp-1B generation pipeline
The project page says SynGrasp-1B is built from more than 10,000 object meshes across 240 Objaverse categories, with stable grasps from BoDex and trajectories planned by CuRobo, then domain-randomized in simulation (GraspVLA).
CoRL 2025 release and labs
The project page lists Galbot, Peking University, the University of Hong Kong, and the Beijing Academy of Artificial Intelligence, and marks the work as a CoRL 2025 paper that will release SynGrasp-1B and pretrained weights (GraspVLA).

Developed by (1)

Powers (1)

Where GraspVLA (Galbot) runs

View the global map

Deployments of robots powered by GraspVLA (Galbot), by place and type.

Free to use on your site, with a link back to DEPLOY.

Mentioned on (1)

Registry pages whose text names GraspVLA (Galbot).

People (1)

He Wang

Common questions

What is GraspVLA (Galbot)?
Galbot's grasping foundation model, GraspVLA (launched Jan 2025): an end-to-end Vision-Language-Action model pre-trained on SynGrasp-1B, a billion-frame synthetic grasping dataset with photorealistic rendering and domain randomization. Enables zero-shot generalization to new grasping tasks without additional training, powering dexterous manipulation on Galbot's robots.
Which robots run on GraspVLA (Galbot)?
1 robot model on the DEPLOY registry runs on GraspVLA (Galbot): Galbot G1.
Who developed GraspVLA (Galbot)?
GraspVLA (Galbot) is credited to Galbot on the DEPLOY registry. Each developer attribution is verified via primary sources.
Is GraspVLA (Galbot) open source?
No. GraspVLA (Galbot) is recorded as proprietary on the DEPLOY registry. Model weights and source are not publicly available.
What type of AI is GraspVLA (Galbot)?
GraspVLA (Galbot) is a foundation model on the DEPLOY registry.
What is GraspVLA (Galbot)'s maturity stage?
GraspVLA (Galbot) is at the research stage on the DEPLOY maturity ladder. Research stage means active development without commercial deployments on file.

Sources (3)

  1. https://arxiv.org/abs/2505.03233
  2. https://roboticsandautomationnews.com/2025/07/03/galbot-raises-151-million-to-scale-embodied-ai-humanoid-robots-partners-with-bosch-investment-arm/92864/
  3. https://pku-epic.github.io/GraspVLA-web
Methodology: Verified · 3 sources (no primary) · last reviewed 2026-10-10

Verification posture

Verified

Low confidence

Review state

Stable

Last reviewed 2026-10-10

Maturity + lifecycle

Maturity stage: research

Sources by quality tier

2
unclassified
Unclassified source
1
preprint
Preprint

The framework is documented at /methodology. Corrections at /corrections. Reviewer: DEPLOY editorial team.

Methodology surface for GraspVLA (Galbot).

Canonical ID 6ab6bd66-6794-4509-bb25-12479104d5a2