Preview

Apr 14, 2026

ProducerGoogle DeepMind

Hosted UIedgeCloud

Gemini Robotics On-Device

Robotics FMVLA

VLA model by Google DeepMind optimized to run locally on robotic devices with low-latency inference.

Technical specification

Context window

1088 tokens

Tools

No

Fine-tuning

Yes

Weights access

Closed

Last updated: May 2, 2026

Modalities

Input

Text

Image

action

Output

action

Capabilities

2

Image understanding★

Vision

Multimodal understanding★

Multimodality

Robotic AI capabilities

Dexterous manipulation Robot manipulation Robot control Motion planning

Applications

Sources

Website1Technical report1Research paper1

01

Gemini Robotics On-Device - Google DeepMind

deepmind.googleWebsite

02

Gemini Robotics On-Device Model Card

storage.googleapis.comTechnical report

03

Gemini Robotics: Bringing AI into the Physical World (arXiv:2503.20020)

arxiv.orgResearch paper