model · model--gemini-robotics-on-device-2

Gemini Robotics On-Device 2

On-device VLA for standing bi-arm manipulation, distributed to select Trusted Testers.

Identity and state

PublicationRestricted
DeploymentExternal Evaluation
Canonical claims7
Last reviewed2026-09-04

Reviewed relationships

StateRelated entityRelationScopeBoundary
VerifiedGoogle DeepMindpublishesmodel cardAccess is limited to select Trusted Testers.

Canonical evidence claims

Claims are projected from authored tables. A dash remains an explicit missing value; it is never filled from a neighboring product or name match.

MetricValueClass
Artifact kindvision-language-action model based on on-device Gemma models [software-m27]robot-learning-runtime
Published data pathimages, text, robot sensor data, and action data; dataset details are not published [software-m27]robot-learning-runtime
Distribution / accessselect Trusted Testers only [software-m27]robot-learning-runtime
Execution topologygeneral-purpose manipulation runs on local devices [software-m27]robot-learning-runtime
Input / output contracttext, images, and numerical proprioception → numerical robot actions [software-m27]robot-learning-runtime
Published boundaryhigh-DoF generalization is limited; mobile-platform and whole-body risks are outside the evaluated scope [software-m27]robot-learning-runtime
Stated scopestanding bi-arm manipulation across several robot types [software-m27]robot-learning-runtime

Source ledger

What is not inferred

No fuzzy entity matching, adjacent-column borrowing, universal compatibility, commercial maturity, or checkpoint-to-product identity. Absence of a relation means unreviewed or unsupported here—not proven incompatibility.