Research Google Robotics Proprietary (research)

Google RT-2 (Robotics Transformer)

Google's vision-language-action model — robots that reason from web knowledge

Released 2023-07-28 · ~55B

Overview

Google's vision-language-action model for robotic control

Specifications

DeveloperGoogle
Release date2023-07-28
Model typeRobotics
Parameters~55B
ArchitectureTransformer (PaLI-X / PaLM-E based)
LicenseProprietary (research)
Input modalitiestext, image
Output modalitiesrobot actions

Benchmarks

Benchmark RT-2 (Robotics Transformer)
Success Rate 62%
Novel Tasks 3x improvement
Emergent Skills High

Availability

Research Only

Official references