Vision-language-action

RT-2

RT-2 studies transferring pretrained vision-language representations into robot control by treating robot action tokens as another output language.

Source contextPrimary specReviewed 2026-09-28

Sources 1

RT-2 research project

Google DeepMind · Publication date not disclosed · Source accessed 2026-09-28

Reviewed recordLast reviewed 2026-09-281 sources ↗4 sourced fields

Official source Suggest a correction

Architecture and scope

The project evaluates PaLM-E and PaLI-X based variants. Those are distinct model configurations; the RT-2 name alone cannot identify the model size, training recipe or evaluation embodiment.

Source contextPrimary specReviewed 2026-09-28

Sources 1

RT-2 research project

Google DeepMind · Publication date not disclosed · Source accessed 2026-09-28

Integration and evaluation

The authors examine unfamiliar objects, semantic reasoning and task generalization. These research tests are narrower than an operational claim about unsupervised customer use.

Source contextPrimary specReviewed 2026-09-28

Sources 1

RT-2 research project

Google DeepMind · Publication date not disclosed · Source accessed 2026-09-28

Evidence boundary

Action-token predictions and qualitative demonstrations do not prove safety or deployment reliability.

Source contextPrimary specReviewed 2026-09-28

Sources 1

RT-2 research project

Google DeepMind · Publication date not disclosed · Source accessed 2026-09-28

Model identity

Developer: Google DeepMind
Primary specReviewed 2026-09-28
Sources 1
RT-2 research project
Google DeepMind · Publication date not disclosed · Source accessed 2026-09-28
Exact version / scope: RT-2 PaLM-E and RT-2 PaLI-X research variants
Primary specReviewed 2026-09-28
Sources 1
RT-2 research project
Google DeepMind · Publication date not disclosed · Source accessed 2026-09-28
Category: Vision-language-action
Primary specReviewed 2026-09-28
Sources 1
RT-2 research project
Google DeepMind · Publication date not disclosed · Source accessed 2026-09-28
Access model: Public research project; public checkpoint access not verified
Primary specReviewed 2026-09-28
Sources 1
RT-2 research project
Google DeepMind · Publication date not disclosed · Source accessed 2026-09-28
License: Not publicly disclosed

Inputs, outputs & embodiment

Inputs: Not publicly disclosed
Outputs: Not publicly disclosed
Embodiment: Not publicly disclosed

Implementation & training

Compute: Not publicly disclosed
Training-data disclosures: Not publicly disclosed
Integrations: Not publicly disclosed

Reported benchmark results

Context before scores. Results from different benchmarks are not directly comparable. A simulation result does not establish real-world reliability, safety or commercial availability.

No result meets our complete revision and methodology requirements for this record yet. Inspect the original benchmark documentation before comparing published scores.

What this evidence does not establish

Action-token predictions and qualitative demonstrations do not prove safety or deployment reliability.
Pricing and deployment service terms were not verified.

Relationships & deployments

Related reading is an editorial crosslink. Sourced connections describe relationships reported in the cited material. A link to a versioned profile does not establish compatibility with that version unless the connection note explicitly identifies it.

develops

Google DeepMind → RT-2

Sources 1

RT-2 research project

Google DeepMind · Publication date not disclosed · Source accessed 2026-09-28

Connection reviewed 2026-09-28