New: the Timeline is live. Track world model releases, papers, and benchmark updates in real time.
world-models.io
The Knowledge Hub for AI World Models

RT-2

RT-2 (Robotic Transformer 2) is a Vision-Language-Action model by Google DeepMind that transfers knowledge from web-scale vision-language pretraining directly to robotic control policies.

robotics model-based-rl simulation embodied-ai

Key Attributes

Static HTML snapshot of the model record for crawlers and no-JS readers.

AttributeValue
ModelRT-2
Lab / OrganizationDeepMind
CategoryFoundation World Model
SubtypeVision-Language-Action Model
World Model TypeWeb-knowledge transfer model for robotics
Primary DomainRobotics
ArchitecturePaLI-X / PaLM-E backbone fine-tuned with action tokenization
ModalityVisual + Language + Proprioceptive
Training MethodCo-fine-tuning on web VLM data + robot demonstration trajectories
Statusactive
Year2023
Performance Index72/100 (medium confidence, v1.1)

About RT-2

Main editorial body preserved directly in static HTML.

RT-2 (Robotic Transformer 2) is a vision-language-action (VLA) model that co-fine-tunes a large vision-language model on both web data and robotic trajectories. By tokenizing robot actions as text tokens, RT-2 enables a single model to perform visual question answering, image captioning, and robot control simultaneously. It demonstrates emergent capabilities such as reasoning about previously unseen objects, interpreting abstract instructions, and performing multi-step manipulation tasks, all without task-specific training.

RT-2 is a web-knowledge transfer model for robotics developed by Google DeepMind in 2023 for robotics.

Editorial Snapshot

Short extractable facts for answer engines and no-JS readers.

SignalValue
DefinitionRT-2 is a web-knowledge transfer model for robotics developed by Google DeepMind in 2023 for robotics.
Short DescriptionA vision-language-action model that transfers web-scale knowledge directly to robot control.
Benchmark Rows2
FAQ Entries2
Related Models4
Related Guides0
Related Research Topics1
Last Updated2026-04-07

Notable Features

Key capabilities associated with this model.

  • Emergent reasoning about novel objects
  • Action tokenization within language model
  • Zero-shot generalization to unseen instructions
  • Combines web knowledge with physical grounding

Use Cases

Representative applications attached to this model record.

Robotic manipulationInstruction-following robotsObject reasoningMulti-step task execution

Strengths and Limitations

Balanced assessment surfaced in static HTML.

Strengths

  • Web-scale knowledge transfer
  • No task-specific modules needed
  • Handles novel objects and instructions
  • Unified architecture for perception and action

Limitations

  • Requires large-scale compute
  • Limited to manipulation tasks in evaluation
  • Single-robot arm setting
  • Inference latency for real-time control

Benchmarks

Published benchmark evidence attached to this model record.

BenchmarkMetricResultSource
Emergent Skills EvalSuccess Rate 62 %62% success on novel objectsSource
Language TableTask Success 90 %90% successSource

References and Citations

Primary references preserved in static HTML for citation extraction.

ReferenceLink
Brohan et al., 2023. RT-2: Vision-Language-Action Models Transfer Web Knowledge to Robotic Control. arXiv:2307.15818Open source

Related Models

Nearby models linked from the current editorial record.

ModelCategoryWorld Model TypeIndex v1.1
3D-VLAFoundation World Model3D-aware embodied world model47/100
AMI World ModelFoundation World ModelMultimodal generative world model38/100
TD-MPC2Model-Based RLImplicit dynamics + MPC planner80/100

Direct Comparisons

Side-by-side comparisons already connected to this model.

ComparisonMatchupSummary
RT-2 vs 3D-VLART-2 vs 3D-VLATwo approaches to vision-language-action models for robotics. RT-2 leverages web-scale VLM knowledge through action tokenization, while 3D-VLA integrates explicit 3D spatial understanding for embodied reasoning.

Research Topics Referencing This Model

Connected research areas surfaced directly in static HTML.

TopicSummary
Language-Conditioned World ModelsHow language-conditioned world models use text prompts or natural-language actions to control simulation, planning, and embodied behavior across Pandora, 3D-VLA, RT-2, and hybrid systems.

Frequently Asked Questions

FAQ answers rendered directly into static HTML for extractable responses.

How does RT-2 differ from RT-1?

RT-1 was trained only on robot data. RT-2 co-fine-tunes a large vision-language model (PaLI-X or PaLM-E) on both web and robot data, enabling emergent reasoning capabilities that RT-1 lacks.

Can RT-2 handle objects it has never seen?

Yes. By leveraging web-scale visual knowledge, RT-2 can reason about and manipulate objects not present in its robot training data, a key emergent capability.

Quick Answer

Short extractable summary preserved directly in static HTML.

  • RT-2 is a web-knowledge transfer model for robotics developed by Google DeepMind in 2023 for robotics.
  • Use this page when you need a fast read on how RT-2 fits into the foundation world model landscape, then validate the details in the benchmarks, citations, and related pages.
  • A key strength surfaced in the editorial record is web-scale knowledge transfer.

Editorial Trust Signals

Editorial provenance and refresh policy preserved directly in static HTML.

Published by world-models.io editorial board.

Lead editor Tyler D. - Technical editor, methodology and benchmark analysis.

This model page synthesizes primary papers, official model pages, benchmark evidence, and related world-models.io context into a reference resource.

Each editorial page is assembled from primary sources, normalized into extractable summaries, checked for factual drift, and reviewed before publication or major refreshes. Last reviewed: 2026-04-07.

Pages are refreshed when a new paper, benchmark, release, architecture update, or stronger primary source materially changes the answer a reader or AI system should retrieve.

Each page links back to relevant primary sources and keeps a stable canonical URL so readers can verify claims, trace context, and reference the most up-to-date version. See the editorial policy.

Primary sources onlyLast reviewed date visibleMethodology documentedSource links included

External Sources

Primary model and lab sources embedded in static HTML.

References

  1. [1] Brohan et al., 2023. RT-2: Vision-Language-Action Models Transfer Web Knowledge to Robotic Control. arXiv:2307.15818