Brain
Robotics Diffusion Transformer (RDT-1B): a 1B parameter diffusion-based VLA for bimanual manipulation. Developed by Tsinghua University. Competitive with pi0 and OpenVLA on real-world dexterous tasks.
Research model · Maturity: Research · Open source
Machine-readable surfaces
- Markdown mirror: /brains/rdt-1b.md
- RSS feed: /brains/rdt-1b/feed.xml
- JSON-LD: embedded in this page’s head
- REST API: /v1/brains/2cd7e049-d607-4549-bafc-5ebef5e52734
- Revision history: /brains/rdt-1b/history
- Data documentation: /data
- Query this programmatically: Deploy MCP
Architecture
Diffusion Transformer, 1B params
Key facts
- Model scale and pretraining (primary project page)
- The RDT-1B project page presents Robotics Diffusion Transformer with 1.2B parameters as a diffusion-based foundation model for robotic manipulation, pretrained on a multi-robot collection of 46 datasets with 1M+ episodes, then fine-tuned on 6K+ ALOHA dual-arm episodes. Authors are from Tsinghua University. Treat largest framing and dataset counts as author-stated (RDT project page)
- Capabilities claimed (primary project page)
- The page says RDT supports control of modern manipulators including dual-arm, joints, end-effectors, and wheeled locomotion, and reports demos of dexterity, zero-shot generalization to unseen objects and scenes, instruction following, few-shot learning with 1-5 demonstrations, and real-time inference at 6 action chunks/sec or 381 actions/sec. Treat performance claims as project demos, not a fleet census (RDT project page)
- Official code and model release
- The RDT project links its implementation and training materials through the RoboticsDiffusionTransformer repository and publishes an RDT-1B model entry on Hugging Face (RDT GitHub, RDT-1B model)
- Bimanual manipulation research scope
- The RDT project presents RDT-1B as a diffusion foundation model for bimanual manipulation and documents real robot demonstrations and training resources on its project page (RDT project)
- Diffusion policy
- The RDT project presents RDT-1B as a diffusion foundation model for bimanual robotic manipulation (RDT-1B project).
- Multi-robot pretraining
- The RDT project says its model was pretrained on a multi-robot collection spanning more than one million episodes and 46 datasets (RDT-1B project).
- Open implementation
- The RDT team publishes implementation and training materials in its RoboticsDiffusionTransformer repository (RDT GitHub).
Common questions
What is RDT-1B?
Robotics Diffusion Transformer (RDT-1B): a 1B parameter diffusion-based VLA for bimanual manipulation. Developed by Tsinghua University. Competitive with pi0 and OpenVLA on real-world dexterous tasks.
Is RDT-1B open source?
Yes. RDT-1B is recorded as open-source / open-weights on the DEPLOY registry, meaning model weights or source code are publicly available.
What type of AI is RDT-1B?
RDT-1B is a research model, built on a Diffusion Transformer, 1B params architecture on the DEPLOY registry.
What is RDT-1B's maturity stage?
RDT-1B is at the research stage on the DEPLOY maturity ladder. Research stage means active development without commercial deployments on file.
Which robots run on RDT-1B?
No robot models on the DEPLOY registry are recorded as running RDT-1B. DEPLOY wires brain-to-model connections only when the wiring is verifiable from primary sources; absence may reflect pre-deployment or unverified manufacturer claims.
Sources (3)
Methodology: Verified · 3 sources (no primary) · last reviewed 2026-10-09
Verification posture
Verified
Low confidence
Review state
Stable
Last reviewed 2026-10-09
Maturity + lifecycle
Maturity stage: research
Sources by quality tier
- 1
- unclassified
- Unclassified source
- 1
- code-repository
- Code repository
- 1
- model-repository
- Model repository
The framework is documented at /methodology. Corrections at /corrections. Reviewer: DEPLOY editorial team.
Methodology surface for RDT-1B.Canonical ID 2cd7e049-d607-4549-bafc-5ebef5e52734