# RDT-1B
RDT-1B is a 1B-parameter (1.2B by the paper's count) diffusion transformer for bimanual robot manipulation, released in October 2024 by the RDT team of the TSAIL group at [Tsinghua University](https://www.edgepedia.org/tsinghua-university) and described by its authors as the largest diffusion-based foundation model for robotic manipulation at the time.[1](https://arxiv.org/html/2410.07864v1) At inference it takes a language instruction, RGB images from up to three camera views, the control frequency, and the robot's proprioceptive state, and predicts the next 64 robot actions as a chunk, which the robot then executes.[2](https://huggingface.co/robotics-diffusion-transformer/rdt-1b/blob/main/README.md) Unlike vision-language-action (VLA) models such as [OpenVLA](https://www.edgepedia.org/openvla), it has no vision-language model backbone: a frozen SigLIP vision encoder and a frozen T5-XXL text encoder feed a diffusion policy head directly.[3](https://www.ay-robots.com/arena/rdt-1b)
| Fact | Value |
|---|---|
| Developer | RDT team, TSAIL group, Tsinghua University[2](https://huggingface.co/robotics-diffusion-transformer/rdt-1b/blob/main/README.md) |
| First release | October 2024 (repository created 7 October 2024)[4](https://github.com/thu-ml/RoboticsDiffusionTransformer) |
| Parameters | 1B (name and README); 1.2B in the paper[4](https://github.com/thu-ml/RoboticsDiffusionTransformer) • [1](https://arxiv.org/html/2410.07864v1) |
| Training data | 46 datasets, 1M+ episodes, 21 TB; 6K+ ALOHA bimanual fine-tuning trajectories[5](https://www.ay-robots.com/blog/en/rdt-1b-diffusion-foundation-model) • [1](https://arxiv.org/html/2410.07864v1) |
| Inference rate | 6 action chunks/sec (381 actions/sec) on an onboard RTX 4090 24GB[1](https://arxiv.org/html/2410.07864v1) |
| License | MIT for code, weights, and data; commercial use permitted[2](https://huggingface.co/robotics-diffusion-transformer/rdt-1b/blob/main/README.md) |
| Publication | ICLR 2025[6](https://proceedings.iclr.cc/paper_files/paper/2025/file/49f80e4d2471ad4f2edf4f5f1ab62339-Paper-Conference.pdf) |
## Architecture and training
RDT is a diffusion transformer trained to denoise action chunks conditioned on language, images, and proprioception (the body's internal sense of its own position and movement). It deliberately avoids a VLM backbone: a frozen siglip-so400m-patch14-384 vision encoder and a frozen t5-v1_1-xxl text encoder supply the conditioning signals, and the policy emits a chunk of 64 future actions from up to three camera views plus proprioception.[3](https://www.ay-robots.com/arena/rdt-1b) • [2](https://huggingface.co/robotics-diffusion-transformer/rdt-1b/blob/main/README.md)
To train across many robots, the authors introduce a Physically Interpretable Unified Action Space of 128 dimensions that unifies action representations across embodiments while preserving physical meanings such as joint angles and end-effector poses.[1](https://arxiv.org/html/2410.07864v1) Pre-training covered 46 datasets including the RT-1 Dataset, RH20T, DROID, BridgeData V2, RoboSet, and a subset of [Open X-Embodiment](https://www.edgepedia.org/open-x-embodiment), totaling more than 1 million trajectories and 21 TB of data.[2](https://huggingface.co/robotics-diffusion-transformer/rdt-1b/blob/main/README.md) • [5](https://www.ay-robots.com/blog/en/rdt-1b-diffusion-foundation-model) To strengthen bimanual capability, the team collected 6K+ episodes on the ALOHA dual-arm robot covering 300+ tasks, 100+ objects, and 15+ rooms for fine-tuning.[1](https://arxiv.org/html/2410.07864v1) • [7](https://rdt-robotics.github.io/rdt-robotics/) Pre-training ran on 48 H100 80GB GPUs for one month (1M steps); fine-tuning took three days for 130K steps on the same hardware.[1](https://arxiv.org/html/2410.07864v1)
## Release timeline and versions
The GitHub repository was created on 7 October 2024. The RDT-170M checkpoint, a smaller variant positioned as a more VRAM-friendly option, was released on 23 October 2024. ManiSkill evaluation scripts followed on 17 December 2024, and the paper was accepted at ICLR 2025.[4](https://github.com/thu-ml/RoboticsDiffusionTransformer) • [6](https://proceedings.iclr.cc/paper_files/paper/2025/file/49f80e4d2471ad4f2edf4f5f1ab62339-Paper-Conference.pdf) The repository's last push was 21 January 2026; new work moved to the RDT2 repository.[5](https://www.ay-robots.com/blog/en/rdt-1b-diffusion-foundation-model)
## Benchmark results: vendor versus independent
Vendor-reported results come from the paper and the project page. The authors report a 56% improvement in success rates over baselines including ACT, OpenVLA (7B), and Octo (93M), with zero-shot generalization to unseen objects and scenes and few-shot learning from 1 to 5 demonstrations.[1](https://arxiv.org/html/2410.07864v1) Real-robot numbers include 87.5% on a Wash Cup task with an unseen cup (best baseline ACT: 37.5%), 100% on a 5-shot Handover task, and 48% on a robot-dog walk-straight task (best baseline Octo: 4%).[3](https://www.ay-robots.com/arena/rdt-1b) The project page also demonstrates zero-shot instruction following on unseen water-level quantities (1/3, 2/3) after training only on "little", "half," and "full".[7](https://rdt-robotics.github.io/rdt-robotics/)
In the [ManiSkill](https://www.edgepedia.org/maniskill) simulation benchmark (250 trials per task, 10 seeds), the repository reports RDT averaging 53.6% success across five tasks versus 4.8% for OpenVLA, 0.0% for Octo, and 30.2% for Diffusion-Policy.[4](https://github.com/thu-ml/RoboticsDiffusionTransformer) Independent analysis qualifies this mean: it is carried by the easier tasks. RDT scored 1.2% on PlugCharger and 13.2% on PegInsertionSide, and the AY-Robots analysis notes that contact-rich insertion remained unsolved for all four methods.[4](https://github.com/thu-ml/RoboticsDiffusionTransformer) • [5](https://www.ay-robots.com/blog/en/rdt-1b-diffusion-foundation-model) The same analysis points out that most real-robot evaluation rows use 8 trials per task, so individual percentages carry wide confidence intervals.[3](https://www.ay-robots.com/arena/rdt-1b)
Leaderboard standing also eroded over time. The repository's README cites the [RoboTwin](https://www.edgepedia.org/robotwin) 2.0 leaderboard with RDT ranking second only to Pi0 among listed single-task baselines, but that is a 2025 snapshot. At the leaderboard's 18 August 2026 update, RDT sat eleventh on the hard clean2random ranking at 13.72%, behind pi0 at 16.34% and well behind 2026 co-trained models led by GigaBrain-0.7 at 67.9%.[4](https://github.com/thu-ml/RoboticsDiffusionTransformer) • [5](https://www.ay-robots.com/blog/en/rdt-1b-diffusion-foundation-model)
## How it compares with other robot foundation models
RDT-1B's main design contrast is with VLA models. OpenVLA is a 7B model that discretizes actions and builds on a vision-language backbone; Octo is a 93M transformer; ACT is a small imitation-learning policy. RDT sits between them in size at 1B/1.2B parameters and avoids the VLM backbone entirely, feeding frozen encoders into a diffusion policy that outputs continuous 64-action chunks.[1](https://arxiv.org/html/2410.07864v1) • [3](https://www.ay-robots.com/arena/rdt-1b) In ManiSkill it outscored both OpenVLA and Octo by wide margins, though the contact-rich tasks were poor for all methods.[4](https://github.com/thu-ml/RoboticsDiffusionTransformer) On the RoboTwin 2.0 leaderboard it trailed pi0 in both the 2025 snapshot and the 2026 update.[4](https://github.com/thu-ml/RoboticsDiffusionTransformer) • [5](https://www.ay-robots.com/blog/en/rdt-1b-diffusion-foundation-model)
## Licensing, availability and running it
All code, pre-trained weights, and data are MIT-licensed, which permits commercial use.[2](https://huggingface.co/robotics-diffusion-transformer/rdt-1b/blob/main/README.md) The released checkpoint is about 2.46 GB; independent computation puts the bf16 weights at about 2.2 GB, small enough for an 8 GB card to hold for inference.[5](https://www.ay-robots.com/blog/en/rdt-1b-diffusion-foundation-model) • [3](https://www.ay-robots.com/arena/rdt-1b) The practical memory bottleneck is the T5-XXL text encoder: the repository recommends precomputing language embeddings offline because on an RTX 4090 or lower GPU, memory may be too low to load t5-v1_1-xxl directly.[4](https://github.com/thu-ml/RoboticsDiffusionTransformer) The authors report 6 Hz chunk inference (381 actions/sec) on an onboard RTX 4090 24GB using [DPM-Solver](https://www.edgepedia.org/dpm-solver)++ to cut sampling from 100 to 5 steps, and state this supports arm speeds comparable to human operators.[1](https://arxiv.org/html/2410.07864v1) • [7](https://rdt-robotics.github.io/rdt-robotics/) The repository ships [DeepSpeed](https://www.edgepedia.org/deepspeed) training scripts and a [Mobile ALOHA](https://www.edgepedia.org/mobile-aloha) deployment example; all real-robot evaluation was on a Cobot Mobile ALOHA, with the mobile base unused during training and inference.[4](https://github.com/thu-ml/RoboticsDiffusionTransformer) • [5](https://www.ay-robots.com/blog/en/rdt-1b-diffusion-foundation-model)
## Reception, limits and criticisms
The model card itself states the central limitation: due to the embodiment gap, RDT cannot yet generalize to new robot platforms not seen in the pre-training datasets, and the authors recommend fine-tuning on a small target-robot dataset.[2](https://huggingface.co/robotics-diffusion-transformer/rdt-1b/blob/main/README.md) Data scarcity for bimanual tasks is the second structural limit: the 6K+ ALOHA fine-tuning set, described as one of the largest collected to date, is still small next to the 1M+ single-arm pre-training episodes.[7](https://rdt-robotics.github.io/rdt-robotics/)
On evaluation rigor, the independent analysis notes that no official VRAM or millisecond latency figure was published, only the 6 Hz chunk rate on an RTX 4090, and that 8-trial real-robot evaluations give wide confidence intervals.[3](https://www.ay-robots.com/arena/rdt-1b)
## References
1. RDT-1B: a Diffusion Foundation Model for Bimanual Manipulation (arXiv, Liu et al., October 2024) — https://arxiv.org/html/2410.07864v1
2. Hugging Face model card: robotics-diffusion-transformer/rdt-1b — https://huggingface.co/robotics-diffusion-transformer/rdt-1b/blob/main/README.md
3. RDT-1B: VRAM, Latency, Parameters and Benchmarks (AY-Robots arena page) — https://www.ay-robots.com/arena/rdt-1b
4. thu-ml/RoboticsDiffusionTransformer — official GitHub repository — https://github.com/thu-ml/RoboticsDiffusionTransformer
5. RDT-1B: the 1.2B Diffusion Foundation Model for Bimanual Robots (AY-Robots, independent technical analysis) — https://www.ay-robots.com/blog/en/rdt-1b-diffusion-foundation-model
6. RDT-1B — ICLR 2025 proceedings version — https://proceedings.iclr.cc/paper_files/paper/2025/file/49f80e4d2471ad4f2edf4f5f1ab62339-Paper-Conference.pdf
7. RDT-1B official project page — https://rdt-robotics.github.io/rdt-robotics/
---
*Topic: Encyclopedia › Technology and the built world › Computing and digital systems › Modern AI: foundation models, generative AI, and the AI industry › Model families and named models › Multimodal, vision, and world models*
*Initially written Sep 17, 2026 · Edited: Oct 11, 2026; Oct 15, 2026; Oct 21, 2026*
*Copyright 2026 EdgeChat AI, a subsidiary of Biostate AI.*
License: Edgepedia Community License 1.0, https://www.edgepedia.org/edgepedia/license