# thu-ml/RoboticsDiffusionTransformer

RDT-1B: a Diffusion Foundation Model for Bimanual Manipulation

Repository: https://github.com/thu-ml/RoboticsDiffusionTransformer
Canonical: https://ross.abutalabs.com/products/roboticsdiffusiontransformer
Language: Python
License: MIT
License Family: permissive
Last push: 2026-01-21T09:48:33+00:00

## Health v2 (maintenance only)
Score: 50/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 63, release rhythm 35, longevity 49
- inputs: {"age_days": 695, "days_push": 224, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1778, forks 162 (observed 2026-08-28T04:05:35.158632+00:00)

## What it is
RDT-1B is a 1B-parameter diffusion foundation model for robot bimanual manipulation, pre-trained on 1M+ multi-robot episodes to predict robot actions from language instructions and RGB images. The repo provides the official PyTorch implementation with model code, pre-trained checkpoints, training/sampling scripts, and real-robot deployment examples.

## Use cases
- fine-tune a robot manipulation foundation model on my own dataset
- deploy a learned policy on a dual-arm ALOHA robot
- evaluate imitation learning policies in ManiSkill simulation
- predict robot actions from camera images and language instructions
- train a diffusion transformer for robotics with DeepSpeed
- run zero-shot manipulation with a pre-trained robotics model

## When to choose
- you need a pre-trained foundation model for bimanual or single-arm manipulation
- you want to fine-tune a large robotics policy on custom episodes
- you need compatibility with varied manipulators (joint/EEF, position/velocity control)
- you want a state-of-the-art imitation learning baseline for manipulation benchmarks

## When to avoid
- you need lightweight models for low-VRAM hardware (though RDT-170M partially addresses this)
- your task is not robot manipulation (e.g., navigation or locomotion only)
- you lack GPU resources for training or inference of billion-parameter models
- you need a plug-and-play product rather than a research codebase

## Facets
- artifact type: library
- maturity: active
- function: machine-learning, deep-learning, llm-training, simulation
- domain: robotics, machine-learning, deep-learning, artificial-intelligence
- platform: python
- tags: diffusion-transformer, imitation-learning, robot-manipulation, foundation-model, bimanual-manipulation, pytorch, deepspeed, aloha-robot, maniskill, gpu, linux

## Member repositories
- thu-ml/RoboticsDiffusionTransformer (main) score 50

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:05:35.158632+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T03:25:10.988159+00:00, confidence not recorded.
  - readme: https://github.com/thu-ml/RoboticsDiffusionTransformer (fetched 2026-08-28T04:05:35.158632+00:00, sha 195c7c7d0c00)
- Data as of 2026-08-30T08:39:29.467469+00:00.