RDT Team

RDT-1B

Checkpoint

1.2B diffusion foundation model for bimanual manipulation.

Specifications

Architecture components: Diffusion action decoder
Backbone: T5 language encoder, SigLIP vision encoder and diffusion transformer
Frameworks: PyTorch / Transformers
Input modalities: Language instruction
Input modalities: RGB camera images
Input modalities: Robot state / proprioception
Languages: English
Model type: Vision-language-action model
Output action space: Continuous bimanual action chunks
Parameter count: 1.2B parameters
Pipeline task: Robotics

Unknown

trajectory · continuous

Unknown

Continuous bimanual action chunks

Default checkpoint

openMIT License

Other releases

published checkpoints

No verified robot compatibility recorded.

No published benchmark results.

PyTorch / Transformers · NVIDIA GPU

bfloat16 · unknown

Submit info