XR-1-Stage1-UVMC

[Project Page] [Paper] [GitHub]

This repository contains the Stage 1 weights for the XR-1 (X Robotic Model 1) project, specifically the Unified Vision-Motion Codes (UVMC) tokenizer.

🤖 Model Description

The UVMC is a core component of the XR-1 Vision-Language-Action (VLA) framework. It utilizes a dual-branch VQ-VAE (Vector Quantized-Variational Autoencoder) architecture to encode both environmental visual dynamics and robot actions into a shared, discrete latent space.

Key Features

Unified Representation: Maps high-dimensional images and continuous robot actions into a single discrete vocabulary.
Cross-Embodiment: Trained on diverse datasets (RoboMIND 2.0) to support multiple robot configurations (e.g., Tien Kung, Franka, UR-5e).
Foundation for VLA: Provides the essential tokenization layer for the Stage 2 Transformer-based policy training.

🛠 Usage

To use these weights, please clone the official XR-1 GitHub Repository and follow the installation instructions.

📝 Citation

If you find this model or the XR-1 framework useful in your research, please cite our work:

@article{fan2025xr,
  title={XR-1: Towards Versatile Vision-Language-Action Models via Learning Unified Vision-Motion Representations},
  author={Fan, Shichao and Wu, Kun and Che, Zhengping and Wang, Xinhua and Wu, Di and Liao, Fei and Liu, Ning and Zhang, Yixue and Zhao, Zhen and Xu, Zhiyuan and others},
  journal={arXiv preprint arXiv:2411.02776},
  year={2025}
}