Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

GR-Dexter Technical Report

Ruoshi Wen, Guangzeng Chen, Zhongren Cui, Min Du, Yang Gou, Zhigang Han, Liqun Huang, Mingyu Lei, Yunfei Li, Zhuohang Li, Wenlei Liu, Yuxiao Liu, Xiao Ma, Hao Niu, Yutao Ouyang, Zeyu Ren, Haixin Shi, Wei Xu, Haoxiang Zhang, Jiajun Zhang, Xiao Zhang, Liwei Zheng, Weiheng Zhong, Yifei Zhou, Zhengming Zhu, Hang Li

TL;DR

GR-Dexter advances generalist dexterous-hand manipulation by integrating a compact $56$-DoF bimanual platform with ByteDexter V2 and a multi-source VLA policy. The approach combines a $4$-billion-parameter Mixture-of-Transformer model pretrained on vision-language data with cross-embodiment and human-trajectory datasets, then fine-tuned on teleoperated robot trajectories to adapt to the target hardware. A unified cross-embodiment retargeting pipeline enables transferring diverse demonstrations to ByteDexter V2, supporting long-horizon tasks and generalization to unseen objects and commands. Real-world experiments show strong in-domain performance and improved robustness to unseen layouts, objects, and instructions, highlighting a practical path toward generalist dexterous-hand manipulation and scalable data collection through teleoperation and cross-embodiment supervision.

Abstract

Vision-language-action (VLA) models have enabled language-conditioned, long-horizon robot manipulation, but most existing systems are limited to grippers. Scaling VLA policies to bimanual robots with high degree-of-freedom (DoF) dexterous hands remains challenging due to the expanded action space, frequent hand-object occlusions, and the cost of collecting real-robot data. We present GR-Dexter, a holistic hardware-model-data framework for VLA-based generalist manipulation on a bimanual dexterous-hand robot. Our approach combines the design of a compact 21-DoF robotic hand, an intuitive bimanual teleoperation system for real-robot data collection, and a training recipe that leverages teleoperated robot trajectories together with large-scale vision-language and carefully curated cross-embodiment datasets. Across real-world evaluations spanning long-horizon everyday manipulation and generalizable pick-and-place, GR-Dexter achieves strong in-domain performance and improved robustness to unseen objects and unseen instructions. We hope GR-Dexter serves as a practical step toward generalist dexterous-hand robotic manipulation.

GR-Dexter Technical Report

TL;DR

GR-Dexter advances generalist dexterous-hand manipulation by integrating a compact -DoF bimanual platform with ByteDexter V2 and a multi-source VLA policy. The approach combines a -billion-parameter Mixture-of-Transformer model pretrained on vision-language data with cross-embodiment and human-trajectory datasets, then fine-tuned on teleoperated robot trajectories to adapt to the target hardware. A unified cross-embodiment retargeting pipeline enables transferring diverse demonstrations to ByteDexter V2, supporting long-horizon tasks and generalization to unseen objects and commands. Real-world experiments show strong in-domain performance and improved robustness to unseen layouts, objects, and instructions, highlighting a practical path toward generalist dexterous-hand manipulation and scalable data collection through teleoperation and cross-embodiment supervision.

Abstract

Vision-language-action (VLA) models have enabled language-conditioned, long-horizon robot manipulation, but most existing systems are limited to grippers. Scaling VLA policies to bimanual robots with high degree-of-freedom (DoF) dexterous hands remains challenging due to the expanded action space, frequent hand-object occlusions, and the cost of collecting real-robot data. We present GR-Dexter, a holistic hardware-model-data framework for VLA-based generalist manipulation on a bimanual dexterous-hand robot. Our approach combines the design of a compact 21-DoF robotic hand, an intuitive bimanual teleoperation system for real-robot data collection, and a training recipe that leverages teleoperated robot trajectories together with large-scale vision-language and carefully curated cross-embodiment datasets. Across real-world evaluations spanning long-horizon everyday manipulation and generalizable pick-and-place, GR-Dexter achieves strong in-domain performance and improved robustness to unseen objects and unseen instructions. We hope GR-Dexter serves as a practical step toward generalist dexterous-hand robotic manipulation.
Paper Structure (35 sections, 9 figures)

This paper contains 35 sections, 9 figures.

Table of Contents

  1. Introduction
  2. ByteDexter Robotic Hand
  3. Hand Design
  4. Bimanual System and Control
  5. Bimanual teleoperation
  6. Policy rollout
  7. The GR-Dexter Model
  8. Training Recipe
  9. Vision-language data
  10. Cross-embodiment data
  11. Human trajectories
  12. Cross-Embodiment Motion Retargeting and Transferring
  13. Transferring cross-embodiment trajectories
  14. Transferring human trajectories
  15. Experiments
  16. ...and 20 more sections

Figures (9)

  • Figure 1: GR-Dexter performs dexterous long-horizon daily tasks and generalizes to out-of-domain settings by learning from four data sources: vision--language, cross-embodiment, human-trajectory, and robot-trajectory data.
  • Figure 2: Capabilities. GR-Dexter robustly completes long-horizon daily tasks. It also learns to grasp unseen objects, and follow unseen, abstract language instructions.
  • Figure 3: The ByteDexter V2 hand. We show the DoF distribution, tactile fingertips, and the thumb's opposition capability.
  • Figure 4: The bimanual robotic system comprising two Franka Research 3 arms equipped with ByteDexter V2 hands. Data are collected via a teleoperation interface using a Meta Quest VR headset, Manus gloves with mounted VR tracking controllers, and a set of global RGB-D cameras.
  • Figure 5: Teleoperation capability in long-horizon dexterous grasping and bimanual manipulation tasks.
  • ...and 4 more figures