UniBiDex: A Unified Teleoperation Framework for Robotic Bimanual Dexterous Manipulation

Zhongxuan Li1,* Zeliang Guo2,* Jun Hu2 David Navarro-Alarcon3 Jia Pan1 Hongmin Wu4,† Peng Zhou2,†
1 Department of Computer and Data Science, The University of Hong Kong
2 Dongguan Key Laboratory of Intelligent Equipment and Smart Industry, School of Advanced Engineering, The Great Bay University
3 Department of Mechanical Engineering, The Hong Kong Polytechnic University
4 Guangdong Academy of Sciences
* Equal contribution, † Corresponding author
Paper Code

Abstract

We present UniBiDex, a unified teleoperation framework for robotic bimanual dexterous manipulation that supports both VR-based and leader–follower input modalities. UniBiDex enables real-time, contact-rich dual-arm teleoperation by integrating heterogeneous input devices into a shared control stack with consistent kinematic treatment and safety guarantees.

The framework employs null-space control to optimize bimanual configurations, ensuring smooth, collision-free and singularity-aware motion across tasks. We validate UniBiDex on a long-horizon kitchen-tidying task involving five sequential manipulation subtasks, demonstrating higher task success rates, smoother trajectories, and improved robustness compared to strong baselines.

By releasing all hardware and software components as open-source, we aim to lower the barrier to collecting large-scale, high-quality human demonstration datasets and accelerate progress in robot learning.

System Overview

The proposed system supports two teleoperation input modalities: VR headsets and leader–follower arms. For the VR mode, we use the Meta Quest 3, while the leader–follower mode is based on the hardware design of the GELLO system. A unified dual-arm control layer minimizes collisions and inverse-kinematics failures in both modes, enabling precise and reliable bimanual manipulation.

Framework Architecture

UniBiDex Framework Architecture

The framework integrates heterogeneous input devices into a shared control stack with consistent kinematic treatment. The name "UniBiDex" reflects two core principles: universal device support (any input modality can be integrated) and unified constraint handling (all safety and task-related constraints are enforced through a single bimanual control module). The overall architecture consists of four decoupled modules: input preprocessing, motion retargeting, bimanual motion control, and haptic feedback.

Experimental Results

Kitchen-Tidying Task

We conducted a comprehensive user study with four participants performing a long-horizon bimanual manipulation task simulating a household kitchen-tidying routine. The task involves five sequential, contact-rich sub-tasks requiring dexterous coordination:

  1. Item Unpacking: Pick up assorted items from a cloth bag and place them on the table
  2. Shelf Organization: Sort and place items into designated positions in a multi-level kitchen shelf
  3. Towel Folding: Fold the towel neatly along a marked midline using both grippers
  4. Towel Placement: Place the folded towel onto the rack section of the shelf
  5. Clamp Attachment: Attach the clamp onto the rack to secure the towel in place

Performance Comparison

Method Overall Success Rate Completion Time (s) Improvement
UniBiDex (VR) 60% (24/40) 672 ± 20 +15% success, -18% time
Naive VR Baseline 45% (18/40) 816 ± 24 -
UniBiDex (Leader-Follower) 75% (30/40) 319 ± 8 +18% success, -5% time
Naive LF Baseline 57% (24/40) 335 ± 9 -

Citation

@article{li2025unibidex,
    title   = {UniBiDex: A Unified Teleoperation Framework for Robotic Bimanual Dexterous Manipulation},
    author  = {Zhongxuan Li, Zeliang Guo, Jun Hu, David Navarro-Alarcon, Jia Pan, Hongmin Wu, and Peng Zhou},
    year    = {2025},
    journal = {IEEE International Conference on Robotics and Biomimetics (IEEE ROBIO)},
    note    = {Under Review},
    url     = {https://github.com/Dieselmarble/UniBiDex}
}