Abstract
This paper introduces GEX, an innovative low-cost dexterous manipulation system that combines the GX11 tri-finger anthropomorphic hand (11 DoF) with the EX12 tri-finger exoskeleton glove (12 DoF), forming a closed-loop teleoperation framework through kinematic retargeting for high-fidelity control. Both components employ modular 3D-printed finger designs, achieving ultra-low manufacturing costs while maintaining full actuation capabilities. Departing from conventional tendon-driven or underactuated approaches, our electromechanical system integrates independent joint motors across all 23 DoF, ensuring complete state observability and accurate kinematic modeling. This full-actuation architecture enables precise bidirectional kinematic calculations, substantially enhancing kinematic retargeting fidelity between the exoskeleton and robotic hand. The proposed system bridges the cost-performance gap in dexterous manipulation research, providing an accessible platform for acquiring high-quality demonstration data to advance embodied AI and dexterous robotic skill transfer learning.
Introduction
While AI has mastered complex robotic locomotion, real-world dexterous manipulation—the ability to use hands with human-like precision—remains a significant bottleneck. The primary barrier is hardware accessibility. High-performance robotic hands, such as the Shadow Hand, can cost over $100,000 and are notoriously difficult to maintain due to complex tendon-driven systems. Even "affordable" alternatives like the Allegro Hand retail for over $16,000. GEX aims to democratize this field by providing a robust, anthropomorphic system that costs approximately $1,200—roughly 1/20th of its competitors—without sacrificing performance.
System Design and Architecture
The GEX system consists of two primary hardware components: the GX11 robotic hand and the EX12 exoskeleton glove, unified by a custom software stack called libgex.
GX11 Dexterous Hand
The GX11 is an 11-Degree-of-Freedom (DoF) hand featuring a 3-DoF thumb and two 4-DoF fingers. Unlike many robotic hands that use tendons or underactuated joints, the GX11 is fully actuated. Every joint is equipped with its own independent Dynamixel XL330 motor, ensuring complete state observability and precise control. Most structural components are 3D-printed using ABS plastic, allowing for rapid assembly and low-cost maintenance.
EX12 Exoskeleton Glove
The EX12 is a 12-DoF wearable glove designed for intuitive human teleoperation. It supports bidirectional force feedback, allowing the user to "feel" when the robotic hand makes contact with an object. By utilizing motors with lower reduction ratios to minimize friction, the glove provides a natural tactile experience essential for high-quality data collection.
Kinematic Retargeting and Workspace
A core innovation of the GEX system is its kinematic retargeting algorithm. This software maps human finger movements captured by the EX12 glove to the GX11’s joint space in real-time. Because every joint in the GX11 is directly driven, the system can perform precise bidirectional kinematic calculations, ensuring high-fidelity motion transfer with minimal jitter.
As shown in the workspace analysis, the fully-actuated architecture allows for a broad range of motion across all three fingers, enabling the hand to perform complex manipulation tasks that require fine-grained coordination.
Key Results and Performance
The GEX system demonstrates that ultra-low-cost hardware can achieve high-precision results:
- Cost-Efficiency: The total build cost is ~$1,200 ($600 for the hand and $600 for the glove).
- Precision: Integrated 12-bit absolute magnetic encoders provide a resolution of 0.088°, allowing for extremely fine motor control.
- Task Success: The system has successfully performed representative dexterous tasks, including grasping fragile paper cups, reorienting composite objects, and performing controlled twisting movements.
Conclusion and Future Work
GEX provides an accessible, open-source platform for dexterous manipulation research. By replacing expensive mechanical complexity with modular 3D-printed parts and full actuation, it enables any laboratory to collect high-quality demonstration data for imitation learning. Future work will focus on integrating dedicated tactile sensors for delicate manipulation and exploring autonomous control algorithms using the GEX platform.
Summary updated