GEX: Democratizing Dexterity with Fully-Actuated Dexterous Hand and Exoskeleton Glove
Yunlong Dong, Xing Liu, Jun Wan, Zelin Deng · 2025
Read GEX: Democratizing Dexterity with Fully-Actuated Dexterous Hand and Exoskeleton Glove by Yunlong Dong, Xing Liu, Jun Wan, Zelin Deng (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2506-04982.txt.
Introduction
Yunlong Dong, Xing Liu, Jun Wan, and Zelin Deng present GEX, a low-cost system pairing a robot hand with an exoskeleton glove for teleoperation, in a 2025 paper. Skilled robot hands are usually pricey, which limits the demonstration data that learning systems need. Their hand has three fingers and eleven driven joints; the glove has three fingers and twelve.
Both use modular printed fingers and a motor at every joint, 23 in all, avoiding the cables and partial actuation of cheaper designs. Full actuation makes the glove-to-hand motion mapping exact in both directions. The authors offer the pair as an affordable way to gather high-quality demonstrations.
Catalog record
| CREATOR | Yunlong Dong, Xing Liu, Jun Wan, Zelin Deng |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 15 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2506-04982
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
GEX: Democratizing Dexterity with Fully-Actuated Dexterous Hand and Exoskeleton Glove Yunlong Dong, Xing Liu, Jun Wan, Zelin Deng 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. The code is published in 1 1 1 https://github.com/orgs/Democratizing-Dexterous/repositories . I Introduction Hand dexterity is fundamental to human cognition, enabling active manipulation, tool use, and the way we learn from our environment. Replicating this level of dexterity in robotic systems remains a significant challenge [ 1 ] . Although machine learning has driven notable advances in robotic locomotion across varied terrains, real‐world manipulation has largely been limited to simple, single‐degree‐of‐freedom parallel‐jaw grippers [ 2 ] . Truly dexterous in‐hand manipulation has been investigated primarily in simulation, with only a handful of demonstrations on physical hardware [ 3 ] . A major barrier to widespread research in dexterous manipulation is the lack of affordable, robust robotic hands [ 4 ] . High‐performance tendon‐driven designs such as the Shadow Hand achieve human‐like capabilities but carry price tags well over six figures and require substantial maintenance due to their mechanical complexity. For many researchers, direct‐drive hands offer a more practical alternative. The Allegro Hand, in particular, has seen broad adoption: however, it remains expensive—retailing for over $16,000. Consequently, only a small number of laboratories can access hardware capable of complex dexterous tasks. In contrast, the ubiquity of affordable two‐finger grippers and legged locomotion platforms has accelerated progress by enabling easy replication and iterative improvement. To democratize research in learned dexterous manipulation, a robotic hand platform must be durable, repeatable, low cost, versatile, and anthropomorphic, thus lowering the barrier for broader community engagement. We introduce the GEX Series, a general‐purpose, low‐cost robotic dexterous hand with 11 DoF and a complementary force‐feedback exoskeleton glove, both constructed from off‐the‐shelf or 3D‐printed parts as robust platforms for robot learning. Each device can be assembled in under four hours for $600—making the hand 1/20 the cost of the Allegro Hand and 1/125 that of the ShadowHand—while the glove likewise takes only hours to build for under $600. While robustness and low cost are important, they should not come at the expense of dexterity and anthropomorphism. We believe that a truly versatile system must combine a nimble, human‐like hand with an equally responsive teleoperation interface. The human hand can adeptly manipulate a vast array of tools and objects; accordingly, the GX11 hand is designed with an anthropomorphic three‐finger structure that maximizes flexibility while closely mirroring human kinematics. Experimental results confirm that our three‐finger dexterous hand achieves manipulation capabilities. Efficiently collecting high-quality demonstrations remains a persistent challenge in robotic learning. Teleoperation is a common solution, but the high DoF and intrinsic complexity of dexterous hands impose stringent demands on control precision and motion-capture fidelity [ 5 ] . To address this, our complementary EX12 exoskeleton glove to deliver intuitive, high-fidelity capture of human finger motions, enabling ultra-precise remote operation of the GX11 hand and seamless transfer of demonstrations to the robot platform. All mechanical designs, embedded firmware, assembly instructions, URDF models, and dexterous retargeting algorithm are released open‐source. II GEX Hardware II-A Dexterous Hand Building upon the Leap Hand [ 4 ] design, We present a redesigned three-finger anthropomorphic fully-driven dexterous hand named GX11. The GX11 hand adopts the same direct joint drive method as the Leap Hand, featuring a 3 DoF structure for the thumb while employing identical 4 DoF configurations for both the index and middle fingers, collectively forming a total of 11 DoF. This innovative design not only simplifies the mechanical architecture but also achieves significant reductions in overall weight and physical dimensions. The joint configuration and physical diagram of the GX11 hand are presented in Fig. 1 and Fig. 2 . The motion space of the fingertip positions for the three fingers of the GX11 hand is shown in Fig. 4 . Fig. 1 : The joint configuration of the GX11 hand. Fig. 2 : The physical diagram of the GX11 hand. (a) Thumb finger (b) Index finger (c) Middle finger Fig. 3 : The motion space of the fingertip positions for the three fingers of the GX11 hand. (a) Thumb finger (b) Index finger (c) Middle finger Fig. 4 : The motion space of the fingertip positions for the three fingers of the EX12 exoskeleton glove. The GX11’s joint actuation employs Dynamixel XL330-M288-T, a high-performance, cost-effective servo actuator. These XL330-M288-T units deliver 0.53Nm rated torque, providing sufficient power density for driving all phalangeal joints. The integrated 12-bit absolute magnetic encoder (AS5601) ensures high-precision position feedback with 0.088° resolution, guaranteeing motion accuracy across all articulations. Through a 288:1 reduction ratio, the actuators achieve enhanced torque amplification while maintaining operational stability and reliability during diverse manipulation tasks. Operating at 5V DC, the motor system utilizes an ARM Cortex-M0+ microcontroller for real-time control, implementing TTL Multidrop Bus communication protocol that simplifies circuit architecture while improving system scalability and hardware compatibility. The actuators support multiple control modalities including current control mode, velocity control mode and position control mode. This multi-modal flexibility enables adaptive parameter tuning for various operational requirements. With a total mass of merely 18g and compact dimensions (24.5 × 33 × 25 mm), the optimized form factor permits seamless integration into the anthropomorphic joint design without compromising dexterity. II-B Exoskeleton Glove The EX12 exoskeleton glove adopts Dynamixel XL330-M077-T motor, which has a smaller reduction ratio (77:1) to reduce the friction from the gear box during human finger motion. Considering the high DoF of the human thumb, the EX12 exoskeleton glove incorporates a 4-DoF thumb design to better accommodate the comfort requirements of human teleoperation. The other two fingers maintain the same 4-DoF configuration as the GX11 hand. The joint configuration and physical diagram of the EX12 exoskeleton glove are presented in Fig. 5 and Fig. 6 . The motion space of the fingertip positions for the three fingers of the EX12 exoskeleton glove is shown in Fig. 4 . In the GX11 dexterous hand and EX12 exoskeleton glove, all structural components except the joint motors are fabricated using acrylonitrile butadiene styrene (ABS) material through 3D printing. ABS demonstrates notable advantages including cost-effectiveness, high mechanical strength, and excellent wear resistance, making it particularly suitable for manufacturing structural components of dexterous hands. The adoption of 3D printing structural components exhibit simplified assembly procedures and maintenance accessibility, thereby significantly reducing both overall manufacturing costs and maintenance complexity. Fig. 5 : The joint configuration of the EX12 exoskeleton glove. Fig. 6 : The physical demonstration of the EX12 exoskeleton glove worn on a