ART-Glove: Articulated Tactile Glove for Contact-Grounded Dexterous Interaction Capture
ART-Glove integrates 16 rigid functional surfaces and 22-DoF joint sensing with 2048-taxel tactile arrays, enabling contact-grounded dexterous demonstration capture at 120Hz.
Key Findings
Methodology
ART-Glove employs 16 rigid functional surfaces covering fingers, thumb, and palm, connected via 22 anatomically aligned joints. Encoder-based sensors track joint angles, while dense piezoresistive tactile arrays record contact signals. The system achieves synchronized data collection at 120Hz, capturing 22-DoF joint angles and 2048 tactile taxels. Mechanical design ensures high dexterity, with adaptive joint mechanisms addressing complex hand geometries. The hardware layout balances natural human motion preservation with precise contact and motion tracking, enabling detailed demonstration data for robotic learning.
Key Results
- The glove provides finger flexion ranges up to 70°-88°, with joint angle errors below 1.4°, maximum 2.4°. Tactile sensors reliably detect contact shape, force, and distributed contact patterns. During multi-finger tasks like rotation, screwing, and pressing, the system captures synchronized surface motion and tactile signals at 120Hz, demonstrating high fidelity in contact-rich scenarios. Experimental data shows accurate surface geometry, surface motion, and tactile response, validating the system’s capability for detailed dexterous demonstration collection.
- In quantitative evaluations, joint angle sensing achieves mean absolute errors of less than 0.5°, with high consistency across different joint types. Tactile responses accurately distinguish contact shapes, from localized forces to distributed grasp patterns. The system successfully records complex manipulation tasks, with contact location and force distribution captured in real-time. These results confirm the system’s robustness, high spatial resolution, and suitability for downstream robot learning and dexterous manipulation tasks.
- The integrated capture of motion and tactile data enables detailed modeling of contact-grounded interactions. The system’s high sampling rate and multi-modal sensing facilitate precise analysis of contact dynamics, supporting advanced robot imitation and reinforcement learning. Demonstrations include object rotation, screwing, and pressing, where the system captures full surface trajectories and tactile signals, providing comprehensive datasets for developing dexterous robotic policies. The results highlight the potential for real-world application in robotic skill acquisition and human-robot collaboration.
Significance
This work advances the state-of-the-art in human demonstration capture by combining explicit contact geometry, high-density tactile sensing, and natural hand motion preservation. It addresses critical limitations of prior methods—such as unreliable visual inference and limited haptic feedback—by providing a reliable, high-fidelity data source. The system’s ability to record contact-grounded information at high temporal resolution opens new avenues for robot learning, especially in contact-rich tasks like assembly, manipulation, and virtual prototyping. It bridges the gap between human dexterity and robotic perception, enabling more effective transfer of skills and behaviors from humans to robots, with broad implications for automation, teleoperation, and assistive robotics.
Technical Contribution
The paper introduces a novel hardware architecture integrating 16 rigid functional surfaces with 22-DoF anatomical joints, coupled with dense piezoresistive tactile arrays. The design employs adaptive joint mechanisms—shaft-sleeve, shaft-bearing, and arc-slot—to realize complex hand geometries while maintaining mechanical simplicity. The system achieves synchronized 120Hz data acquisition of joint angles and tactile signals, with high spatial resolution. This comprehensive integration of rigid contact surfaces, multi-DoF articulation, and dense tactile sensing represents a significant leap over existing soft gloves or vision-based methods, enabling reliable, contact-grounded demonstration capture for robotic learning.
Novelty
This is the first system to combine explicit rigid functional surfaces with dense tactile sensing in a high-DoF articulated glove, providing accurate contact geometry and synchronized surface motion data. Unlike prior soft gloves or visual inference approaches, ART-Glove offers direct, reliable contact measurement while preserving natural dexterity. Its innovative joint mechanisms and sensor integration address complex hand geometries and motion constraints, setting a new standard for contact-grounded demonstration data collection. This dual focus on explicit contact geometry and high-density tactile feedback distinguishes it from existing methods, opening new possibilities for dexterous robot learning.
Limitations
- The mechanical complexity and cost of the system limit large-scale deployment. The rigid surfaces and joint mechanisms, while precise, add bulk and may restrict certain extreme motions. Tactile sensors may experience signal drift or noise under high-pressure contact. The current design focuses on static contact states; dynamic contact sensing remains challenging. Future work should aim to simplify hardware, reduce costs, and improve robustness under dynamic conditions.
Future Work
Future efforts will focus on miniaturizing the hardware, integrating more advanced sensing modalities such as force and slip detection, and developing deep learning models for automatic interpretation of tactile and motion data. Enhancing system robustness for dynamic tasks and reducing manufacturing costs are also priorities. Additionally, extending the system to multi-user scenarios and integrating it with autonomous control algorithms will broaden its application in robotic skill learning, teleoperation, and assistive technologies.
AI Executive Summary
Deep Dive
Plain Language Accessible to non-experts
想象你在厨房里做菜，你的手非常灵巧，可以抓住各种不同形状的东西，比如瓶子、刀子、菜板。为了教会机器人帮你做菜，你需要让它知道你手指怎么弯、手掌怎么转，还要知道你碰到的东西的形状和用力大小。以前的方法就像用模糊的图片或只看视频，难以捕捉这些细节。而ART-Glove就像给手戴上了带有传感器的特殊手套，不仅能追踪手指的弯曲，还能感受到你碰到的东西的形状和压力。这样，机器人就能学会像人一样灵巧地操作各种厨房用具。它的设计让你的动作自然流畅，又能详细记录接触的细节，为机器人学习提供了宝贵的“示范”。
ELI14 Explained like you're 14
想象你在玩一个超级酷的游戏，你用手指操作虚拟世界里的东西。普通的手套只能让你看见手在屏幕上的动作，但不能告诉你手碰到了什么或者用多大力。ART-Glove就像给你的手装上了高科技的“感应手套”，它能同时知道你的手指弯了多少、手在动，还能感受到你碰到的东西的形状和压力。这样，机器人就能学会你做的那些复杂动作，比如拧瓶盖、按按钮，甚至用手指调节东西。它的设计特别聪明，既让你的动作看起来自然，又能详细记录每一个接触细节，帮助机器人变得更聪明、更灵活。未来，这种技术还能用在虚拟现实、远程操控甚至康复治疗中，带来很多惊喜！
Abstract
We present ART-Glove, an articulated tactile glove designed to capture contact-grounded dexterous demonstrations while preserving human dexterity. ART-Glove makes hand-side contact geometry explicit with 16 rigid functional surfaces covering the fingers, thumb, and palm. Twenty-two anatomically aligned joints connect these surfaces and allow them to follow human hand motion during dexterous manipulation. Encoder-based sensing tracks surface motion, while dense piezoresistive tactile sensing records contact over the same surfaces. The complete system captures synchronized 22-DoF joint measurements and 2048-taxel tactile measurements at 120 Hz. We evaluate ART-Glove across experiments on motion freedom, joint sensing, tactile sensing, and contact-rich interaction capture, demonstrating its ability to preserve human dexterity while recording contact-grounded information that can support downstream dexterous robot learning.