WearSkill: Personalized and Interchangeable Input with Wearables for Users with Motor Impairments
Ovidiu-Andrei Schipor, Laura-Bianca Bilius, Radu-Daniel Vatavu · 2022 · Proceedings of the 19th International Web for All Conference (W4A) · doi:10.1145/3493612.3520455
Summary
This paper introduces WearSkill, a web-based application designed to provide personalized and interchangeable input for wearable computing, with a specific focus on users with motor impairments. The system addresses a fundamental challenge in wearable accessibility: most commercial wearables are not designed with motor accessibility in mind, and the type and severity of a user's motor condition determines which input modalities they can effectively use. WearSkill tackles this by allowing users to switch between different input methods across different wearable devices — for example, touch input on a smart ring, mid-air hand gestures on a smartwatch, or voice commands via smartglasses — based on their individual motor abilities and preferences. The application is built on the WISE framework, a set of recommendations for improving wearable accessibility that emphasizes exploring diverse wearable designs, new input modalities, more user studies, and extending interactions across devices. WearSkill's software architecture follows SOLID principles and uses the Euphoria middleware for event-driven communication between heterogeneous devices via HTTP and WebSocket protocols. The system comprises six components: user profile entry (motor symptoms), preference recommendations via machine learning, device registration, input detection and recognition, command mapping, and runtime monitoring. The front-end uses Vue.js while the back-end runs on Node.js, with over 200 end-to-end tests implemented using Cypress.
Key findings
The study evaluated WearSkill's recommendation engine with 21 participants who had motor impairments including spinal cord injuries, spina bifida, and traumatic brain injury, aged 28-59 years. Users self-reported eleven motor symptoms (slow movements, spasm, low strength, tremor, poor coordination, rapid fatigue, difficulty gripping, difficulty holding, lack of sensation, difficulty controlling direction, and difficulty controlling distance), which were used as features for machine learning classifiers. The best-performing classifiers — Decision Tree, Gradient Boosting, and Ada Boost with Random Forest — all achieved 85.3% mean accuracy in predicting users' preferred input modality and wearable device combinations using leave-one-out cross-validation. Performance varied significantly across device-modality pairs: smart ring touch input achieved 100% prediction accuracy, while smartglasses hand motion prediction was lowest at 57.1%. Voice input prediction was notably inconsistent, ranging from 57.1% to 95.2% depending on the device. The system supported three wearable types (smartwatch, smartglasses, smart ring) and four input modalities (touch, hand motion, head motion, voice), tested with Samsung Galaxy Watch 3, Gear Fit 2 (adapted as smart ring), and Vuzix Blade smartglasses.
Relevance
This research is directly relevant to practitioners working on accessible interaction design for wearable devices and IoT environments. The key practical takeaway is that personalized input recommendations based on self-reported motor symptoms can accurately match users' own preferences over 85% of the time, validating the ability-based design approach for wearables. The open architecture of WearSkill, built on web standards and event-driven middleware, provides a replicable model for developers building accessible multi-device systems. However, the study's sample size of 21 participants is small, and the binary classification of preferences (recommend vs. not recommend) may oversimplify nuanced user preferences. The work also highlights an important gap: voice input prediction accuracy was notably lower and more variable than other modalities, suggesting that voice interaction accessibility for people with motor impairments deserves further investigation. For organizations deploying wearable technology, this research reinforces the importance of offering multiple input modalities rather than assuming one-size-fits-all interaction.
Tags: wearable technology · motor impairments · personalization · input modalities · gesture input · voice input · touch input · machine learning · accessibility · ability-based design
Standards referenced: ISO/IEC 25010:2011 SQuaRE · WISE Framework