Automatic Task Assistance for People with Cognitive Disabilities in Brushing Teeth - A User Study with the TEBRA System
Christian Peters, Thomas Hermann, Sven Wachsmuth, Jesse Hoey · 2014 · ACM Transactions on Accessible Computing · doi:10.1145/2579700
Summary
This paper presents TEBRA (TEeth BRushing Assistance), an Assistive Technology for Cognition (ATC) system that supports people with moderate cognitive disabilities in brushing their teeth independently. The research addresses a fundamental challenge: people with cognitive disabilities such as dementia, intellectual disabilities, and autism often struggle with "task sequencing"—coordinating the substeps of Activities of Daily Living (ADLs) in the proper order. This dependency on human caregivers reduces independence and places significant burden on care providers. TEBRA uses a context-aware approach that infers what the user is doing without requiring explicit feedback. The system tracks user behavior through environmental sensors: two cameras (overhead and frontal), a water flow sensor at the tap, and a sensor-equipped electric toothbrush with gyroscope and accelerometer. The brushing task is decomposed into eight behaviors: paste_on_brush, rinse_mug_fill, rinse_mouth_wet, brush_teeth, rinse_mouth_clean, rinse_mug_clean, clean_brush, and use_towel. A key technical challenge is handling the variance in how people with cognitive disabilities perform tasks. The system addresses "spatial variance" (different movement characteristics between users) through a Bayesian network classifier that infers behaviors from object states rather than tracking specific movements. It handles "temporal variance" (different execution speeds) through a dynamic timing model that categorizes users into fast, medium, or slow velocity categories and adapts timing thresholds accordingly. When the system detects inconsistent behavior, it delivers prompts through a two-level escalation hierarchy: first a pictogram with audio command, then a real-life video with audio if the user doesn't respond.
Key findings
The study evaluated TEBRA with seven participants (ages 41-56) with moderate cognitive disabilities living in a German residential care home. Using an AB single-subject design, researchers compared performance with caregiver assistance (baseline) versus TEBRA system assistance (intervention) across 55 trials over nine days. The results showed a statistically significant increase in independent task steps when using TEBRA (p = 3.5 × 10^-9, Mann-Whitney U-test). With caregiver assistance, participants completed an average of 2.7 steps independently; with TEBRA, this increased to 7 steps on average. All users showed improvement, though the magnitude varied—one user achieved perfect independence (all 8 steps) in every TEBRA trial, while another showed a smaller but still meaningful increase of 1.6 additional independent steps. The behavior recognition component achieved an average classification rate of 69.3%, with rates varying from 97.8% for paste_on_brush to 41% for rinse_mouth_clean. The dynamic timing model successfully adapted to individual user velocities, enabling appropriate prompt timing. Users rated the system's helpfulness at 4.1/5 and acceptance at 4.5/5, with some users treating TEBRA as an interaction partner—responding verbally to prompts with "ok" or "I will." Caregivers rated helpfulness at 3.8/5, validating the system from a professional perspective.
Relevance
This research demonstrates that technology can meaningfully increase independence for people with cognitive disabilities in essential daily activities. The findings have practical implications for residential care settings, where the high ratio of caregivers to residents often means people receive less personalized assistance than they need. The design methodology is instructive for accessibility practitioners. The researchers used Interaction Unit (IU) analysis to systematically decompose the brushing task through observation of real caregiver-assisted sessions, rather than relying on assumptions about how the task should be performed. Caregiver interviews informed prompt design, revealing that short audio commands paired with pictograms work best for most users, while some need the richer information in video prompts. Several limitations warrant consideration. The small sample size (n=7) and short study duration (five weeks) limit generalizability. The behavior recognition rate of 69.3% means roughly one-third of behaviors are misclassified, leading to some inappropriate prompts. The study also acknowledges a trade-off: while TEBRA reduced caregiver physical intervention, it also reduced social interaction during the task—a reminder that independence and social connection are both important for wellbeing. Future work aims to extend TEBRA to other ADLs like hand washing and shaving.
Tags: cognitive disability · assistive technology · activities of daily living · dementia · intellectual disability · autism · behavior recognition · context awareness · task assistance
Standards referenced: ICF