Localization of Vibrotactile Stimuli on the Face

TL;DR

This work explores using the face, specifically the cheek, as a practical channel for vibrotactile haptic feedback in contexts where the hands are unsuitable. It employs a 4×3 grid of coin ERM motors delivering discrete 200 ms stimuli to 12 cheek sites, tested across 10 participants to gauge localization accuracy. Findings indicate feasible localization with a notable lateral bias toward the ear and higher confusion for middle-column sites, though overall comfort was reported; results inform design considerations for compact, socially acceptable facial haptic devices. The study lays groundwork for optimizing site count and encoding schemes (including spatiotemporal patterns) to enable robust facial haptic feedback for prosthetics and assistive technologies.

Abstract

The face remains relatively unexplored as a target region for haptic feedback, despite providing a considerable surface area consisting of highly sensitive skin. There are promising applications for facial haptic feedback, especially in cases of severe upper limb loss or spinal cord injury, where the face is typically less impacted than other body parts. Moreover, the neural representation of the face is adjacent to that of the hand, and phantom maps have been discovered between the fingertips and the cheeks. However, there is a dearth of compact devices for facial haptic feedback, and vibrotactile stimulation, a common modality of haptic feedback, has not been characterized for localization acuity on the face. We performed a localization experiment on the cheek, with an arrangement of off-the-shelf coin vibration motors. The study follows the methods of prior work studying other skin regions, in which participants attempt to identify the sites of discrete vibrotactile stimuli. We intend for our results to inform the future development of systems using vibrotactile feedback to convey information via the face.

Figures (5)

• Figure 1: Left: the array of actuators on a participant's cheek; Right: the interface shown to participants for collecting responses.
• Figure 2: Confusion matrices showing correct responses along the diagonal, and misclassifications off the diagonal.
• Figure 3: The number of correct responses per site, for each participant. L-R is medial-lateral in each grid.
• Figure 4: The directionality of errors by each participant; errors along columns and rows shown separately.
• Figure 5: The frequencies of responses for each stimulus site, aggregated over participants. L-R is medial-lateral.