Mind the Gaze: Improving the Usability of Dwell Input by Adapting Gaze Targets Based on Viewing Distance

TL;DR

<3-5 sentence high-level summary> GAUI tackles the problem of distance-induced performance degradation in dwell-based input on handheld devices by dynamically resizing targets to maintain a consistent visual angle. The authors implement a distance-aware, gaze-driven media player and validate it through a two-phase study: a controlled lab phase showing performance and preference benefits over static UIs, and a mobile-posture phase assessing real-world use with Experience Sampling Method. Key contributions include empirical evidence that distance-adaptive targets can improve task time, navigation efficiency, and error rates at longer distances, plus design guidelines for context-aware gaze interfaces on mobile devices. The work also reveals posture and movement effects, highlighting the need for personalization and mode-switching to optimize GAUI in everyday use.

Abstract

Dwell input shows promise for handheld mobile contexts, but its performance is impacted by target size and viewing distance. While fixed target sizes suffice in static setups, in mobile settings, frequent posture changes alter viewing distances, which in turn distort perceived size and hinder dwell performance. We address this through GAUI, a Gaze-based Adaptive User Interface that dynamically resizes targets to maximise performance at the given viewing distance. In a two-phased study (N=24), GAUI leveraged the strengths of its distance-responsive design, outperforming the large UI static baseline in task time, and being less error-prone than the small UI static baseline. It was rated the most preferred interface overall. Participants reflected on using GAUI in six different postures. We discuss how their experience is impacted by posture, and propose guidelines for designing context-aware adaptive UIs for dwell interfaces on handheld mobile devices that maximise performance.

Figures (8)

• Figure 1: In our experiment, we used four different interface sizes. Besides the three static interfaces shown above, we also created an adaptive interface. The adaptive interface adjusts according to the distance between the participant and the phone to maintain a visual angle of approximately 4° across all distances. At the closest distance, the interface switches to the static-small. As the distance increases, the interface size scales up. We calculated the number of soundtracks that could be displayed on each screen based on the size of the interface elements.
• Figure 2: In Phase 2, participants interacted with GAUI in six different postures and responded to in-app questions (E), triggered once an adaptation occurred to reflect on their experience with the adaptive UI in the mobile context. The figure shows a participant exploring the media player while (A) slouching, (B) sitting (desk), (C) sitting (hands-free), and (D) walking.
• Figure 3: Before the experiment, participants responded to questions using a scale (1 = Never, 2 = Rarely, 3 = Sometimes, 4 = Often, 5 = Always) regarding how frequently they used their phones in various postures. This indicates that participants were familiar with using their phones across various postures, making them suitable for evaluating our interface in realistic contexts.
• Figure 4: Mean task time by levels of Interface Types, Face-to-Screen Distance and Task Difficulty. For the hard tasks, both Interface Type and Face-to-Screen Distance showed a significant main effect on task time. Error bars represent standard deviation. The adaptive and static-small resulted in significantly faster task time than the static-large UI for the hard task. As expected, hard tasks took longer than easy ones across all conditions.
• Figure 5: Mean navigation time by levels of Interface Type and Face-to-Screen Distance. At the farthest distance between the participants and the phone, the adaptive and static-large interfaces significantly reduced navigation time compared to the static-medium and static-small interfaces. Error bars represent standard deviations.