Get In Touch with Your Seat: Accessible Seat Localization for Blind and Low Vision Travelers in Autonomous Shuttles
Paul D. S. Fink, Justin R. Brown, Raina M. Movalia, Kyle J. James, Margaret E. Kastelein, Jacob Bond, Morgan E. Andrulis, Nicholas A. Giudice · 2026 · Extended Abstracts of the 2026 CHI Conference on Human Factors in Computing Systems (CHI EA '26) · doi:10.1145/3772363.3798714
Summary
Fink, Brown, Movalia, and colleagues (University of Maine VEMI Lab, Grand Valley State University, and General Motors Global Research and Development) tackle a segment of the autonomous-vehicle accessibility problem that has received little attention: helping a blind or low-vision (BLV) traveller find their assigned or available seat after entering a driverless shuttle, where there is no driver or human operator to offer guidance. They compare two interaction conditions in a within-subjects study: a traditional natural-language (NL) audio-only description delivered via smartphone (analogous to how VoiceOver, TalkBack, and JAWS currently present spatial information), and a multisensory condition combining the same NL descriptions with a prototype Vibro-Audio Map (VAM) rendered on the phone's touchscreen plus spatialised environmental audio chimes from the entrance and the target seat. The VAM represents the entrance, bench lines, and target seat through specific vibration patterns and paired audio cues (continuous vibration for bench lines, pulsed 4 Hz for the target seat, pulsed 7 Hz for the entrance). Eight legally blind participants (aged 23-54, M=37.63) each completed the seat-finding task in a physical AV cabin simulator with six 'campfire' seats arranged in a semicircle, using a Wizard-of-Oz iPhone-controlled delivery. After each condition they rated clarity, helpfulness, and timing on 5-point Likert scales, completed the System Usability Scale (SUS), and participated in a structured interview.
Key findings
All eight participants (100%) identified the correct seat in the multisensory condition; only four of eight (50%) did so with NL audio alone — the remaining four incorrectly sat on the wrong side of the AV cabin. Seat-finding was dramatically faster in the multisensory condition (mean 18.13 s, SD 13.73) than in NL-only (mean 52.38 s, SD 37.07), a statistically significant difference (t(7)=2.53, p=.04, d=.89). The NL condition generated more requests for repeated information (4 of 8 participants vs 2 of 8 in multisensory). SUS scores trended higher for the multisensory condition (87.5 vs 82.2, not significant but grading 'A+' vs 'A' on Lewis and Sauro benchmarks). Seven of eight participants preferred the multisensory condition overall. Qualitative themes: participants valued haptic feedback as a confidence-building cross-check against spoken descriptions ('left most seat' is hard to verify without a map); participants disliked information overload and poorly timed cues; and participants asked for fine-grained user control to toggle specific information channels on and off ('I would say like every single thing that could possibly give you information, make sure that there's an on and off switch for every single piece'). The paper thus challenges the common assumption that BLV travellers prefer audio-only interfaces: when designed well, multisensory interfaces were both faster and more accurate while maintaining or improving perceived usability.
Relevance
Seat localisation is one piece of a longer BLV travel pipeline (wayfinding to the vehicle, door identification, entry, seat-finding, in-ride awareness, exit), and this study closes a specific gap: AV rideshares and shuttles remove the human driver who previously offered informal guidance, so BLV users either fall back on audio-only descriptions that are known to be error-prone for layout tasks, or they travel with a sighted companion. The Vibro-Audio Map prototype demonstrates that consumer smartphones — without any additional hardware — can deliver layout information accurately enough to support independent seat selection, which has direct implications for accessible AV design, airport seating kiosks, event-venue seat finders, and any indoor navigation app where the layout fits on a phone screen. The user-control theme also generalises: accessible information should be toggleable at fine granularity, not delivered as a monolithic audio stream. Limitations are straightforward: N=8, a physical but static simulator with six seats, Wizard-of-Oz rather than a deployed app, and no evaluation of the full trip. Nonetheless, this is the first study to experimentally compare a Vibro-Audio Map against the natural-language description logic previously proposed for BLV access to AVs, and it lays the foundation for on-road testing.
Tags: autonomous vehicles · blind and low vision · transportation accessibility · haptics · spatial audio · vibro-audio maps · multimodal interaction · orientation and mobility · shared mobility · multisensory interface