From Selfie Stick to Virtual Cane: Enabling Blind Exploration through Mobile Virtual Reality
Hao Tang, Hong Zhao, Xinpeng Liu, Zhenchao Xia, William Seiple, Zhigang Zhu · 2026 · Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems (CHI '26) · doi:10.1145/3772318.3790883
Summary
Consumer VR is a visual-first medium, which in practice shuts out blind and low-vision (BLV) users. Tang and colleagues argue that existing accessible VR systems do help BLV users explore virtual environments (VEs), but they depend on head-mounted displays, external tracking rigs, haptic gloves, or treadmills - all expensive, specialised, and not realistic for home or community use. The paper introduces Virtual Cane, a fully mobile audio-haptic VR system built on consumer-grade hardware (an iPhone, AirPods, and a cheap selfie stick) that lets BLV users physically walk through an empty room and use the phone as a simulated white cane to explore a 3D virtual scene. The system uses AR Foundation (ARKit) for six-degree-of-freedom tracking of the phone, AirPod gyroscope data for head orientation, and 3D spatial audio plus phone vibration for multimodal feedback. A novel dual-layered virtual object model produces pseudo-haptic feedback: each virtual object has an outer 'surface zone' that fires a hit sound + vibration on contact, and an inner 'alert zone' that hums to signal imminent penetration. This lets the system simulate the tactile sensation of boundaries without any mechanical brake. The app supports the three standard cane techniques (constant contact, two-point touch, shorelining) so BLV users can transfer real-world orientation-and-mobility skills directly. A mixed-methods user study with 13 BLV adults (ages 23-72, all cane users trained in O&M) compared exploration of a real room (RS), a 4.6x9.1m virtual office (VE), and a cardboard-box mock-up of that office (RMS).
Key findings
Participants could independently explore the VE using cane techniques and build transferable mental maps. In the VE exploration task they correctly identified 6.92/7 structural components and 7.77/8 objects, and in the real-world mock-up task (entered with no prior visit, using only the VE-derived mental map) they correctly identified 6.31/7 structural components and 5.38/9 objects. Cognitive mapping performance was not significantly different between real and virtual exploration (structural t=1.90, p=0.082). Object identification accuracy of 87.5% exceeded the 75% reported by prior HMD-based work, and exploration was roughly three times faster per square metre (0.26 vs 0.70 min/m²). The dual-layer model drove behaviour change: cane-sweeping angles dropped from 72.92° without collisions to 67.00° with the humming alert layer (p<.01), mirroring how real cane users narrow sweeps after contact. Immersion and realism ratings were strong (experienced realism 4.3, presence 5.2 on 7-point scales) and 12/13 participants said they would reuse the system. Two substantive challenges emerged: high cognitive load when parsing multi-step auditory instructions, and a skill-transfer conflict where participants with non-standard cane techniques (e.g., P3, P9) became disoriented because the system was tuned to canonical shorelining. Selfie-stick grip also varied between participants, suggesting stick-balance tuning is needed. The authors close with concrete design guidelines (landmarks/reset, phased goals, structured cane training, on-demand orientation queries, progressive O&M modules, and cane-motion tracking for training).
Relevance
This is one of the first VR accessibility papers that treats ubiquity as a first-class design constraint. If a BLV user already owns a smartphone and wireless headphones, a $10 selfie stick is all that separates them from a working VR pre-exploration tool. That radically lowers the barrier for indoor wayfinding previews, O&M training at home, virtual tours of public buildings (airports, train stations, museums), and even entertainment use cases - 11 of 13 participants wanted this for video games. The dual-layer pseudo-haptic pattern is an interesting design primitive that other accessible-VR researchers could adopt without needing specialised hardware. For practitioners, the strongest takeaway is that O&M skill transfer only works if the system accommodates the cane technique the user already has, rather than forcing a canonical version of it. Limitations are substantive: 13 participants, static indoor layouts, no stairs or moving obstacles, a 1:1 physical-to-virtual mapping that breaks down in cramped physical spaces, and no longitudinal evaluation of learning effects or retention. Outdoor scenes and crowded dynamic environments are still open problems.
Tags: blind and low vision · virtual reality · mobile accessibility · orientation and mobility · white cane · spatial audio · haptics · mental map · assistive technology · embodied interaction