NavCog: A Navigational Cognitive Assistant for the Blind
Dragan Ahmetovic, Cole Gleason, Chengxiong Ruan, Kris Kitani, Hironobu Takagi, Chieko Asakawa · 2016 · Proceedings of the 18th International Conference on Human-Computer Interaction with Mobile Devices and Services (MobileHCI '16) · doi:10.1145/2935334.2935361
Summary
This paper introduces NavCog, an iPhone-based turn-by-turn navigation assistant designed to guide blind users through unfamiliar indoor and outdoor environments without requiring expensive infrastructure or structural modifications. The authors argue that while blind travellers can routinely handle familiar routes by drawing on mental maps built up with orientation-and-mobility training, new or rarely visited spaces remain difficult because most environmental cues (signs, distant landmarks) are visual. GPS is too coarse indoors (errors of tens of metres) and even outdoors struggles to distinguish adjacent doors or corridors, so the authors instead rely on a network of Bluetooth Low Energy (BLE) beacons stuck to walls with velcro and an RSSI-fingerprinting approach trained per-edge with a K-nearest-neighbour model. NavCog has three components: a web-based map-authoring tool that lets operators draw a graph of nodes (destinations, transitions such as elevators, POIs) and walkable edges over a floorplan; the beacon network itself; and an iOS app tightly integrated with VoiceOver that downloads the map and performs all localisation on-device without any network connection. The user interface is deliberately stripped down — four large corner-of-screen buttons (Stop, Previous Instruction, Accessibility Info, Surrounding Info) — and instructions can be delivered as synthesised speech or as non-verbal clicking sounds whose pitch rises as the user approaches the next waypoint.
Key findings
In the localisation evaluation, collecting 12 RSSI samples per fingerprint point across 24 beacons along a 16 m corridor produced a mean error of 0.53 m and a maximum error of 2.5 m, with 83% of test positions localised within 1 m of ground truth. Even a single sample per point yielded 0.85 m average error and 67% within-1 m accuracy — usable on long straight corridors but not at decision points. The authors recommend 12 samples per fingerprint as a practical balance between deployment workload and precision. The user study deployed NavCog over 530 m of paths across a Carnegie Mellon campus (two buildings, two bridges, a stairway, and an outdoor quad) and evaluated it with six blind participants. Participants recovered unaided from 66 of 76 observed navigation errors; only 10 events required experimenter intervention. Missed turns were the most common problem (25 events), often in the outdoor quad where beacons could be placed on only one side of the path, and over-turns occurred when users rotated past the correct bearing. Participants reported satisfaction with the core turn-by-turn paradigm and particularly valued the POI descriptions about artwork, history, and restrooms.
Relevance
NavCog is a landmark demonstration that accurate indoor wayfinding for blind users can be delivered on commodity smartphones using inexpensive, non-invasive infrastructure — an important contrast to systems that demand custom hardware, RFID-embedded floors, or building-wide Wi-Fi retrofits. For accessibility practitioners in venues such as airports, campuses, hospitals, and transit stations, the paper offers a concrete blueprint: graph-based route maps, RSSI fingerprinting, and an audio-first interface that supplements rather than replaces the white cane or guide dog. The authoring tools also make the system feasible for small accessibility teams to deploy, and the POI and accessibility-alert features point beyond pure navigation to richer environmental awareness. Limitations include the open-area veering problem (straight-edge maps don't model plazas well), a small participant pool, and dependence on beacon maintenance. NavCog seeded the broader CMU/IBM line of work that later produced the BBeep collision-avoidance system and the AI Suitcase.
Tags: blind navigation · indoor navigation · wayfinding · orientation and mobility · assistive technology · mobile accessibility · turn-by-turn navigation · bluetooth low energy · localization · points of interest · blindness and low vision