How Context and User Behavior Affect Indoor Navigation Assistance for Blind People
João Guerreiro, Eshed Ohn-Bar, Dragan Ahmetovic, Kris Kitani, Chieko Asakawa · 2018 · Proceedings of the 15th International Web for All Conference (W4A 2018) · doi:10.1145/3192714.3192829
Summary
This short paper presents a fine-grained thematic analysis of user behaviour during indoor assisted navigation to understand when and why navigation errors occur even when the underlying localisation system is accurate. The research uses NavCog, a smartphone-based indoor navigation system that combines BLE beacon fingerprinting with a pedestrian motion model, achieving an average localisation error under 1.65 metres across three instrumented buildings at Carnegie Mellon University covering 58,800 m². NavCog provides turn-by-turn guidance with distance countdowns, "approaching" warnings before turns, and vibration/sound confirmation upon completing a turn. The study analyses video recordings from two user studies: the first with 13 blind participants (6 guide dog users, 7 white cane users) on routes of 200-690 feet with 4-11 turns, and the second with 8 blind white cane users on routes of 250-500 feet with 7-8 turns. Both first-person (participant) and third-person (researcher) GoPro video perspectives were analysed. Rather than reporting aggregate metrics like task completion time or error rates, the paper focuses on identifying the specific situational contexts and user behaviours that lead to navigation errors — information that navigation system designers can use to build more adaptive and resilient interfaces.
Key findings
Six problematic scenarios were identified. (1) Close sequence of instructions: when two turns are in close spatial and temporal proximity (e.g., "turn left, proceed 10 feet, turn right"), faster users (especially guide dog users) missed the second turn. The recommendation is to combine close instructions into a single compound instruction ("turn right, and after 10 feet turn left"). (2) Approaching turns: the "approaching" warning was highly valued, but problems arose when multiple alternative paths existed near a turn. Guide dogs sometimes stopped slightly before the announced turn point, and close sequences of intersections could cause premature turns. (3) Error recovery: guide dog users rarely needed to recover from errors because the dog follows the correct path even when instructions are slightly late. White cane users recovered quickly in narrow corridors by sweeping the cane to find openings, but struggled in wider corridors and open spaces where there were no walls to reference. The system should detect erroneous behaviour (e.g., going back and forth) and intervene sooner rather than waiting for a distance threshold. (4) Slight turns: instructions for turns between 30 and 60 degrees caused over-turning for white cane users, who tended to execute approximately 90-degree turns regardless. Guide dog users handled slight turns well. Additional feedback is needed when over-turning occurs. (5) Corridors and open spaces: white cane users veered from the intended path in wide corridors and open areas without nearby walls, while narrow corridors self-corrected veering when the cane hit a wall. Localisation accuracy should be prioritised in open areas. (6) Landmarks and obstacles: users valued information about POIs (water fountains, restrooms), obstacles (chairs, columns), and landmarks (floor changes), but the value depended on navigation aid type — guide dog users did not need obstacle warnings since the dog avoids them, while cane users found them essential.
Relevance
This paper provides uniquely granular, behaviour-level insights for designers of indoor navigation systems for blind users. Most navigation system evaluations report aggregate performance metrics, but this analysis reveals the specific interaction patterns that cause failures — enabling targeted design improvements. The distinction between guide dog and white cane users is particularly valuable: these two groups have fundamentally different navigation strategies, error recovery mechanisms, and information needs, yet most systems treat them identically. For accessibility practitioners, the six design recommendations are immediately actionable: combine closely spaced instructions; provide context about alternative paths near turns; detect and intervene on erroneous behaviour earlier; provide additional feedback for slight turns and over-turning; prioritise localisation accuracy in open spaces; and adapt environmental information to the user’s navigation aid type. The observation that users with echolocation skills recovered from errors faster by detecting they had passed a corridor opening connects O&M skills with technology design — navigation systems should complement rather than replace existing orientation skills.
Tags: indoor navigation · blind · orientation and mobility · guide dog · white cane · Bluetooth beacons · wayfinding · user behavior · error recovery · navigation strategies