Field Trials of Autonomous Navigation Robot for Visually Impaired People
Hironobu Takagi, Kakuya Naito, Daisuke Sato, Masayuki Murata, Seita Kayukawa, Chieko Asakawa · 2025 · Extended Abstracts of the CHI Conference on Human Factors in Computing Systems (CHI EA '25) · doi:10.1145/3706599.3706663
Summary
This paper reports field trials of AI Suitcase, an autonomous navigation robot shaped like a conventional carry-on suitcase and designed to guide blind and visually impaired (BVI) travellers safely through unfamiliar public spaces. The system traces its origins to the CaBot research project begun at Carnegie Mellon in 2017 and is positioned as a socially implementable alternative to staff escorts, guide dogs, and white-cane-only travel in places where acquiring environmental information is particularly challenging (shopping malls, museums, transit hubs). The user grips a handle containing a touch sensor — releasing it stops the robot — while a smartphone app is used to select destinations or predefined tours by voice or menu, and three vibration motors on the handle pre-cue turns and stops. The suitcase form factor was chosen to blend into urban environments, to let the user lift the robot over obstacles, and to naturally detect stairs or corners in front of the user. Indoor (2022, ~15 kg) and outdoor (2023, ~30 kg with large wheels) models share LiDAR, RGB-D cameras, hub motors, and on-board GPU. The authors report three trials: a pilot in Tokyo's Nihonbashi Muromachi commercial complex with 38 BVI participants; permanent daily operation at Miraikan science museum (launched 18 April 2024) with 1,288 users logged to 30 September 2024, about a quarter visually impaired; and an outdoor 400 m route between the museum, a park, and a railway station with 34 BVI participants across two sessions.
Key findings
Across all three trials, AI Suitcase consistently received high satisfaction ratings from BVI participants. At the Miraikan daily operation, 83% of 89 visually-impaired respondents were satisfied or very satisfied with overall experience (Q1), and 71% felt safe or very safe (Q2). Sighted co-visitors rated the experience equally highly (94% satisfied, 79% feeling safe). The outdoor trial returned a SUS score of 83.2 ('excellent'), with the worst-rated items being curb negotiation and zebra-crossing navigation (~3.5 and 3.4), and safety ratings for obstacle avoidance of 4.5–4.7. Qualitative comments surfaced two recurring issues: the robot's frequent stops in crowded conditions caused anxiety because users were not notified when the suitcase halted, and 20 cm wheels sometimes failed to climb even 3 cm curbs. In comparison with guide dogs, participants praised the robot for not needing daily care and for reliable destination-based navigation, but noted that guide dogs make quicker decisions in pedestrian traffic and can navigate steps. The paper also reports a provocation: a segment of BVI participants at the commercial complex felt unsafe when the suitcase made them less visible, referencing Thomas Carroll's 'loss of obscurity' concept and the Japanese legal requirement that solo BVI pedestrians use a white cane or guide dog under Article 14.1 of the Road Traffic Act.
Relevance
For accessibility practitioners — particularly those working in public buildings, transport, cultural venues, or municipal wayfinding — this paper is valuable as the first published account of a permanently deployed autonomous navigation robot for BVI users in a real museum, not a lab study. The authors name the operational challenges that any organisation considering this technology must plan for: fragmented ownership requires separate permissions from every facility owner along a route; outdoor deployment in Japan required legal classification as a mobility aid analogous to a powered wheelchair; elevators require manual button pressing because robots cannot yet reliably call and select floors; and near-misses with inattentive pedestrians remain a significant risk. The 'loss of obscurity' discussion is especially important: the question of whether a BVI person should be visibly identifiable in public space is contested within the BVI community itself and cannot be settled by designers alone. Limitations: the paper is an 8-page CHI extended abstract rather than a controlled study; the outdoor tele-operation detail is understated; and all three trials took place in Japan, where social norms, infrastructure, and legal frameworks differ markedly from other regions.
Tags: assistive robotics · blindness and low vision · navigation · indoor navigation · wayfinding · field study · museum accessibility · social acceptability · human-robot interaction