Robot-Assisted Group Tours for Blind People
Yaxin Hu, Masaki Kuribayashi, Allan Wang, Seita Kayukawa, Daisuke Sato, Bilge Mutlu, Hironobu Takagi, Chieko Asakawa · 2026 · Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems (CHI '26) · doi:10.1145/3772318.3790425
Summary
Hu and colleagues investigate how an assistive mobile robot can support blind people's participation in mixed-visual group tours, where blind and sighted visitors share the same guided experience. The motivation is that group activities are central to social life, but blind participants in group tours routinely struggle to approach the guide in noisy environments, follow group movement at its visual pace, position themselves correctly relative to the group, ask questions at appropriate moments, and benefit from spontaneous visual discoveries that sighted peers make. Existing tour-accessibility solutions either rely on scarce dedicated staff or treat blind visitors as a separate group, neither of which scales. The authors run a three-phase study at Miraikan: (1) semi-structured interviews with five blind people and five museum science communicators to surface needs and challenges; (2) design and implementation of a CaBot-based suitcase-shaped robot with a four-button handle (left: surrounding GPT-4o description, right: guide location, right-long: nearby people locations, down: notify the guide); (3) a field study in which eight blind participants each joined a real museum tour with one human guide and two unfamiliar sighted participants. The robot addresses three design goals: environmental awareness (vision-language scene descriptions), group-interaction awareness (guide and bystander positions via UWB and on-board cameras), and independent navigation (autonomous following with LiDAR collision avoidance, tele-operated in this study for reliability). System logs, video recorded with ELAN coding, SUS, RoSAS, and post-tour interviews captured how blind participants, sighted co-tourists, and the tour guide experienced robot-mediated participation.
Key findings
Blind participants reported a strong sense of safety from the robot's navigational support (SUS mean 90.6, SD 12.9 — well into the 'excellent' band) and described tours as 'surprising', 'enjoyable', and 'fun'. RoSAS scores were positive across warmth and competence and low on discomfort. Feature usage was uneven and revealing: the Guide Location feature was used by all eight participants (mean 4.88 triggers per tour), Surrounding Description by all (mean 4.13), Nearby People Locations by seven (mean 2.75), and Call Guide rarely (mean 0.75) because the small three-person tour group made direct conversation easy. Participants triggered features mostly during transitions between exhibits (37%) or on arrival at a new exhibit (40%), and especially when the robot got stuck (a recurring problem in afternoon crowds, sometimes triggered when the robot's LiDAR mistakenly detected the user's white cane or feet as obstacles). Qualitative findings split on the robot-versus-human-companion comparison: some participants valued the schedule independence and reduced 'sense of obligation' that came with not relying on a sighted family member, while one participant felt the robot left him 'the same as sitting in a seat and listening to a talk', struggling to keep up with sighted-visitor understanding. Two participants reported heightened self-consciousness from being 'treated specially' or 'being different'. Sighted co-tourists rated the experience positively, but several reported confusion about how to position themselves around the robot, and the tour guide noted that the robot's slower movement set a slower tour pace overall.
Relevance
For accessibility practitioners, this paper extends the assistive-robot research programme from solo navigation (CaBot, AI Suitcase, WanderGuide) to multi-stakeholder group settings — a much harder design problem because the robot must mediate among the blind user, a tour guide who is responsible for the group, and other visitors who did not opt into the system. The empirical contribution is a concrete inventory of robot features that map to specific group-tour challenges, plus the negative finding that some features (Call Guide) only become valuable at larger group sizes and emergencies. The design implications are directly actionable for museums, art galleries, theatres, and gardens that already run accessibility programmes: the robot supplements the human guide rather than replacing them, lowers the social cost of asking for help, and lets blind visitors share standard tours rather than requiring separate scheduled sessions. Limitations to flag: tele-operated following (not full autonomy in the wild), one museum and one well-trained accessibility-aware tour guide, three-person group size, and short-term exposure. Practitioners considering similar interventions should attend particularly to the heightened self-consciousness reported by some blind participants — the robot should not become a stigmatising marker of difference. The convergence of this paper with the Hata et al. delegation study and the Kuribayashi et al. WanderGuide work signals a maturing research thread that practitioners building accessible-tour or accessible-museum programmes should watch closely.
Tags: assistive robotics · museum accessibility · blindness and low vision · visual impairment · group activities · social inclusion · mixed-ability · human-robot interaction · navigation · inclusive design