The Future of Urban Accessibility: The Role of AI
Jon E. Froehlich, Chu Li, Maryam Hosseini, Fabio Miranda, Andres Sevtsuk, Yochai Eisenberg · 2024 · Proceedings of the 26th International ACM SIGACCESS Conference on Computers and Accessibility (ASSETS '24) · doi:10.1145/3663548.3688550
Summary
This workshop paper examines the emerging role of artificial intelligence in designing equitable and accessible cities, transportation systems, and navigation tools for people with disabilities. The authors — spanning HCI, urban planning, public health, and accessibility research — argue that while AI has transformed many domains, its potential to address the systemic inaccessibility of the built environment remains underexplored. The paper identifies six thematic areas where AI intersects with urban accessibility: (1) assessing pedestrian pathways using computer vision and street-level imagery to detect barriers like missing curb ramps, broken sidewalks, and obstructions; (2) indoor accessibility mapping and navigation; (3) autonomous vehicles and their potential to transform transportation for people who cannot drive; (4) intelligent wheelchairs and assistive mobility devices; (5) assistive human-robot interaction in urban settings; and (6) overarching challenges related to ethics, bias, data privacy, and the digital divide. The paper draws on the authors' prior work including Project Sidewalk, a crowdsourcing platform where volunteers label accessibility barriers in Google Street View imagery, and urban network analysis tools that model pedestrian accessibility across city-scale infrastructure. The workshop brought together researchers from HCI, disability studies, gerontology, social work, community psychology, and law to identify open challenges and spur interdisciplinary collaboration.
Key findings
The paper highlights several critical findings and challenges. Current approaches to assessing urban accessibility are largely manual, with cities relying on costly and infrequent physical audits that quickly become outdated. AI-powered tools using computer vision and LiDAR can automate the detection of accessibility barriers at scale — Project Sidewalk alone has collected over 500,000 crowdsourced labels across multiple cities, and machine learning models trained on this data can now automatically identify problems like missing curb ramps and surface quality issues from street-level imagery. However, significant challenges remain: AI models trained on data from one city may not generalise to others with different infrastructure; street-level imagery has gaps in coverage, particularly in lower-income neighbourhoods; and the relationship between detected barriers and the actual lived experience of disabled pedestrians is complex and context-dependent. For autonomous vehicles, the potential is transformative — in the US alone, over 600,000 people with disabilities cannot leave their homes due to transportation barriers — but current AV development has largely ignored accessibility in vehicle design, pickup/dropoff logistics, and interaction modalities. The paper also raises concerns about AI bias: training data may underrepresent certain disability types, and algorithmic systems risk encoding existing inequities if not designed with disability communities.
Relevance
This paper is valuable as a landscape survey of AI applications in urban accessibility, connecting work that is often siloed across disciplines. For accessibility practitioners, it highlights that the built environment — sidewalks, crossings, transit stops, buildings — remains one of the most significant barriers to participation for disabled people, and that AI offers scalable approaches to assessing and addressing these barriers. The Project Sidewalk model of combining crowdsourcing with machine learning for accessibility auditing is directly applicable to cities and organisations seeking to map their infrastructure. The autonomous vehicle discussion is a timely reminder that emerging transportation technologies must be designed with disability from the outset, not retrofitted. As a workshop paper, it is necessarily broad rather than deep, but it effectively maps the research landscape and identifies critical gaps including the need for longitudinal data on how urban accessibility changes over time, better integration of subjective disability experience into AI-driven assessment, and governance frameworks for urban AI that centre disability rights.
Tags: urban accessibility · artificial intelligence · smart cities · autonomous vehicles · pedestrian infrastructure · wayfinding · built environment · computer vision
Standards referenced: ADA