The mandate was not a routine condition survey. It was something more structurally ambitious: to build a digital foundation for long-term road asset management across a 100-kilometre network in Riyadh — one that could support budget planning, maintenance prioritisation, and infrastructure lifecycle decisions at scale using AI-powered road assessment and predictive road analytics.
The problem with traditional inspection at that scale is not accuracy. It is consistent. Manual methods depend on field teams who classify defects differently on different days, generate reports on different timelines, and produce outputs that resist the kind of structured, comparable analysis that asset management systems require. At 5 kilometres, those inconsistencies are manageable. At 100 kilometres, they compound into a dataset that cannot be trusted for financial decisions.
Specifically, five structural limitations existed with the prevailing approach: subjective defect categorisation varying between inspectors, delayed reporting cycles that allowed early-stage defects to escalate, limited chainage-level visibility masking localised deterioration within healthy corridors, difficulty prioritising intervention zones without comparative condition data, and high dependency on field documentation that was difficult to audit or verify.
The solution had to go beyond counting defects. It had to deliver the kind of structured, reproducible, granular condition intelligence that turns inspection into infrastructure governance through AI-driven infrastructure intelligence.
RoadVision AI implemented a mobile-based AI road inspection system across the client's 100-kilometre network. The operational approach was deliberately simple: field teams mounted mobile devices on regular vehicles and drove each corridor at normal traffic speeds. No laser profiling equipment. No specialist survey trucks. No manual defect tagging in the field.
Mobile devices captured continuous roadway imagery throughout each survey run. Once uploaded to the RoadVision cloud platform, the AI engine processed the footage automatically — scanning every frame using trained computer vision models to detect, classify, and geo-tag surface distresses without any manual intervention in the classification workflow, enabling automated road condition detection.
Every 10-metre segment of the inspected network was evaluated independently and assigned a PCI (Pavement Condition Index) score using automated pavement condition modelling. This granularity is the critical differentiator from corridor-level reporting: it reveals localised deterioration pockets that aggregate scores conceal, enabling targeted maintenance decisions rather than blanket resurfacing programmes.
Each defect record was geo-tagged, severity-classified, chainage-mapped, and supported with visual evidence extracted directly from the survey footage. The output was structured for direct integration into asset management systems — not a PDF report to be manually re-entered, but a structured dataset ready for capital planning workflows.
The pilot stretch of 1.19 kilometres yielded 751 surface defects — a density that underlines how much traditional inspection misses. The majority were cracking-related distresses, which carry a specific implication: distributed cracking across a corridor signals surface fatigue, not localised structural collapse. That distinction is not semantic. It separates a corridor that needs preventive microsurfacing from one that requires full structural rehabilitation — a difference of several orders of magnitude in cost.
Across the broader corridor, the overall PCI score was 95.83, placing the network at Level 1 (Good). But chainage-level analysis revealed localised 'Fair' pockets within that same healthy average — micro-deterioration zones that would have been invisible in traditional corridor-level reporting, and that represent precisely the early-intervention opportunities that prevent expensive future rehabilitation.
The most significant outcome of this deployment was not a defect count or a condition score. It was the transition from reactive maintenance to data-driven infrastructure governance — a structural shift in how road asset decisions get made. With chainage-level PCI visibility and structured defect clustering, the organisation moved from responding to visible failure to anticipating deterioration before it escalates.
This deployment demonstrates what happens when road inspection is redesigned from the ground up as a smart infrastructure monitoring system rather than a field survey.
For Riyadh's road network, the implications extend beyond the 100-kilometre pilot. The mobile-based methodology requires no specialist hardware, no traffic management, and no procurement lead time. It is directly replicable across every additional corridor in the network using the same operational approach. What was built for 100 kilometres is a framework for the entire city.
More broadly, the deployment validates a model that infrastructure authorities across the Gulf region are beginning to recognise: that the most cost-effective road maintenance strategy is not more frequent inspection, but smarter inspection — one that produces data structured for decisions, not just documentation. This was not a survey. It was the construction of an infrastructure intelligence system.
The Autonomous Road Engineer
AI that is transforming road infrastructure, from reactive to predictive.
