Every year, municipalities around the world spend billions of dollars maintaining their road networks. Yet budgets never stretch far enough to fix everything, and deciding which pothole to repair first has traditionally relied on citizen complaints, inspector intuition, or whoever happens to call city hall loudest. A research team at the University of Texas at Dallas (UT Dallas) is challenging that status quo with an AI-powered system designed to bring objectivity and efficiency to one of urban governance's most unglamorous problems.
Why Potholes Are a Bigger Deal Than They Look
A pothole is rarely just a nuisance. According to the American Automobile Association, road damage caused by potholes costs U.S. drivers approximately $3 billion per year in vehicle repairs alone. Beyond financial impact, potholes cause accidents, accelerate deterioration of drainage infrastructure, and disproportionately affect lower-income neighborhoods — where formal complaints tend to be filed less frequently, meaning those areas are systematically underserved by traditional response systems.
The core challenge for city managers is not simply detecting potholes — modern inspection vehicles and even smartphone apps handle that reasonably well — but deciding which ones to fix first when the maintenance budget runs out long before the repair list does. That prioritization problem is exactly what the UT Dallas team tackled in collaboration with a private urban infrastructure management company.
How the AI Prioritization System Works
The system ingests multiple data streams to generate a priority score for each identified pothole across a city's road network. Key inputs include:
- Damage severity: assessed through high-resolution imagery captured by camera-equipped municipal vehicles or sensor data collected from fleet cars already in service.
- Traffic volume: a pothole on a main arterial road handling 20,000 vehicles per day carries far greater collective risk than an identical defect on a quiet residential street.
- Repair history: the model learns from past maintenance records to flag zones where pavement degrades faster than average, which can signal deeper structural issues beneath the surface.
- Equity indicators: some model configurations incorporate socioeconomic data to ensure that resource allocation doesn't simply mirror historical patterns of neglect in underserved areas.
The underlying machine learning model was trained on real inspection data and historical work orders from partner municipalities. This gives it the ability not only to rank current damage, but also to predict which road segments face the highest deterioration risk over the coming months — shifting the city's posture from reactive patching to proactive planning.
Pilot Results and Real-World Impact
UT Dallas researchers reported that pilot testing demonstrated meaningful reductions in maintenance response times and a more defensible distribution of available budget. Instead of routing crews toward the most visible or most complained-about potholes, city managers could deploy resources toward the cases with the highest collective impact — whether or not those roads had generated a single 311 call.
One of the project's deliberate design choices was scalability. The system was built to integrate with existing municipal management platforms rather than requiring cities to overhaul their technology stack. That makes it accessible to mid-sized cities operating on tight budgets, not just large metros with dedicated smart-city offices.
A Broader Shift: AI Meets Urban Infrastructure
The UT Dallas project fits within a wider global movement applying machine learning to what engineers call critical infrastructure systems. Cities like Pittsburgh, Amsterdam, and Singapore have spent years experimenting with AI for traffic management, water distribution, and public transit optimization. Road maintenance, however, has lagged behind — partly because the problem seemed too tactile and localized to benefit from algorithmic approaches.
This system challenges that assumption. What makes it particularly noteworthy is its explicit attention to algorithmic fairness. The researchers acknowledged upfront that a model trained purely on complaint data would likely replicate historical bias, prioritizing wealthier neighborhoods simply because their residents are more likely to report issues through official channels. Building equity considerations directly into the scoring model represents a meaningful step toward AI that serves all residents, not just the most vocal ones.
The intersection of operational efficiency and distributional justice makes this a compelling case study not only for civil engineers, but for public policy professionals and anyone thinking seriously about how automated systems shape city life. Roads are one of the most tangible expressions of how a government values its citizens' time, safety, and mobility.
The team's next steps include publishing the full pilot results and pursuing agreements with additional municipalities to scale deployment. The takeaway is straightforward: fixing potholes has never been a glamorous job, but deciding which ones to fix just got a lot more scientific.
