AI Applications in Road Pavement Technology

A Bengaluru engineer built AI that can trace a pothole to its contractor. Gaurav Sen, founder of InterviewReady, is building a system that turns pothole detection into ready-to-file civic complaints. In 2024, 2,385 people lost their lives in pothole-related road accidents across India. The challenge was finding who was responsible for fixing the error. Sen’s system uses a dashcam, GPS and accelerometer to capture road data. AI then detects potholes, classifies their severity and maps them to the exact road stretch. The most interesting part? It scans around 2,900 government road contracts to identify the contractor responsible and the concerned officer. And in around seconds, it can create a complaint with images, location, tender details and supporting evidence. During one test drive, it detected 12 potholes and generated 12 complaint records. This is the kind of AI use case that deserves attention. Not because it automates a task. Because it connects a problem with accountability. The real power of this idea is simple. It turns “someone should fix this” into “this person is responsible for fixing this.” #AI

Developing an end-to-end computer vision project in a single tab, it’s truly incredible, but now it’s possible with the Ultralytics Platform. Autonomous vehicles don’t fail because of flawed models. The reason for their failures is elements that the model wasn’t trained to recognize. Now I’ll tell you about a project I developed on the Ultralytics Platform, where I handled the entire workflow end-to-end in a single tab, from labeling with SAM 3 to cloud GPU training, from real-time loss tracking to exporting. Cracks. Potholes. Patches. Traffic cones. Road barriers. These 5 classes sit at the exact intersection of vehicle safety and driving comfort, yet they are almost never addressed as a single detection problem. That’s why I developed Road Sense-Seg to change that. Segmentation using YOLO26x-seg on an RTX PRO 6000. Trained end-to-end on the Ultralytics Platform. Post-training results: • mAP50(M): 91.0% • mAP50-95(M): 67.8% • Sensitivity(M): 86.3% • Recall(M): 85.2% But the standout feature is the browser-based Prediction tab. Raw frames are input. Segmentation masks are generated. Instantly. No local GPU is required for inference testing. The video below shows exactly how this looks in the real world: raw vehicle camera images on top, RoadSense-Seg segmentation output on the bottom. If you’re working on ADAS, autonomous driving, or road infrastructure monitoring, this is the foundational perception layer that makes all the difference. #Ultralytics #UltralyticsPlatform #YOLO26 #ComputerVision #AutonomousDriving #Segmentation #SAM3 #ADAS #RoadSafety #VisionAI #MachineLearning

Dubai is leveraging AI to detect road defects, enhancing the safety and maintenance of its road network. AI algorithms analyze images and sensor data to identify issues like cracks, potholes, and uneven surfaces. The models, trained on extensive datasets, can detect early signs of deterioration invisible to the human eye, enabling timely repairs. This system allows for continuous monitoring, generating reports that help prioritize maintenance efforts and minimize traffic disruptions. The approach is faster and more accurate than traditional inspections, improving road longevity and safety. Video Credit: The National News #mehrtashsoltani

We used a large multimodal model to analyze a saw-cut asphalt pavement cross-section for image-based material characterization. The model was used to generate a binary segmentation for detecting aggregate particles from binder and voids, and then run a gradation analysis through particle size estimation based on established digital image analysis methods. This example shows how multimodal models can estimate aggregate size distribution from high-quality images or videos and complement conventional testing. #MultimodalAI #ComputerVision #ImageAnalysis

More on AI Use Cases with State DOTs As a follow an earlier post, I wanted to share a bit more about the types of AI Use Cases that were recently shared by State DOTs across the nation. Use cases can be largely grouped into two categories: the “hard” side of infrastructure management and the “soft” side of enterprise operations. Across both areas, the focus is on utilizing AI as a strategic force multiplier rather than a replacement for human expertise. On the hard 'operations' side, machine learning and computer vision are shifting our approaches from reactive to proactive—accelerating incident detection and automating the evaluation of pavement conditions. On the soft 'administrative' side, tools like workflow automation and document digitization are cutting processing times and returning thousands of staff hours to our agencies. Based on these examples, a few themes stand out: • Enhanced Safety and Operations: AI systems are increasingly effective at ingesting real-time data to identify roadway incidents and changing hazards, allowing operators to respond faster. • Proactive Asset Stewardship: Predictive analytics are processing massive amounts of imagery and fleet data, helping agencies anticipate maintenance needs before physical failures occur. • Human-in-the-Loop Governance: By taking over routine data processing, AI allows transportation professionals to focus on high-level decision-making while retaining ultimate accountability for verifying and validating all outputs. #Transportation #StateDOTs #AI #Infrastructure #TransportationPlanning #AASHTO

The next breakthrough won’t come from stacking agents on top of AI that understands language… The next frontier is teaching AI to compute physical reality 🌍 Bridges deform. Foundations scour. Roads settle. Yet critical infrastructure is still managed reactively, with periodic snapshots of systems that never stop changing… The question isn’t “What’s the condition today?” It’s “What happens next ⁉️ ” How will this bridge behave after three more hurricane seasons? How much will this foundation scour after five more flood cycles? What does this corridor look like in 2035 if nothing changes? At Geolava, we’re working with DOTs (Departments of Transportation) to build spatial AI that learns from the physical world, running thousands of simulations every day to forecast defects before they become failures. The future of infrastructure isn’t reactive. It’s predictive ⚡ #SpatialIntelligence #WorldModel #SpatialAI