Road section reviews, product updates, and field insights from Mata Garuda.
FeaturedA proud milestone in AI-powered road infrastructure management: Mata Garuda was named among the winners of the AI-Based Road Damage Detection Competition at National Road Day (Hari Jalan) 2023, organized by the Directorate General of Highways, Ministry of Public Works and Housing (PUPR).
Read moreMata Garuda's detection engine has moved from YOLOv8x to YOLOv11. The newer architecture improves both detection accuracy and inference efficiency, and the gain carries across all 27 supported defect types — on flexible and rigid pavement alike.
Our training set now draws on more than 45,000 annotated images extracted from over 200 survey videos, collected across 19 regions of Indonesia — from Aceh and Sumatra Utara, through Java and Bali, to Kalimantan, Sulawesi and Papua Barat. The regional spread is the point: it exposes the model to the full range of Indonesian road conditions, surface materials, lighting and weather rather than one province's idea of a road.
Benchmarking puts average accuracy at around 90% for the five most common road defects in Indonesia, with up to 90% overall performance across all defect types. Consistency matters as much as the headline figure: an automated assessment applies identical criteria to every kilometre, removing the inspector-to-inspector variation that manual surveys inevitably carry.
Mata Garuda classifies 27 distinct distress types in a single pass. On flexible (asphalt) pavement that spans longitudinal, lateral, alligator and edge cracking, potholes, rutting, raveling, bleeding, corrugation, depression and shoving. On rigid (concrete) pavement it covers corner breaks, faulting, punch-outs, joint and corner spalling, divided slabs, scaling, shrinkage and polished aggregate — alongside shared conditions such as patching, delamination, shoulder drop-off and railroad crossings.
Mata Garuda ran AI-assisted assessment over footage from a Bina Marga road survey. Every detection is written back with its GPS coordinates, road chainage, timestamp, and the source image and video frame — so any finding on the map can be traced to the exact stretch of road it came from, and defended when a maintenance budget depends on it. Footage from the survey is featured on our homepage.
Mata Garuda works from video captured by ordinary vehicles — dash cameras, smartphones, fleet and patrol vehicles, or public transport — with no purpose-built survey rig required. That is where the operational saving comes from: roughly 70% less inspection time and around 60% lower assessment cost compared with conventional manual surveys, and a road network that can be re-surveyed as often as vehicles happen to drive it.
Beyond Indonesia, Mata Garuda's training data now includes road imagery from Japan, New Zealand and the United States, widening the range of pavement standards, marking conventions and climates the model has seen. Datasets covering Europe, the Middle East and North Africa, and South Asia are in progress.