AIBRIEF
The South Korean carrier's in house vision system watches video from its own service vans and flags 11 classes of network faults, with no dedicated inspection fleet.
SK Telecom has commercialized an in-house AI called Topda (톺다) that turns video from cameras already mounted on its service vehicles into a passive inspection layer for the carrier's nationwide network. Announced on September 1, 2026 and demonstrated the same day at SKT's Bundang building in Seongnam, the system runs on footage the fleet already records during routine fault-response and civil-complaint runs. No dedicated inspection vans are required.
Topda is built on SKT's VISTA (Visual AI Inspection System for Telco Assets) platform, which pairs vision AI for object recognition with context AI for situation reading and spatial AI for location judgment. The hardware stack is cameras, GPS antennas, and neural processing units on the work vehicles, with no LiDAR or other high-cost sensors. SKT says the same pipeline also runs on drone video.
The system flags 11 classes of abnormality, including damaged facilities and defective construction sites. In one demo frame, Topda spotted construction equipment near 17 cables, one enclosure (함체), and one utility pole (전주), letting a network operator decide whether to dispatch. The scale it is meant to cover is the wired network itself: roughly 200,000 km of optical cable and 2.35 million utility poles, a footprint two-person manual crews cannot keep up with.
What the company has not yet disclosed is how reliably those 11 classes fire from a moving windshield. The load of false positives a control room has to triage is the open operational question.