“Construction site detected.”
A high-precision camera and neural processing unit (NPU) installed on a vehicle collect and analyze surrounding footage, identifying potential threats to telecommunications infrastructure. While the vehicle is in motion, the system evaluates 11 inspection categories—including construction sites, leaning utility poles, damaged manholes and sagging cables—to determine their potential impact on nearby network facilities and help maintain a safe, reliable communications environment.
SK Telecom on Sept. 1 unveiled TOPDA, an AI-powered solution for automated telecommunications-infrastructure inspections, at a media briefing at its Bundang headquarters in Gyeonggi Province.
TOPDA uses proprietary AI to inspect the surroundings from SK Telecom work vehicles and assess the safety of telecommunications infrastructure in real time. The rollout marks SKT's latest step in expanding AI-based operations and management from wireless and core networks to fixed-line infrastructure, accelerating its companywide network AI transformation, or AX.
Installed on SKT work vehicles, TOPDA autonomously identifies and evaluates inspection targets. To develop the system, SKT used AI to analyze and classify 22,000 field photographs collected over five years. Following a four-month pilot program, it has achieved a detection accuracy of about 88% and aims to raise that figure to 95% next year.
The solution combines SKT-developed Vision AI, Context AI and Spatial AI. This allows it to sense, analyze and provide the information required for inspections without relying on costly hardware such as LiDAR. The underlying VISTA platform functions as a digital-twin system, enabling real-world telecommunications assets to be mapped and analyzed in a virtual environment.
SKT also demonstrated TOPDA in a near-real-world setting, using a large bus equipped with high-precision cameras, GPS and an NPU server. During a short 3-kilometer route, the vehicle detected potential hazards in real time, including excavators and construction sites.
When it detected a hazard near an SKT-managed fiber-optic cable, the system immediately alerted the relevant personnel. It also provided additional operational information, such as the number of users served by the cable—in one demonstration case, 1,820 subscribers.
“Over the past four months, we have covered 50% of facilities in the Seoul metropolitan area,” said Bok Jae-won, head of network operations at SK Telecom. “We plan to expand coverage to more than 90% by 2028. To support this, we will increase the number of operating vehicles from six today to 150 next year.”
SKT is prioritizing AX in network operations and management in order to respond more quickly to traffic fluctuations in an AI-driven environment and overcome the physical limitations of conventional inspections. These limits include underground fiber-optic cables, transmission towers and other infrastructure that can be difficult or unsafe for personnel to inspect directly.
For its wireless network, SKT has already developed and deployed A-One, an AI operations system for predicting event-related traffic and monitoring network quality, as well as VISTA-Drone, an AI inspection system that analyzes drone footage to assess transmission towers and antennas. For its core network, it is using Spider, an AI-based solution that detects faults and recommends optimal corrective action.
A-One has reduced preparation time for major events with large crowds from five days to 30 minutes. Spider, meanwhile, has cut the actual fault rate by 53% and reduced recovery time by 24% after deployment, according to SKT.
With TOPDA, SKT has established an AI-based operations and management framework spanning wireless, fixed-line and core networks. The company plans to eventually integrate data from each network domain and develop a more advanced, end-to-end network-control system.
“We are currently collecting data separately from fixed-line, wireless and core-network domains, and are developing a plan for integrated monitoring and control,” said Yang Jeong-dae, head of the Network Control Team. “We will determine how to manage cross-domain operations, including how faults in fixed-line networks may affect the core network.”
