8B-Parameter Multimodal Urban Safety Foundation Model to Be Tested in Public Spaces Across Four Districts of Daejeon

Demonstrations Planned at Five or More Locations, Including Daejeon Station, Jungang Market, and the Gapcheon River, with Nationwide Expansion Targeted by 2030

(WAVUS signed a memorandum of understanding with Daejeon Metropolitan City and participating ICT companies and organizations at Daejeon City Hall on September 8 to support the development and field demonstration of an urban safety AI model. Courtesy of WAVUS)

[By Sim-geon Lee, Chungcheong Today]

WAVUS Co., Ltd., a company specializing in geospatial information and artificial intelligence, signed a memorandum of understanding (MOU) with Daejeon Metropolitan City on September 8 to cooperate on field demonstrations for the Urban Safety AI Model Development and Demonstration Project.

The project is being implemented as part of the Regional Innovation Cluster Advancement Program (Leading R&D), supported by the Ministry of Trade, Industry and Energy (MOTIE) and the Korea Institute for Advancement of Technology (KIAT).

The agreement formally establishes Daejeon as the field demonstration site for the Urban Safety AI project. The signing ceremony was held in the conference room of Daejeon City Hall.

The Regional Innovation Cluster Advancement Program is designed to generate region-led innovation by concentrating local industry, academia, and research capabilities in strategically specialized sectors.

WAVUS serves as the lead organization for the Urban Safety AI project, which brings together ten participating organizations, including the Korea Advanced Institute of Science and Technology (KAIST) and DataMaker Co., Ltd. The research project will run for four years and nine months, from April 2026 through December 2030.

◆One AI Connecting Detection, Prediction, Response, and Recovery

The project’s central objective is to enable a single AI model to continuously perform the decision-making tasks that have traditionally been fragmented across different stages of disaster response.

(WAVUS signed a memorandum of understanding with Daejeon Metropolitan City and 11 participating companies and organizations in the information and communications technology (ICT) sector at Daejeon City Hall on September 8 to collaborate on the development and field demonstration of an urban safety AI model. Courtesy of WAVUS)

The research team will develop an 8-billion-parameter multimodal foundation model capable of simultaneously processing heterogeneous data, including video footage from CCTV cameras and drones, audio data, IoT, weather and water-level sensor data, geospatial information such as GIS and BIM data, and text.

The model will establish a continuous reasoning framework covering the entire disaster management process—from disaster detection and impact-spread prediction to recommendations for evacuation routes and resource deployment, as well as the prioritization of recovery activities.

The system will also incorporate a geospatial knowledge graph (GeoKG), which structures the spatial relationships among buildings, roads, and infrastructure as a graph. This will enable the model to identify blind spots in evacuation planning and infer cascading effects in which an initial disaster spreads to adjacent areas.

The project’s target performance indicators include an area under the curve (AUC) score of 0.90 for disaster detection accuracy, a real-time inference latency of no more than 150 milliseconds, and a score of 90 for the quality of Korean-language safety knowledge responses.

Two versions of the model will be deployed: a full-scale 8B+ model for emergency operation centers and a lightweight 2B–3B model for on-site edge devices.

◆Field Demonstrations in Public Spaces Across Four Districts of Daejeon

The demonstrations will take place not in controlled experimental settings, but in public spaces used daily by Daejeon residents.

In Dong-gu, Daejeon Station and the Jungangno Underground Shopping Center will be used to test responses to underground fires and flooding.

In Jung-gu, the project will address fire risks in aging facilities and crowd congestion around Jungang Market, Sungsimdang, and Hanwha Life Ballpark.

In Seo-gu, demonstrations will focus on crowd congestion and river flooding along the Daejeoncheon and Gapcheon rivers and at the Gapcheon Ecological Lake Park.

In Yuseong-gu, the Yuseongcheon River and Shinsegae Department Store will serve as demonstration sites for complex disaster scenarios.

During the fifth and final project year, comprehensive demonstrations involving fire authorities, police, and local government officials will be conducted at five or more locations.

(WAVUS signed a memorandum of understanding with Daejeon Metropolitan City and 11 ICT companies and organizations at Daejeon City Hall on September 8 to collaborate on research, development, and field testing of an urban safety AI model. Photo courtesy of WAVUS.)

Much of the infrastructure required for the demonstrations is already in place. Available resources include 9.2 square kilometers of 3D mesh data covering the Daejeon area, BIM data for Daejeon Station and the underground shopping center, 3D data on underground infrastructure—including water supply, sewerage, stormwater, and wastewater pipelines—and connectivity with CCTV systems operated by the Daejeon Transportation Corporation.

These existing assets will allow the consortium to begin designing the field demonstrations immediately following the signing of the agreement.

◆An Operational System for Emergency Control Centers and Preventive Administration

WAVUS stated that the project has been designed to ensure that its outcomes do not remain limited to research reports, but are implemented as an operational system that can be used directly by Daejeon Metropolitan City.

Data currently distributed across weather, hydrological, video, and infrastructure systems will be integrated into a single 3D digital twin interface. Based on AI-generated risk indices, the system will then provide recommended response measures.

Final decisions and approvals will remain the responsibility of authorized public officials, while the basis and confidence level of each AI-generated recommendation will be recorded automatically.

The project will also support preventive disaster management. Facility registry information will be combined with vibration and displacement sensor data collected in the field, enabling authorities to prioritize inspections of facilities showing signs of abnormality instead of relying solely on fixed inspection schedules.

In addition, the system will analyze historical flood records, infrastructure deterioration, and pedestrian density to produce area-specific priorities for disaster prevention investment. These recommendations will be provided annually to support the municipal budgeting process.

The project also includes plans to feed historical disaster data back into the system and use it as a virtual response training environment for firefighters, police officers, and public officials.

◆Developing Local Talent and Data Assets

The project plans to create 150 new jobs during its implementation period, including 45 positions for young professionals.

Priority will be given to graduates and master’s and doctoral students from universities in Daejeon, including KAIST, Chungnam National University, and Hanbat National University. Recruited personnel will be assigned to key technology areas such as artificial intelligence and digital twins.

Disaster-related data accumulated in Daejeon over the five-year project period will be refined and standardized to establish a regional public data asset. The data will subsequently be made available for use by local companies and research institutions.

(WAVUS signed a memorandum of understanding with Daejeon Metropolitan City and 11 ICT companies and organizations at Daejeon City Hall on September 8 to collaborate on the research, development, and field demonstration of an urban safety AI model. Photo courtesy of WAVUS.)

This approach aligns with the objective of the Regional Innovation Cluster Advancement Program, which seeks to concentrate the capabilities of regional companies and universities in specialized industries.

Participating companies based in Daejeon will directly engage in the development of foundation models, platforms, and edge technologies, enabling them to build in-house expertise in AI-based disaster and safety management.

The consortium also plans to connect the project with the technology commercialization infrastructure of the Daedeok Innopolis research and development cluster to support follow-up commercialization initiatives.

The research team has presented a roadmap to package the results of the Daejeon demonstrations into a standardized deployment solution. Following early-stage pilot projects in neighboring local governments, the solution will be rolled out gradually to municipalities nationwide.

“The greatest challenge in disaster response is not a lack of information, but the inability to connect fragmented information at the right time,” said Chang-hoon Park, CEO of WAVUS. “We will ensure that the results validated in Daejeon become a standard solution that local governments across the country can readily adopt.”

Sun-woo Park, Project Director and Principal Investigator, added, “Because the demonstration sites are part of citizens’ everyday living environment, this project must validate not only technical performance indicators but also practical acceptance in real-world settings.”

Source: Chungcheong Today

Original article by Sim-geon Lee (beotkkot@cctoday.co.kr)

[Access the original press release]