Hosei University, SSIL, and Pacific Consultants to Develop Physical AI Training Materials to Address Aging Infrastructure and Shortage of Inspectors; Selected for NEDO's "GENIAC" Program; Bridge and Sewer Inspection Data to Be Used as a Common Platform for AI and Robot Learning and Evaluation
Hosei University (Chiyoda-ku, Tokyo; President: Diana Khor; Principal Investigator: Professor Ryuichi Imai, Faculty of Engineering and Design), Space Service Innovation Laboratory (Chuo-ku, Tokyo; Representative of Board of Directors: Naohiko Kohtake, hereinafter SSIL), and Pacific Consultants Co., Ltd. (Chiyoda-ku, Tokyo; Representative Director, President Executive Officer: Satoko Okano) have been selected as operators for the "Research and Development on Data Ecosystem Construction" under the national project "GENIAC (Generative AI Accelerator Challenge)," led by the Ministry of Economy, Trade and Industry (METI) and NEDO (New Energy and Industrial Technology Development Organization), aimed at strengthening domestic generative AI development capabilities and social implementation.
This project develops AI-ready data that can be used for training and evaluating physical AI for infrastructure inspections, including inspection records of bridges and water supply and sewage systems, as well as 3D data. The core outcome is data for training and evaluating physical AI (AI operating in real spaces through devices such as robots) and a mechanisum for its continuous provision. We support the development and evaluation of inspection AI, autonomous robots, and drones from data.
[Background of this project] Addressing aging infrastructure and labor shortages through data utilization.
Currently, there are approximately 730,000 road bridges nationwide, of which about 240,000 are over 50 years old. Among bridges for which the construction year is known, the proportion is approximately 44% (as of the end of March 2026). Sewerage pipelines amount to approximately 500,000 km (as of the end of fiscal year 2024), and 25% of municipalities lack civil Engineer or Engineer in their general administrative Dept. (as of April 1, 2025).
Furthermore, records accumulated on-site lack consistency in format, terminology, and management, making them difficult to reference for individual facilities and utilize for AI training. Hosei University and the other two organizations will work to support human knowledge and experience with data and AI, and to build a foundation for sustainably protecting essential infrastructure.
Source: Ministry of Land, Infrastructure, Transport and Tourism, "Annual Report on Road Maintenance (FY2025)" and relted materials. Details are provided in the attached document. Road bridges are defined as bridges with a length of 2m or more, and sewer pipelines exclude urban drainage channels. The 50-year age since construction does not directly indicate the level of risk or condition of individual facilities.
[Key points of this project] Converting on-site records into AI data. Building a system that goes beyond simply collecting data.
- Transforming dispersed inspection records into data that AI can learn from.
We organize drawings, ledgers, inspection photographs, and repair histories, and link them through location and structural information. - Synthetic data is used to supplement conditions that are difficult to capture with real-world data.
By reproducing changes in lighting and reflections on the water surface, it can be used for learning and evaluation in special environments. - Data will be continuously updated and provided, rather than simply collected and archived.
We will establish a system that connects data holders and developers, and provides data that has been verified for use and quality.
(1) Business Overview | Connecting three types of data: location, condition, and special environment
We will organize the formats, items, terminology, quality, and usage conditions of on-site records, and prepare them in a state that can be used for AI learning and evaluation (AI-ready). The target data will be the following three groups:
AI ready data to be developed under this project
| dataset | Content to be represented and data to be compiled |
|---|---|
| A | Spatial Infrastructure | Where are structures located, and what are they like? A reference base for location and shape across the entire country, and high-resolution 3D data for selected regions and sections. |
| B | Structural Infrastructure Inspection | What is their condition? How are they handled in the past? Drawings, ledgers, inspection records, and repair histories are organized in a way that allows them to be traced for individual facilities and components. |
| C | Special Environment Learning | How to learn in environments where data is difficult to obtain? This data combines real-world data, such as the inside of sewer pipes and the underside of bridges, with data obtained through physical simulations. |
1) Use A as a foundation to connect locations, structures, and records.
Using A as a common reference axis, information in B and C corresponding to facilities and sections will be linked by IDs that identify locations and structures. For example, the goal is to be able to trace drawings, inspection photographs, and repair histories of the same bridge, not only for AI learning but also for inspection planning and verification of past actions.
The level of detail in the correlation varies depending on the source data, and not everything will be precisely positioned in small sections. For sewer infrastructure, survey footage will be linked to pipe sections between manholes. Further details are provided in the attached document.
Please note that the nationwide spatial infrastructure is intended solely to provide a common location reference and does not include detailed geometry and inspection data for all facilities nationwide.
2) Synthetic data is used to supplement conditions that are difficult to capture.
Synthetic data is not a substitute for real-world data. We aim to distinguish between information obtained through actual surveys and conditions reproduced through simulation, and to evaluate the effectiveness of learning with synthetic data using real-world data.
3) The roles of the three ornanizations: Connecting research, data infrastructure, and field operations.
By combining research expertise from Hosei University, spatial information integration technology from SSIL, practical infrastructure expertise from Pacific Consultants, and collaboration with local public organizations and infrastructure operators, the three organizations will advance an integrated approach covering everything from data development to continuous provision.
[Main roles]
* Hosei University: Data structuring, correlation with location and structures, design and technical supervision of quality evaluation.
* SSIL: Construction of a data infrastructure connecting spatial information, search, distribution, and system operation.
* Pacific Consultants: Coordination with local public organizations and infrastructure operators, collection of real-world data, management of rights and usage conditions, and data management.
(2) Expected benefits
This project aims to reduce the burden on AI developers in searching for and organizing data, while promoting the development of technologies for searching inspection records, extracting damage information, and recognizing conditions in special environments. By making facility information more accessible to local public organizations and infrastructure operators, the project will support the planning of inspections and the consideration of preventive maintenance measures.
Reduced inspection time and costs, and improved safety are expected benefits. However, the actual benefits require verification according to the technologies used and the application conditions. Performance and safety verification are required for the autonomous operation of AI and robots in real-world environments.
(3) Basic information on the selected projects
| Business name | Post-5G Information and Communication Systems Infrastructure Enhancement R&D Project/Research and Development Related to the Construction of Data Ecosystems (GENIAC) |
|---|---|
| Selected theme name | Building AI-ready hybrid spatial and synthetic datasets to support improved infrastructure inspection efficiency and autonomous robot control. |
| Implementer | Hosei University / Space Service Innovation Laboratory / Pacific Consultants Co., Ltd. |
| Date of announcement of selection results | July 2, 2026 |
| Implementation Period | Planned for two years, starting in September 2026. |
The project name, selected theme name, implementer, and date of announcement of selection results are based on NEDO-published materials. The source is indicated in the attached document.
For further details, please refer to the PDF document.