Partnerships
J-POWER, Hitachi and Macnica Agree on Physical AI Platform Study

Key PointsThe platform trains AI in simulation then validates it on real equipment through a…
View more
- The platform trains AI in simulation then validates it on real equipment through a Sim2Real cycle.
- Edge inference on NVIDIA Jetson lets drones operate without continuous cloud connections at infrastructure sites.
- J-POWER will supply real power facilities as testing grounds for the first drone demonstrations.
J-POWER, Hitachi and Macnica announced on November 19, 2026 that they had agreed to jointly study a physical AI development platform for critical infrastructure, with next-generation transmission-line inspection drones as the first proposed use case.
At this stage, the agreement covers joint studies and demonstration planning. Physical AI applies artificial intelligence to machines operating in the physical world, such as robots and drones. The companies expect it to support more advanced and autonomous inspection and maintenance in Japan, where critical infrastructure sectors face shortages of skilled workers and aging equipment.
Simulated Training for Real Equipment
The proposed platform would train AI in simulation for infrastructure operation, maintenance and control, validate the models and install them on equipment and robots. Field data would then feed further training and validation, creating a simulation-to-reality cycle known as Sim2Real.
The companies say training and validating models solely in real environments creates constraints in time, cost and safety. The design therefore combines digital twins, which replicate real environments, with a platform intended to manage confidential data securely while supporting continuous AI training, validation and operation.
Technologies under consideration include NVIDIA’s Omniverse simulation libraries, Cosmos world-model platform and Isaac robotics development platform. Hitachi would combine these with HMAX Data Fabric, which organizes field data and expert knowledge for AI use, and HMAX AI Operations, which monitors AI operating conditions and model-specific risks.
An edge environment based on NVIDIA Jetson would support studies of real-time equipment recognition, anomaly detection, autonomous movement and path planning. The companies intend for those inference tasks to run on field equipment, so robots and drones would not need a continuous cloud connection at infrastructure sites.
Three Roles in the Drone Study
The first initiative will examine demonstrations of next-generation transmission-line inspection drones using real J-POWER Group power generation and transmission facilities as testing environments. J-POWER will plan use cases, provide those facilities, and evaluate and verify the study results.
Hitachi’s Forward Deployed Engineers team will refine and evaluate use cases, study the platform architecture, and build and verify the platform, drawing on power-sector knowledge and expertise in IT, operational technology and AI. Macnica, an authorized NVIDIA distributor, will support the design, construction and operation of the Sim2Real demonstration environment and help examine robotics use cases and applications.
Expansion Beyond Power Lines
Results from the initial drone work are intended to guide practical applications in inspection and maintenance automation, advanced autonomous control and disaster-recovery support. The companies also plan to expand partnerships beyond electricity into other critical infrastructure fields, including railways and industrial plants.
Implementation on the AI data center that J-POWER and Hitachi are separately studying is also under consideration. The companies’ AI Factory concept spans data collection, learning, validation, inference and continuous improvement.
Running the platform on that data center is intended to integrate operation of the digital infrastructure supporting AI with power infrastructure. That would deepen coordination between power and communications that the companies call watt-bit collaboration.
J-POWER executive officer Shingo Koizumi described the aim of developing the initiative into a landmark model for physical AI in critical infrastructure. The proposed drone demonstrations are the first practical test of that ambition: they must establish whether simulation-trained models can work dependably on operating power assets before the same development process can support the partners’ wider plans.












