Physical AI News (Japan)

Physical AI, robotics and compute — reported from Japan.

日本語
· Partnerships & M&A· 日本語で読む

J-POWER, Hitachi and Macnica Explore Physical AI Platform for Critical Infrastructure

J-POWER, Hitachi and Macnica are exploring the development of a platform for using physical AI in critical social infrastructure. The concept is to run AI trained and verified through simulation on equipment and robots, then feed field data back into retraining. As an initial use case, they are considering a demonstration of power line inspection drones using J-POWER Group facilities.

Key facts

  • On November 19, 2026, J-POWER, Hitachi and Macnica agreed to jointly explore how to build a physical AI development platform for critical social infrastructure.
  • As a first step, they are considering a demonstration involving next-generation power line inspection drones using J-POWER Group's power generation, transmission and substation facilities.
  • They are considering the use of NVIDIA Omniverse libraries, NVIDIA Cosmos and NVIDIA Isaac, among other technologies.
  • They are exploring an edge AI environment using NVIDIA Jetson, with a plan to process tasks such as equipment recognition, anomaly detection, autonomous movement and route planning in real time on site.
  • J-POWER will provide real-world sites; Hitachi will explore, build and validate the development platform architecture; and Macnica will support the design, construction and operation of Sim2Real demonstration environments, among other tasks.

What happened

On November 19, 2026, J-POWER, Hitachi, Ltd. and Macnica agreed to jointly explore how to build a development platform for physical AI used in critical social infrastructure. They are also considering implementing it in an AI data center (AI-DC) that J-POWER and Hitachi are exploring. As a first step, they are considering a demonstration involving next-generation power line inspection drones, using J-POWER Group's power generation, transmission and substation facilities.

Why it matters

The concept is to run AI validated through simulation in real-world settings for power line inspection drones and feed the resulting data back into training. For companies responsible for inspecting and maintaining power infrastructure, the effort explores a way to connect field data with AI development. The three companies also say they are considering an on-site inference environment using NVIDIA Jetson, with the aim of operating without relying on a constant cloud connection.

Analysis

The central idea is to connect the computing and training environment for AI models with the infrastructure sites where the models operate in a single improvement loop. The AI-DC is being considered as a platform for training, validation and operation; Omniverse, Cosmos and Isaac would support simulation, while Jetson would handle inference on site. The design also feeds data collected from real equipment back into retraining, making this more than an effort to simply equip drones with AI. However, the announcement concerns an exploration of platform development; it does not say that the demonstration or implementation in an AI-DC has been completed. Selecting power line inspection drones as the initial use case points to a division of roles that connects J-POWER's field assets, including power generation, transmission and substation facilities; Hitachi's IT, OT and AI expertise; and Macnica's support for building demonstration environments. The platform's effectiveness will depend on whether equipment and field data can be reflected in simulations and the resulting model validation can be applied to real equipment. Combining Hitachi's data platform and AI operations support is intended to support not only the accumulation of field data but also the handling of AI model operating status and risks. However, the announcement does not specify which data will be handled, at what level of detail, or how confidential information will be managed. In APTO selected for student physical AI challenge in GENIAC-PRIZE 2026, which Physical AI News reported, the challenge involved designing data for imitation learning and a robotic arm. The current plan goes beyond designing training data to cover infrastructure simulation, operation on real equipment and the feedback of field data, taking in a development cycle that includes real-world operation. The two efforts differ in scope and target, but share the question of how to connect physical environments with training data. The industry implication is a division-of-labor model in which infrastructure operators provide real-world sites, IT and OT companies build the platform, and technology support companies connect simulations with real equipment. To expand to multiple critical infrastructure sectors, key questions include whether validation results from power line inspections can be applied to other equipment and operations, and how differences between sites can be reflected in models. The next points to watch are the conditions and evaluation methods for the drone demonstration, the timeline for implementation in an AI-DC, procedures for data management and model updates, and safety and operating records on site. The announcement marks the start of discussions on a demonstration and does not provide performance results or a specific deployment timeline.

Details

The concept is a Sim2Real loop: train and validate AI in digital twins or simulation environments, run validated models on equipment and robots, and use data collected in the field in the next round of training and validation. The companies are considering using NVIDIA Omniverse libraries, NVIDIA Cosmos and NVIDIA Isaac, along with Hitachi's HMAX Data Fabric and HMAX AI Operations. For field use, they are exploring an edge AI execution environment using NVIDIA Jetson, with a plan to perform real-time inference for tasks such as equipment recognition, anomaly detection, autonomous movement and route planning, aiming to operate without relying on a constant cloud connection. J-POWER will plan use cases and provide real-world sites; Hitachi will explore, build and validate the platform; and Macnica will support the design, construction and operation of Sim2Real demonstration environments, among other tasks. The companies say they ultimately aim to integrate training, validation and operation on an AI-DC and expand into critical social infrastructure sectors beyond power.

Implications for Japan

For Japan's power infrastructure operators, the concept connects validation using real-world power generation, transmission and substation facilities with a training and operations platform on an AI-DC. A key question is whether the division of roles, with J-POWER providing field sites and Hitachi and Macnica handling the platform and demonstration environment, can be applied to domestic critical infrastructure beyond power line inspection.

Sources

Translated from the Japanese original: /news/j-pawaa-hitachi-makunika-juuyou-infura-muke-fijikaru-ai-kiba-8b9e5a. The Japanese version is authoritative if they differ.