Macnica and TGES begin joint demonstration project to realize autonomous inspection of plant and utility equipment using physical AI - verifying next-generation inspection and operation models using humanoid robots and digital twins -

Tokyo Gas Engineering Solutions Co., Ltd. (President: Takayuki Uenaka, hereinafter "TGES") and Macnica (President: Kazumasa Hara, hereinafter "Macnica") have signed a Memorandum of Understanding (MOU) to promote the use of physical AI in daily inspection operations of plant and utility facilities, and will commence joint demonstration (hereinafter "this demonstration").

TGES and Macnica collaborate to verify a next-generation inspection and operation model.

Image of a patrol inspection

This demonstration project aims to verify the feasibility of using humanoid robots in the daily inspection of plant and utility facilities, with the goal of realizing autonomous inspections using physical AI in the future. TGES will combine its inspection knowledge and operational expertise, cultivated over more than half a century of O&M for district heating and cooling systems and on-site energy service management, with Macnica 's AI and robotics technologies. The project will begin with verification of basic movements such as autonomous walking. Going forward, the scope of verification will be gradually expanded to include patrols, information acquisition, and equipment operation, with the aim of establishing an inspection model in which humans and robots collaborate.

Verification in a virtual space (digital twin)

Specifically, we envision building a digital twin environment ^*2 by utilizing environmental data of actual facilities acquired through 3D laser scanning and the NVIDIA Omniverse™ ^*1 library. We will gradually improve the robot's operational accuracy through the Sim2Real ^*3 cycle, which reflects data acquired in the actual field into simulations and returns the learning and verification results in the virtual space to the actual robot. We will also utilize simulations in which multiple robots are operated in parallel in the virtual space to improve the efficiency of learning and verification.
Existing plant and utility facilities are designed with the assumption that people will move around and operate them, and therefore have stairs, steps, doors, narrow passages, and switches positioned for human operation. By utilizing humanoid robots with similar forms and ranges of motion to humans, we aim to adapt to these environments without significantly modifying existing facilities, and in the future, have them handle everything from inspection and information acquisition to equipment operation.

■ Background of the initiative
Maintaining stable operation of plant and utility facilities requires daily inspections, including patrols, instrument checks, visual inspections, and record keeping. However, in 24-hour operation environments, securing inspection personnel, passing on skilled techniques, and managing the burden of nighttime and holiday work are challenges to sustaining facility management. To address these challenges, we believe it is necessary to shift operations where robots handle routine movement, inspection, and record-keeping tasks, allowing humans to focus on more advanced tasks such as anomaly detection, root cause analysis, improvement planning, and maintenance planning, rather than simply replacing human work with robots.