Prox Industries Provides Technical Support for Mitsubishi Electric’s Physical AI R&D

We have not been able to verify the primary source, so please refer to the original report for details.

Mitsubishi Electric’s combination of simulation and implementation support in Physical AI research and development can be seen as an effort to strengthen verification before deployment on physical systems. According to the announcement, the work targets coordinated operation by multiple robots and is characterized by an integrated approach to perception, decision-making, communications and control.

What is new in this announcement is that Prox Industries provided technical advisory support for Mitsubishi Electric’s Physical AI research and development from both the simulation and implementation sides. The plan outlined is not simply to get robots moving, but to use NVIDIA Omniverse and Isaac Sim to validate coordination among multiple robots, linking perception, decision-making, communications and control. At the same time, this appears to build on Mitsubishi Electric’s existing Physical AI and robotics R&D rather than represent a ground-up business shift, making that work more concrete through a verification platform. In related coverage, Mitsubishi Electric outlined autonomous screw-tightening and shelf-stocking by humanoid robots, as well as plans for unmanned factories, in [Mitsubishi Electric showcases physical AI at CEATEC 2026, also presenting unmanned factories and next-generation data centers](/en/news/mitsubishi-electric-to-showcase-physical-ai-at-ceatec-2026-alongside-l-35a645). In HoloLab joins Mitsubishi Electric’s MELFA robot partner program, launches remote-operation software “arm-pilot”, the use of remote-operation software and operational data was also highlighted. Prox Industries’ support can be seen as linking such demonstrations and partnerships to simulation-based verification before deployment on physical systems. Across the industry, the focus is shifting from the performance of individual robots to the verification stack, including computing environments, training settings, communications and sensing. Because the work involves quadruped robots, robot arms, VLMs, LLMs and communications together, a key question is how thoroughly operating conditions can be refined before products are deployed. Coordination among multiple robots can make trial and error on physical systems costly, making it important to determine how well simulation-based iteration can reproduce real-world behavior. Open questions include which specific research themes the work was applied to, how many robots and what range of functions were tested, and how much of the trained models and operating conditions can be transferred to physical systems. The announcement describes the support but does not provide production volumes, deployment sites or specific physical-system evaluation results.

In Japan’s factory automation and robotics sectors, simulation may play a growing role in refining perception, communications and control, alongside testing on physical systems. Efforts by companies such as Mitsubishi Electric to consider verification platforms as well as robot hardware could expand the relative roles of domestic controls, sensors and industrial software.