Physical AI's hype vs. reality: inside Yaskawa's data-driven robot factory
A Nikkei Chinese column visits Yaskawa Electric's new factory, opened in July at its Kitakyushu head-office campus in Kurosaki, describing it as possibly Japan's first true "physical AI" factory: it runs with roughly half the workers of a conventional plant and, instead of a conveyor line, uses AI-equipped robots working in cell-sized (about 7.29 sq m) stations to assemble high-performance motors and robots, with robots able to detect errors and rework on their own and unmanned operation seen as a future possibility. The article says the key is linking robots to factory information systems to continuously collect production data — a "data factory" that makes the AI smarter. Chairman and president Hiroshi Ogasawara is skeptical of near-term rollouts, citing two obstacles: robotic hands (human fingers have 27 bones and 37 muscles, with sensor and actuator costs still limiting) and data consolidation — deduplication and unified identifiers across departments and factories took Yaskawa eight years, originally to compete with Chinese manufacturers' costs rather than for AI.
The case suggests physical AI adoption hinges more on data governance and hardware costs than on model capability alone, a useful reference for manufacturers weighing investment timing.
在日本北九州市黑崎地带，工业机器人巨头安川电机7月在面积很大的总部园区内启用了一座新工厂，这可能是日本第一座真正意义上的“物理AI（人工智能）”工厂。带着这样的想法，笔者前去参观。
让AI变得更聪明的工厂机制
该工厂的工人数量为传统工厂的一半左右。并未采用适合大规模生产的传送带方式，而是设置了多个被称为“单元（Cell）”、约四张半榻榻米（译者注：7.29平方米）大小的空间，由配备AI的生产用机器人逐一完成高性能电机和机器人等产品的组装。
每个单元由一名工人负责。工人负责监控作业进度，但即使机器人出现错误，也能自行发现问题并返工。因此，据说未来实现无人化生产也是有可能的。
这就是物理AI。其关键在于，用于生产的机器人与工厂的信息系统相连接，并收集生产过程中的数据。持续生成学习素材，使AI变得更加聪明的“数据工厂”。