Robots are beginning to sell coffee at tourist sites, hand out medicine at unmanned pharmacies, and inspect power transmission equipment in China. Chinese manufacturers shipped 18,500 humanoid robots in the first half of 2026, exceeding the global total for all of 2025 in six months. More than 140 companies in China work on humanoid robots.

China business support firm SUGENA held an online seminar, “Physical AI: A Complete Guide,” on September 29, 2026. Its president, Isao Sugena, mapped humanoid functions to parts of the human body and explained the strengths of China, the United States, and Japan, as well as obstacles to adoption. He named five reasons China can build robots quickly, cheaply, and at scale: government policy; funding concentrated by governments and investment funds; supply chains developed through the electric vehicle industry; demand generated by orders from state-owned companies and local governments; and a willingness to release products at 70% completion and improve quality through real-world use.

Sugena said China’s 2026–2030 15th Five-Year Plan classifies embodied intelligence, or physical AI, as a future industry. He cited new energy vehicles, named in the 13th Five-Year Plan for 2016–2020, as an example of a sector that later became a major industry. However, he cautioned that production capacity alone does not make robots capable of working in real settings.

Sugena grouped robot functions into a “brain” for information processing, “nerves” for transmitting commands, “muscles” for moving joints, “sensory organs” for detecting the surroundings, and a “body” that integrates the components. The brain’s cerebrum recognizes and assesses its surroundings and determines actions; it consists of an AI model and an AI computing board that runs on the robot.

AI models have expanded from large language models (LLMs) to visual language models (VLMs), which add vision, and then to visual language action (VLA) models, which also produce actions. China’s LLM performance is approaching that of the United States, while global leadership in VLA for controlling robots remains unsettled. Research is also progressing in Japan, but few VLAs have been publicly announced as installed on real robots. NVIDIA’s Jetson is becoming the de facto standard for AI computing boards, with many Chinese humanoids using it as Chinese-made alternatives are also being developed. Japan has no company mass-producing AI computing boards for robots.

For posture and joint control, corresponding to the cerebellum, control software and servo drivers manage joints at millisecond-level intervals. Cloud communication is too slow for this task, so processing must happen inside the robot; Sugena identified this as a major challenge. Microcontrollers and power semiconductors that carry commands to the joints do not necessarily need to be cutting-edge, and Japan has strong companies in these fields. Japan also has strengths in sensor components. Sugena said, “For now, the brain has to be bought. But nerves and sensory organs can be made domestically.”

Robots are beginning to sell coffee at tourist sites, hand out medicine at unmanned pharmacies, and inspect power transmission equipment in China. Chinese manufacturers shipped 18,500 humanoid robots in the first half of 2026, exceeding the global total for all of 2025 in six months. More than 140 companies in China work on humanoid robots.

China business support firm SUGENA held an online seminar, “Physical AI: A Complete Guide,” on September 29, 2026. Its president, Isao Sugena, mapped humanoid functions to parts of the human body and explained the strengths of China, the United States, and Japan, as well as obstacles to adoption. He named five reasons China can build robots quickly, cheaply, and at scale: government policy; funding concentrated by governments and investment funds; supply chains developed through the electric vehicle industry; demand generated by orders from state-owned companies and local governments; and a willingness to release products at 70% completion and improve quality through real-world use.

Sugena said China’s 2026–2030 15th Five-Year Plan classifies embodied intelligence, or physical AI, as a future industry. He cited new energy vehicles, named in the 13th Five-Year Plan for 2016–2020, as an example of a sector that later became a major industry. However, he cautioned that production capacity alone does not make robots capable of working in real settings.

Sugena grouped robot functions into a “brain” for information processing, “nerves” for transmitting commands, “muscles” for moving joints, “sensory organs” for detecting the surroundings, and a “body” that integrates the components. The brain’s cerebrum recognizes and assesses its surroundings and determines actions; it consists of an AI model and an AI computing board that runs on the robot.

AI models have expanded from large language models (LLMs) to visual language models (VLMs), which add vision, and then to visual language action (VLA) models, which also produce actions. China’s LLM performance is approaching that of the United States, while global leadership in VLA for controlling robots remains unsettled. Research is also progressing in Japan, but few VLAs have been publicly announced as installed on real robots. NVIDIA’s Jetson is becoming the de facto standard for AI computing boards, with many Chinese humanoids using it as Chinese-made alternatives are also being developed. Japan has no company mass-producing AI computing boards for robots.

For posture and joint control, corresponding to the cerebellum, control software and servo drivers manage joints at millisecond-level intervals. Cloud communication is too slow for this task, so processing must happen inside the robot; Sugena identified this as a major challenge. Microcontrollers and power semiconductors that carry commands to the joints do not necessarily need to be cutting-edge, and Japan has strong companies in these fields. Japan also has strengths in sensor components. Sugena said, “For now, the brain has to be bought. But nerves and sensory organs can be made domestically.”