Japan Bets on Physical AI

Japan's robotics industry is focusing on developing physical AI, which integrates artificial intelligence with machines that can sense, decide, and act in the real world. This strategy aims to leverage Japan's manufacturing expertise against competitors in the United States and China. One example is Cinnamon 2, a humanoid robot under development by Donut Robotics.

Cinnamon 2 can replicate fluid human movements, comprehend spoken commands, identify people and gestures, and perform tasks suited for security, construction, nursing care, and factories. Currently under development, the robot is slated for official release by the end of 2026. Donut Robotics plans to unveil it in October 2026. The robot's AI is developed in-house, while its hardware is initially manufactured in China before assembly in Japan.

The robot features a vision-language model (VLM) that interprets both images and language, enabling it to provide detailed descriptions of its surroundings. This visual system can also aid in security tasks, such as photographing identified individuals and transmitting the images to management for suspicious activity reporting. Cinnamon 2 can recognize specific gestures from registered users, which can trigger predetermined movements.

This gesture control feature could be particularly useful at noisy construction sites where verbal instructions are challenging to hear. The robot's humanoid form offers advantages in environments designed for humans, such as uneven surfaces and locations inaccessible to wheeled robots. Physical AI differs from traditional industrial robotics, as it enables robots to perceive their surroundings, make decisions, and plan their own movements, rather than merely repeating programmed actions.

While Japan once led in humanoid and industrial robotics, other countries like China and the United States have made rapid advancements. Nonetheless, Japan retains an advantage in deploying physical AI due to its manufacturing sector's potential as a source of proprietary training data, especially in safety standards, industrial processes, and robot components.

Japan's government is also investing in a broader physical AI ecosystem by collaborating with major machinery manufacturers to collect learning data for autonomous robot and industrial equipment operation.