Embodied AI unveils wheeled humanoid ‘AIR’ for factory work

Humanoid.Guide reported on November 7, 2026, that Swiss company Embodied AI had unveiled ‘AIR,’ a humanoid for factory work. The company said development took 9 months, from research to hardware. AIR moves on wheels rather than legs and has tendon-driven arms, with motors, gearboxes and electronics housed in its torso. It was designed in line with ISO/TS 15066:2016, and CE marking is planned for 2027.

Embodied AI says AIR is intended for repetitive manual tasks that frequently change and are difficult to handle with conventional automation. The extent to which its system for learning from human operation recordings and its design for yielding to force on contact will work in actual factories will be key to assessing where it can be deployed.

The reported design priorities for AIR are not bipedal walking but factory operations and behavior on contact. The configuration puts the drive system and electronics in the torso, uses tendon-driven arms to reduce their weight, and relies on wheels on flat factory floors. Rather than competing on the novelty of its mobility, the design appears tailored to repetitive tasks performed near workers. However, the description of the arms yielding to force on contact reflects the company’s design intent; it does not provide measured safety results or specify the conditions under which the design applies. We also covered the same robot in Humanoid Guide introduces Embodied AI’s ‘AIR’ the previous day. This article reports the unveiling, and the comparison with the previous day’s coverage does not establish that any new specifications have been added. Rather, while it is clear that articles about the same robot appeared on consecutive days, they do not constitute independent performance evaluations. Another related article, DeepMind executive discusses progress in robotics, covers progress in robotics broadly and does not provide evidence of AIR’s technology or deployment record. On task learning, the article says Embodied AI’s AI stack learns from each human-operated recording and can typically perform tasks autonomously within 1 month. It also says operators remain available to handle unfamiliar situations. If this account is accurate, deployment would involve a stage in which people initially operate the robot, followed by one in which the robot takes over after learning. The focus of evaluation, therefore, includes not only the learning period but also how often retraining is needed when task conditions change and how often human intervention is required. The article does not specify the range of tasks, task duration, success rate or volume of training recordings. Assessing factory use will require information on the arms’ payload and reach, travel speed on wheels, test conditions for contact safety and real-world operating data for each task. The article’s reference to a design in line with ISO/TS 15066:2016 and its planned CE marking in 2027 are separate matters; the latter does not mean certification has been completed. The excerpt provided also does not confirm production scale or operating rates. At this stage, the company’s descriptions of its design approach and learning method should be distinguished from validation of the robot’s effectiveness in factories.