Platform Structure

Actuators determine how robots exert force and move in the real world. Nonlinear characteristics such as backlash, noise, and elastic deformation must be addressed at the actuator layer first.

By correcting these physical characteristics at the actuator layer, ONE PhAI enables the sensor and AI layers above to handle clearer signals and data. This improves the stability and predictability of the entire system.

Through standardized hardware interfaces, various actuator technologies can be integrated, and the same control principles can be maintained even as robot forms change.

Sensors are the key element for detecting the physical state of robots and human intent. On top of the clean signals organized by actuators, sensors play the role of determining what constitutes core data.

For this purpose, joint encoders, IMUs, and motor current sensing are used alongside force and torque information. This is because detecting the force and intent conveyed by humans during expert teaching is considered critical.

A sensor structure designed with an understanding of the physical characteristics of hardware is advantageous in determining what constitutes core data.

What data is created and how it is created is the starting point of Physical AI learning. In ONE PhAI, the value of data depends on the conditions of its creation rather than its quantity.

With actuators correcting physical errors and sensors filtering meaningful signals, expert actions record both human intent (What, Why) and the robot's physical optimization (How). ONE PhAI believes that data created in this way leads to Physical AI learning.

In ONE PhAI, AI is not structured to directly handle all physical variables. Because the actuator layer first eliminates physical errors, expert actions are recorded as data without distortion. AI can then focus on high-level judgment and learning on top of this refined data.

If the actuator layer fails to handle physical errors, AI ends up learning the robot's physical noise along with the expert's intent. A structure where each layer focuses on its own role leads to the stability of the entire system.

In ONE PhAI, humans set the purpose (Why). Robots physically optimize the method (How) to realize that purpose.

In ONE PhAI, the relationship between humans and robots is closer to teaching-learning than command-execution. Humans convey subtle movements and intent to robots, and robots follow by physically compensating for those movements.

As data accumulates, the domain that robots can handle autonomously expands. The proportion of human teaching gradually decreases, and the proportion of robot autonomy naturally grows. ONE PhAI believes that through this process, Physical AI can move beyond the experimental level to be repeatedly operated in actual field environments.