A new generation of robots is moving away from reliance on rigid commands, instead learning directly from experience by observing their environment, making mistakes, and refining their movements while working. The concept mirrors the way humans learn: rather than memorizing a fixed sequence of movements, the robot builds its skills through continuous interaction with its surroundings.

Founded by researchers led by Ross Ted Drake—a global authority in robotics and artificial intelligence—the company is developing a platform capable of adapting machine behavior as new challenges arise on the factory floor. At the core of this system are "large behavior models"—neural networks trained to understand not only images but also the physics of the tasks they perform. Instead of following pre-defined scripts, these models analyze the environment, calculate various movement options, and instantly select the most suitable strategy for each situation.

Collectively, algorithms known as diffusion policies continuously refine the force, torque, and trajectory of movements, enabling robots to handle minor variations without interrupting operations; each attempt generates new data, and when a task is not performed optimally, the system uses that result to adjust its behavior for future executions. Over time, the machine becomes more efficient without the need for lengthy reprogramming processes.

It is precisely this capacity for continuous evolution that distinguishes these new robots from previous generations. Interest in this approach is already driving billions of dollars in investment; meanwhile, companies like Robot.com are accelerating this transformation with humanoid platforms based on the "Robot-as-a-Service" model. One example is Arnoid, developed to perform various functions in restaurants, laundries, and logistics centers; it features arms with seven degrees of freedom and can be deployed within a few weeks following an assessment of the environment.