For some humans, folding towels may seem like the most banal task in the world, but when a humanoid robot manages to do it calmly, patiently, and even with a certain human touch, the story changes completely. That was exactly what we saw in Helix, the vision, language and action model created by the North American company Figure.

They placed the robot in front of a pile of towels and showed that human-shaped machines can be much closer to executing tasks in our daily lives than we imagined. Helix is ​​the brain behind the robot and I'm not talking about rigid programming with limited commands, food, vision, language understanding and movement control in a single generalist system.

This means that the humanoid does not need to be programmed step by step for each action, it is enough to interpret the environment, understand the instruction and execute the task. A detail that stands out is how he manages to manipulate wrists, fingers, torso, even the head and rapid and continuous movements, reminiscent much more of a human being than a machine.

In practice, Figure's humanoid takes the towels one by one, smoothes the fabric, adjusts the folds when necessary and organizes them and a basket. The curious thing is that he does not seek aesthetic perfection, but rather efficiency, exactly as any person would do on a daily basis.

That naturalness comes from a system divided into two blocks, S1, responsible for executing quick actions in real time and S2, which functions as a deeper reasoning layer, trained with language and scene perception.

While S2 thinks, S1 ensures that actions come out at the right time, without interrupting the flow. Another impressive point is that Hélix was trained to obey natural language commands, meaning he can simply say what you want him to do. In addition, it can work in an ensemble with other robots sharing the task in real time.

This opens the possibility of future axis humanoids performing domestic or industrial activities as a team without relying on repetitive demonstrations or constant reprogramming.