Figure 02 has worked only for an hour. The disturbing thing is that your brain already remembers and your hands “feel”

You can disagree. You may think that a lot is still missing. But it is difficult to ignore the direction in which we advance: one in which Humanoid robots They will be part of our daily lives. When technology is mature, when costs cease to be an obstacle, the decision will no longer be technical. It will be ethical. How far can the robots arrive? And who will put the limits?

Imagine the scenarios. Robots turned into soldiers, deployed in conflicts, alone or next to human troops. Robots as emotional companions, not only for older people, but for anyone who lives alone. Robots that clean, cook, organize. Robots in factories, tireless, constant, replacing tasks that we did before.

The surprising thing is that all this is already beginning to happen. Today, the robot Figure 02, that since last year he works in a BMW factorycan operate autonomously, classify objects and do it with precision thanks to what Its developers call “Touch” and “short -term memory”.

In 2018, ‘Detroit: Become Human‘We thought about science fiction. Even then, not its creator, David CageI would have opted that in 2025 we would be seeing something like that in the real world. But it has happened. And the most fascinating thing is that Figure does not walk alone. Behind his gestures, there is something else. A neuronal network that, as we will see soon, is the true protagonist.

Helix, the ‘brain’ behind Figure 02

That neuronal network has its own name: Helix. It is the brain behind the robot. The person responsible for their movements look more and more natural. And what has achieved In just three months In a logistics environment it is difficult to ignore. Helix has not only learned to manipulate objects with skill. Has learned to Understand the context. To adapt. To act as if remembered what he did before. And the most surprising: to make it better and better, faster and less mistakes.

At first, Helix faced a simpler catalog. But little by little he has learned to work with a greater variety of packages: From rigid boxes to Soft envelopes or deformable plastic bags, which are much more difficult to hold and position. Some wrinkle, others bend, others slide easily. For a robot, that is a real challenge.

And yet, Helix adapts. Adjust the way in which each object grabs, changes its strategy if the package is flatter or softer, even knows when it is convenient to take a small turn or use a more precise type of clamp. He does everything on the flight, without anyone telling him what kind of package he has in front.

That behavior has not been programmed line per line. Has learned it by observing. In total, he was trained with 60 hours of human demonstrations. And with each new example, it was better understood what to do in each case. Thanks to that, the average time it takes to process a package has been down significantly.

But Helix is ​​not just fast. It is also necessary. The labels, which were previously oriented before, now 94.4 % of the time are correctly positioned. As? Because the robot has learned, among other things, Small wrinkled envelopes Before trying to scan. A slight pressure on the plastic is enough for the barcode to be visible. It is a minimum gesture, but very revealing.

That learning relies on something fundamental: memory. Helix has a vision system that not only analyzes what he sees at the time, but remembers what he saw a few seconds ago. It is as if I had a Short -term visual memory. Thanks to her, you can make smarter decisions. For example, if it detects that a package was already rotated at a certain angle, it does not turn it again. If you did not find the label at first, remember where it could be and prove in that direction.

Another key change has been to incorporate the history of its own movements. Before, each action was an isolated fragment: see, act, see again. Now, Helix Remember in what position your arms werehis torso or his head makes a moment, which allows him to move more fluidly. If something leaves the plan, if the package slides or the grip is not perfect, it can correct in real time without starting from scratch.

And the most recent: The sense of touch. Not a human touch, of course, but a way to detect how much pressure is exerting on an object. That strength feedback allows Helix Pause to feel resistance, or adjust it if the object weighs less than expected. Thanks to this, you can manipulate more carefully and adapt to differences in weight or rigidity.

And this is not theory. Figure 02 has already been put to the test in real conditions. During an hour session, The robot operated without interruptions In a logistics environment, classifying packages autonomously while using all its new capacities: its sense of touch, its short -term visual memory, its ability to correct errors on the march. A continuous demonstration of fluid work, precise, without direct intervention.

The images of that session, which we have included in the video that accompanies this article, show more than a robot running. They show a neuronal network facing the chaos of the physical world and leaving airy. They show Figure 02 making decisions, adapting, acting with a level of autonomy that until very recently would have seemed science fiction.

Tesla has a OptimusBoston Dynamics to AtlasAgility Robotics A Digit. All compete to lead the career of the Humanoid robots. But in parallel, Figure 02 advances through a lane without events, without background music, but with results. With the passage of time we will see how all this evolves.

Images | Figure ai

In Xataka | Google does not want AI to stay on the screen: Gemini Robotics is its plan for robots to act in the real world

In Xataka | The Nvidia’s supercomputer costs millions of dollars. And to work we use a switch with three kilometers of cable