Site icon Techno | Phil | oSoph

Round 1: How Robot Companies Troll Each Other

I have identified more than 140 manufacturers of humanoid robots for my upcoming book. Around 30 manufacturers were represented at CES 2026 in Las Vegas with 40 humanoids, plus just as many companies that manufacture robot components for them. And now Tesla CEO Elon Musk has announced that he will discontinue production of the Model S and Model X electric vehicles at the Fremont factory so that he can manufacture his own humanoid robot, the Tesla Optimus, there.

The competition is already in full swing, intensifying, and the knives are flying deep. Nothing makes this clearer than the new videos released by Figure.AI, a startup based in San Jose that is just 3.5 years old. In the video, we see the third generation of humanoid robots, the Figure 03, as a kitchen assistant. It opens and empties a dishwasher, stacks plates and glasses on a shelf, fills the dishwasher with glasses and plates, adds a detergent tablet, and closes the dishwasher door. So far, so good.

But at the same time, this video is also an impressive response to another robotics company whose headquarters are located just a few kilometers away from Figure.AI and which attracted a lot of public interest three months ago. 1X Technologies in Palo Alto had unveiled the NEO Gamma humanoid and touted it as a companion and household helper. The sleek promotional videos made everything look good, but the Wall Street Journal was more skeptical after testing it for several hours. The NEO Gamma, remotely controlled by a human operator, quickly revealed its limitations and clumsiness when dealing with a dishwasher.

I couldn’t help but compare excerpts from the two videos and show how Figure 03 and NEO Gamma perform the same task. The two videos were taken just three months apart and clearly show the differences and progress made in such a short time.

I selected the part from the videos where, after filling a dishwasher, they close it.

What makes the Figure 03 so special? Why can it do what the NEO Gamma cannot?

This is explained by Figure on their website. The operating system, or more correctly, the Vision-Language-Action model (VLA), Helix 02, takes care of the movement of the upper and lower body, the manipulation of objects with the hands, movement through space, and the analysis of a large amount of sensor data, ensuring that everything runs smoothly and simultaneously.

Helix 02 is a neural network that has been trained with thousands of hours of video data on such household tasks. Figure distinguishes between three systems:

System 0 (S0): operates at a frequency of 1 kHz and is responsible for balance, contact, and coordination of the entire body.

System 1 (S1): thinks quickly and translates perceptions at a frequency of 200 Hz into targets for the joints of the entire body.

System 2 (S2): thinks slowly about goals: interpreting scenes, understanding language, and putting behaviors in sequence. It translates a sentence such as “Take the bowl to the sink” into smaller individual steps such as “Grasp the bowl with your hands, turn your body 90 degrees, identify where the sink is, walk to the sink, reach into the sink with your arms, and carefully place the bowl down.

The capabilities of Figure 03 with Helix 02 are now also possible thanks to the inclusion of sensor data from the fingers. Videos with and without the finger sensor data can be viewed on the website mentioned above. The following video shows Figure 03 manipulating fragile glasses, which would have resulted in a broken wine glass without the finger sensors.

In any case, the advances in robotics are enormous. Tasks that seemed impossible for them to perform three months ago, where humanoids appeared clumsy and created more problems than they solved, can now be performed quickly and skillfully. And we are only at the beginning of the development of humanoid robots. As with (generative) artificial intelligence, which three years ago was still hallucinating and producing many errors, we are amazed today at the tasks AI is already capable of solving.

Exit mobile version