Since humanoid robots emerged, they were good at two things separately: they knew how to walk or they knew how to manipulate objects, doing both at the same time with fluidity, balance and intention was always one of the biggest bottlenecks of robotics, now Figure claims to have crossed exactly that frontier.

The company revealed the Helix 02, the next generation of its artificial intelligence system capable of controlling a full-body humanoid as a single integrated organism, the announcement was accompanied by a direct video without cuts, showing a Figure 03 robot, opening a dishwasher, removing dishes, organizing utensils and then reloading everything again completely autonomously.

That is, no operator, no teleoperation, no step-by-step command, just vision, sensors and artificial intelligence, according to Figure, Helix 02 represents the definitive transition from robots that execute isolated movements to robots that reason and move at the same time.

At the center of this is an architecture inspired [music] by the human brain itself, a slower layer decides what needs to be done, another transforms moving images and a third extremely fast, takes care of balance, strength and contact with the physical world in milliseconds.

In tests this means that the robot no longer needs to stop to think, then balance, then act, all this happens together, the most impressive result of this approach is the so-called crazy manipulation, while walking the robot already adjusts its arms, while holding an object, it already corrects posture, while reaching for something, it already compensates the weight with the trunk and legs.

In the demonstration, the humanoid even uses its hip to close a drawer and its foot to lift the dishwasher door when its hands are occupied. These are not programmed movements one by one, they are emergent behaviors, as if the entire body were being used as a single tool. This autonomy is also reinforced with increasingly refined sensors in the tips of the fingers. Touch sensors perceive forces as light as the weight of a paper clip. reach, this allows, for example, taking a single pill, holding fragile objects or searching for parts inside messy boxes.

According to Figure, part of this advance comes from abandoning thousands of lines of traditional code, instead of fixed rules, a single neural network was trained with more than 1000 hours of human movements and when a robot begins to coordinate vision, touch, balance and intention, it becomes increasingly difficult to see artificial intelligence only as software.