The latest in Robotics of South Korea is not humanoid or works in factories. Does something out of the ordinary: Parkour

The physical state. As detailed in A video available on YouTube, the first step is taken by the planner, which generates possible routes from a map of the environment. That map is continuously updated with sensor and simulations data. Then, a neuronal network rules out risky options and stays with the most efficient.

The tracker, on the other hand, guides the precise movements of the robot. It was trained through reinforcement learning, a technique based on trial and error, which prepared it to adapt to dynamic and challenging scenarios. To save calculation time, Raibo reuses their own footprints: the hind legs step on where they did the front before.

Raibo training simulation

As they count, the robot was able to run on irregular surfaces, overcome stones, cross inclined ramps, climb stairs and even jump gaps of more than one meter. It reached a speed of 2.7 meters per second. And the most surprising: if the goal moved, the robot detected and recalculated its route without help, without stopping and without losing control.

Meanwhile, robotics does not stop in the rest of the world. Raibo’s advance is not an isolated case. It is part of a global wave of developments in which robotics and AI are more and more intertwined. Without ia, robots would continue to be little more than a set of sensors and engines. With AI, they are able to interpret their environment, make decisions and execute complex tasks with autonomy.

Companies like Google are betting on it. With Gemini Roboticstheir last great project, have designed a system capable of controlling different types of robots in real time, understanding human language, pointing to 3D objects and adapting to new situations without prior training. The search engine giant says that his performance in unforeseen tasks doubles that of previous models. For now, this technology is in the test phase, but Google already collaborates with companies such as Apptronik or Boston Dynamics to integrate it into advanced humanoids.

China is also accelerating. And it is not the only region that is investing strong in this direction. In China, humanoid robots not only train: they compete. A few weeks ago, The country celebrated a Kickboxing tournament Between four G1 robots by Unitree Robotics. He was broadcast live and showed how these machines were able to dodge blows, get up alone after falling and continue fighting with surprising agility. They are 35 kilos robots and up to 23 degrees of freedom, designed with state -of -the -art sensors, and according to organizers, new multisport competitions are already in preparation.

Robots developed in China in a boxing ring

And there are already robots working in real factories. Meanwhile, in the United States, some humanoid robots have left the laboratory and are entering real factories. One of them is Figure 01, that has long worked in a BMW plant in South Carolina. This robot, developed by the Figure company, can open doors, climb stairs and manipulate objects autonomously. Of course, it still moves slowly and needs to be connected by cable permanently.

Parkour as the advancement of the future. All this helps to understand why Raibo’s case is so fascinating. It is not humanoid, nor has it been created for industry or home. But it shows that, combining real -time decision algorithms with light hardware and advanced training, it is possible to create machines that not only execute orders, but also improvise and have an agility that causes vertigo. Touch to wait to know how these advances find a place in really useful applications. There will be the real leap.

Images | Robotics & Artificial Intelligence Lab

In Xataka | Nvidia desperately seeks engineers for its Taiwan R&D center. They even accuse you of “stealing them” to TSMC