Oriented on making people closer

On June 28th, the final of the World Robot Football League, RoBoLeague, took place in Beijing. For the first time, four Chinese teams of autonomous humanoid robots competed in a historic 3v3 match where all decisions were made by AI without human intervention.

Each robot was equipped with multispectral cameras operating at 120 frames per second and a computer vision system based on YOLOv7 neural networks. These networks were trained on 5,000 hours of game scenarios, enabling accurate recognition of the ball and the field. Gyroscopes and accelerometers ensured stability while moving at speeds of up to 2 m/s. The organizers highlighted that this marked China's inaugural match featuring fully autonomous humanoids, serving as a test of their endurance and agility.

The match comprised two 10-minute halves. The robots utilized a modified AlphaZero algorithm for their tactics, independently making decisions regarding passes and shots. Energy-efficient servomotors provided up to 90 minutes of autonomous operation. Over a year of training, the machines learned to recognize game situations with 40% greater accuracy, and their reaction speed tripled, allowing the AI to "think through" a pass faster than a human. The key to this breakthrough lay in a new neural network architecture that not only analyzed the field but also predicted the actions of their "teammates" with 85% accuracy. Lidars ensured millimeter-level shot precision, while thermal sensors monitored heat levels.

Cheng Hao, a representative from Booster Robotics, stated that the organizers are continuing to refine the technology for more complex game tasks. The dynamic nature of recent exhibition games was so high that spectators occasionally forgot about the absence of humans on the field. This match represents a significant milestone ahead of the World Humanoid Robot Games, scheduled for August.

ORIENT

Photo: nypost