A artificial intelligence (IA) has surprised me more and more! In addition to being present in the interfaces of companies such as OpenAI, Google and Microsoft, it reaches unexpected heights, surpassing even human physical abilities.
Recent tests revealed a pioneering innovation: a robot developed by researchers at the ETH Zurich Institute, called CyberRunner, defeated a human in a physical challenge, marking a new point in the evolution of AI.
Historic feat of artificial intelligence
CyberRunner, a revolutionary creation, faced a game that required not only strategic thinking but also physical dexterity.
The objective was to navigate a ball through a mazecontrolling two knobs to guide the sphere along the correct path and avoid traps that would restart the path.
Surprisingly, the robot completed the challenge in 14.58 seconds, thus surpassing the human record time, which was 15.41 seconds, achieved by Lars Göran Danielsson, an experienced competitor of this game.
The robot’s learning was the result of 6 hours of continuous training, allowing it to become an expert in the challenge.
CyberRunner is able to learn and cheat
However, the machine was not limited to simply surpassing the human record. On his journey to victory, CyberRunner cheated by finding shortcuts in the game. This behavior led researchers to tweak the code to thwart the robot’s strategy.
AI has learned to anticipate movements and strategies, improving its spatial perception and making increasingly intelligent decisions.
Such a reinforcement learning approach enabled robot accumulate experiences, refining your skills and strategies in real time.
The advance opens doors for the application of artificial intelligence in the physical world, which can be adopted by other institutes for various studies.
Raffaello D’Andrea, a professor at the university and CEO of Verity, plans to make the robot’s AI code available for more researchers to explore its possibilities in different fields.
