While AI has beaten chess grandmasters in previous man-versus-machine experiments, it has never been able to test its physical ability too.
ADVERTISEMENTArtificial intelligence (AI) has for the first time beaten humans in a physical and mental task.
An AI robot has learned how to play a labyrinth game, using its robot hands to control two knobs to steer a marble through the maze without falling into the holes.
After learning from its mistakes and even trying to cheat the game, the robot successfully managed to complete the task from start to finish in six hours, less time than it took humans to win the game.
Researchers at ETH Zurich University in Switzerland unveiled the AI robot named CyberRunner on Tuesday.
While AI has beaten chess grandmasters in previous man-versus-machine experiments, it has never been able to test its physical ability as well.
Through the labyrinth, the robot can flex its fine motor skills and spatial reasoning muscles.
The robot learns through experience and uses the model-based reinforcement learning algorithm to get better at the game. It then recognises strategies and even masters how to use its motor hands, a skill which improves with each use.
The AI robot outperforms the previously fastest recorded time, achieved by an extremely skilled human player, by over 6 per cent, the researchers said.
“We believe that this is the ideal testbed for research in real-world machine learning and AI,” said one of the researchers professor Raffaello D’Andrea.
The researchers said they will open-source the project and make it available on the website.
“For less than 200 dollars, anyone can engage in cutting-edge AI research. Furthermore, once thousands of CyberRunners are out in the real world, it will be possible to engage in large-scale experiments, where learning happens in parallel, on a global scale. The ultimate in Citizen Science!” D’Andrea said.
