Artificial intelligence (AI) has been used for improved performance over many video games, digital mazes and even chess. However, physical skill is something that AI has not been able to accomplish.
Well, not until researchers at ETH Zurich did it with its AI robot called CyberRunner.
In what the team called a world’s first, the AI robot was tasked with learning and beating a labyrinth marble game, which requires physical skill to steer a marble from a starting point to an ending point without the marble falling into any holes on the board.
The movement of the ball can be indirectly controlled by two knobs that change the orientation of the board, but it requires fine motor skill and spatial reasoning.
How it trained
CyberRunner learns by capturing observations and receiving rewards based on its performance. Through a camera looking at the labyrinth, memory of playing the game is collected and the model-based reinforcement learning algorithm learns how the system behaves and begins to recognize which strategies are more promising.
The robot also uses two motors to play the game continuously and does not stop playing to learn as the AI runs algorithms while the robot plays the game.
The AI robot learned on a real-world labyrinth in about 6.06 hours, comprising 1.2-million-time steps at a control rate of 55 samples per second. The AI robot outperformed the previously fastest recorded time, achieved by a human player, by more than 6%.
The performance could have been better as well as the AI found ways to skip certain parts of the maze. However, the researchers instructed the AI not to take any shortcuts.
“We believe that this is the ideal testbed for research in real-world machine learning and AI,” said Raffaello D’Andrea, professor at ETH Zurich. “Prior to CyberRunner, only organizations with large budgets and custom-made experimental infrastructure could perform research in this area. Now, 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.”
Learn more about the AI roboat at www.CyberRunner.ai.
