Watch an AI-leveraging robot beat humans in this classic maze puzzle game

Artificial intelligence programs easily and consistently outplay human competitors in cognitively intensive games like chess, poker, and Go—but it’s much harder for robots to beat their biological rivals in games requiring physical dexterity. That performance gap appears to be shortening, however, starting with a classic children’s puzzle game.

Researchers at Switzerland’s ETH Zurich recently unveiled CyberRunner, their new robotic system that leveraged precise physical controls, visual learning, and AI training reinforcement in order to learn how to play Labyrinth faster than a human.

Labyrinth and its many variants generally consist of a box topped with a flat wooden plane that tilts across an x and y axis using external control knobs. Atop the board is a maze featuring numerous gaps. The goal is to move a marble or a metal ball from start to finish without it falling into one of those holes. It can be a… frustrating game, to say the least. But with ample practice and patience, players can generally learn to steady their controls enough to steer their marble through to safety in a relatively short timespan.

CyberRunner, in contrast, reportedly mastered the dexterity required to complete the game in barely 5 hours. Not only that, but researchers claim it can now complete the maze in just under 14.5 seconds—over 6 percent faster than the existing human record.

The key to CyberRunner’s newfound maze expertise is a combination of real-time reinforcement learning and visual input from overhead cameras. Hours’ worth of trial-and-error Labyrinth runs are stored in CyberRunner’s memory, allowing it learn step-by-step how to best navigate the marble successfully along its route.

[Related: This AI program could teach you to be better at chess.]

“Importantly, the robot does not stop playing to learn; the algorithm runs concurrently with the robot playing the game,” reads the project’s description. “As a result, the robot keeps getting better, run after run.”

CyberRunner not only learned the fastest way to beat the game—but it did so by finding faults in the maze design itself. Over the course of testing possible pathways, the AI program uncovered shortcuts allowing it to shave off time from its runs. Basically, CyberRunner created its own Labyrinth cheat codes by finding shortcuts that sidestep the maze’s marked pathways.

CyberRunner’s designers have made the project completely open-source, with an aim for other researchers around the world to utilize and improve upon the program’s capabilities.
“Prior to CyberRunner, only organizations with large budgets and custom-made experimental infrastructure could perform research in this area,” project collaborator and ETH Zurich professor Raffaello D’Andrea said in a statement this week. “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.”