An AI-powered robot has beaten a team of professional curling players in a major real-world breakthrough for robotics.
The Curly robot, developed by a joint German-South Korean research team, used deep reinforcement learning to learn the best way to slide stones along the ice and land on the target.
In a match against top-ranked opponents, the robot won three out of the four games.
“These results indicate the gap between physics-based simulators and the real world can be narrowed,” the researchers wrote in a paper.
Curly makes use of two robots – a “Thrower-Curly” and a “Skip-Curly” – in order to both assess the target area and to slide the stone towards it.
The only difference to matches at professional competitions was the lack of sweepers – the players who brush the ice ahead of the stone as it slides in order to influence its speed and trajectory
The researchers from Korea University and the Berlin Institute of technology trained the robots AI on a computer game that simulates the physical properties of the stones and the ice.
Before each real-life game, Curly threw one stone in order to judge the current conditions of the arena’s surface.
“In curling, the environmental characteristics change at every moment, and every throw has an impact on the outcome of the match. Furthermore, there is no time for relearning during a curling match due to the timing rules of the game,” the researchers said.
“Our proposed adaptation framework extends standard deep reinforcement learning using temporal features, which learn to compensate for the uncertainties and nonstationarities that are an unavoidable part of curling.”
The research was published in the journal Science Robotics on Wednesday.