Meet Curly—the curling robot who can beat human teams

curly the AI curling robot throwing a rock down the ice Via Scientific American on YouTube

Curling—the only time throwing rocks at a house is encouraged. Many of us are familiar with curling, but a little recap never hurt anyone. The players slide a rock or stone down 45 meters of ice towards the target, which is called a house (if you didn’t understand the first line of this article go back and read it again. It’s funny, I promise). The goal is to get their team’s rocks closest to the centre of the house while also knocking their opponent’s rocks further from the centre. The sport is normally played by a team of four people, but it can also be played by a robot. Yup, you read that correctly. The machine is named Curly, and it has won 3-out-of-4 of its games when competing against top-ranked human curling teams.

According to Seong-Whan Lee, one of the researchers who co-authored a paper for Science Robotics, Curly is made up from three major parts: curling AI, which uses strategy planning to adapt to the fast-paced, real-time, changing environment that is seen in curling, and two robots to act as the thrower and the skip. In a human team, the thrower is the person delivering the rock, and the skip is the team captain, making the calls on where the thrower should aim to place their rock. Curly is able to combine those two roles into one.

In order to see what is going on during the game, Curly has two cameras, one extends seven feet up to give an aerial view of the ice, whereas the other camera is located by Curly’s front wheels to help with throwing accuracy. These two cameras allow for Curly to take in information as it happens and adjust the force, direction, and curl of its next throw.

When asked if Curly has any other major parts to help the robot function Lee said, “Curly also had anti-slip software. Without it, the machine was more prone to wheelies.”

The only thing Curly can’t do is sweep, so to make the matches fair, the human teams also didn’t use sweepers. However, in other tests, Curly actually teamed up with humans. The robot still acted as the skip and thrower while the humans played on the team as sweepers. The games were an interesting experiment that showed how capable robots and humans are at working together in challenging environments.

Now to get into a little bit of the “why” behind Curly. The short answer is, to learn more about the way humans and robots interact with each other and how they can collaborate. But also to learn about the differences between robots and humans, or in this case, the lack of difference. In curling, the distance you have to throw a rock is the hardest thing to get right. Through their research, Lee and his co-authors found that even elite curlers could be off their target by an average of 3 to 4 feet. Curly was able to match the same margin of error with its throws.

Due to the time constraints in a curling match, there is no time for relearning, meaning Curly must adapt and adjust its movements to compensate for the ever-changing environment that is unfolding before its eyes… err…cameras. In other words, the gap between complex real-world situations and how capable robots are in navigating those situations is narrowing. This is another reason for Curly’s purpose, according to Lee. The better they understand this type of technology, the more useful it is, and it informs “other complex real-world applications, such as drones.”

So what does the future hold for Curly? Lee said that ongoing research is being done to create sweeper robots to hopefully give Curly some teammates. Who knows maybe there will be a fully robotic curling team in the 2026 Olympics.

Featured Video