Every four years during the Winter Olympics, one sport never fails to capture the world’s attention—curling. It is bizarre, fun, and actually takes a lot more skill than it appears to. Although robots have shown that they can excel at a wide variety of sports, none had taken to the ice to try their “hand” at curling.
Now, Korean researchers have made a robot team that curls so well it can compete at a national level. Although the team is made up of two robots, there is sadly no sweeper. The Roomba certainly could have had its moment to shine.
Nonetheless, the curling robot team, aptly named Curly, is designed to test “the interaction between an AI system and a highly nonstationary real-world scenario.” This could be very useful off the ice as researchers learn how to integrate that type of machine learning into applications like self-driving cars, package delivery drones, and more.
The curling robot team was designed by researchers from Korea University in Seoul and the Berlin Institute of Technology. Their research was published in the journal Science Robotics.
The Curly team works in tandem to improve its accuracy. One of the bots observes the position of the stones from the scoring end of the curling sheet. The other does the throwing. Throughout the process, the two robots communicate in real-time to determine the ideal angle and speed for each throw.
Interestingly, the artificial intelligence (AI) system powering the robots was trained entirely on computer simulations. That is, there was no actual curling until it was time to test the robots.
That approach can end either very well or very poorly. Fortunately, in this case, it was effective. Curly needs just one throw in real life to gauge the ice conditions for each game. That’s about on par with a human curler if not better.
To put Curly to the test, researchers pitted the robot team against some of Korea’s leading women’s teams and the national wheelchair team. The robots won three out of four rounds. Of course, having sweepers may have changed the final outcome as that is an important part of human curling. Nonetheless, it is an impressive achievement.
As mentioned, the benefits of developing a system like Curly go far beyond the ice. On the surface, it shows that robots can be competitive in yet another sport. Perhaps they’ll get their own Olympic team one day.
However, the fact that Curly is able to make dynamic observations and adjust in real-time is crucial. Until very recently, this wasn’t something that most AI systems could handle.
The technology will need to continue improving as the stakes get higher. Applications like autonomous vehicles and delivery drones have no room for error. They require algorithms that can make split-second decisions with perfect accuracy based on observations of a constantly changing environment.
Yes, it’s a daunting task. Fortunately, researchers continue to chip away at it. Innovations like Curly show that it is possible to train AI models to interact with an unstable environment in a successful way.
With even more training—including real-world experience—more advanced AI systems will be able to perform admirably in these situations.