Robots have been proven time and time again to be great single-taskers, but generally awful multi-taskers. In scenarios where one relatively simple job needs to be performed relentlessly, robots usually outperform humans. Humans and other animals, however, usually do much better than robots when it comes to complex tasks with many variables. Sports are a great example of this dichotomy; it’s entirely feasible that you could build a robot that shoots free throws better than the best NBA players, but it would be nearly impossible to build a robot that could actually compete in a real NBA game. But, thanks to the mechanics of the game, Curly the robot is able to perform in ice curling matches as well as people.
The curling sport, if you’re not familiar, is like a big icy version of the table shuffleboard games you’ve probably played in bars once or twice. Table shuffleboard is, of course, a small elevated version of shuffleboard. The goal of the game is to slide a heavy (approximately 40 pound) stone across the ice into a circular target marked on the ice. Two teams attempt to reach the same goal, in alternating throws, which means that a major part of the game involves knocking opposing stones out of the goal while simultaneously trying to position your stone so the same can’t be done to you. The result is a curling sheet (the ice game field) that is dynamic and changes between turns. That would traditionally be tricky for robots to compensate for, but Curly is able to handle it with ease.
Curly was built by a team of roboticists from Seoul’s Korea University and the Berlin Institute of Technology. Technically, Curly is two separate robots. One sits near the goal to observe the position of the stones that have already been thrown, and the second actually does the throwing. Both robots ride on a trio of wheels and have an odd swanlike neck that they can lift into the air in order to get a better view of the curling sheet. There is no sweeper — apparently that position is superfluous. Interestingly, the robots’ artificial intelligence was trained entirely on simulated ice with simulated stones and was only given one test throw to get a sense of the feel of the real world ice. That would be catastrophic in most situations, because the simulations rarely match reality closely enough for this to work. But, in this case, the simulations were accurate enough that the AI was able to win three out of four matches played against professional curling teams.