By Kim Young Deok and Lee Jihae
Video = Korea.net
An artificial intelligence (AI) robot developed by Korean researchers has won three out of four games against a curling team comprising top female players and members of the national wheelchair team.
A research team led by professor Lee Seong-whan from Korea University’s Department of Artificial Intelligence told the global journal Science Robotics on Sept. 24 that the robot, named Curly, achieved this result through “deep reinforcement learning.”
Deep reinforcement enables the AI robot to adapt in real time without relearning to its exposed environment, which has a variety of variables and high uncertainty. Through real-time adaptation, the robot stably fulfills its task.
The report said Curly set over three to four days pitching strategies and internalized the strength and direction for pitching and methods to control the swirls of the stone. As a result, it came to possess skills close to that of a curling professional.
Curling is an official event of the Winter Olympics. The ice field can be irregular depending on the venue’s temperature and humidity and ice maintenance. Thus a player needs to consider these elements.
Professor Lee said his team developed the core technology to develop an AI curling robot with the performance level of a skilled athlete. He said the achievement is notable as AI technology based on machine learning can lead to performance at the level of a skilled competitor.
The professor said, “We note that the insights obtained within our framework on how to alleviate challenges such as strong temporal variability, uncertainties, and continuousness are readily transferable for contributing to other real-world applications of comparable complexity in robotics and beyond.”
He first developed Curly back in March 2018 at the time of the PyeongChang Winter Olympics. At the time, the robot lost, 3-0, in its two-end game against a high-school curling team.