Researchers at South Korea's Korea University, and Germany's Berlin Institute of Technology and Max Planck Institute for Informatics, developed a deep learning algorithm that enabled a curling robot to beat human players.
The team trained the robot, Curly, to evaluate and adapt to uncontrollable environmental conditions using a deep reinforcement learning system to help it compensate for uncertainties and take corrective actions.
The scientists integrated this system with a previously developed strategy planning model, with the result that Curly outperformed expert human curlers.
Korea University’s Seong-Whan Lee said, "The game of curling can be considered a good testbed for studying the interaction between artificial intelligence systems and the real world.”
View Full Article
Abstracts Copyright © 2020 SmithBucklin, Washington, DC, USA