An artificial intelligence (AI) robot has beaten the Korean female curling team, who won the silver medal in the 2018 Winter Olympics.
The AI robot named 'Curly' won three out of four official matches against the Korean team, although the machine could only deliver the stone, but does not sweep. The robot was developed by researchers from the Korea University who said this progress reduces the gap between the real world and computer simulators.
Curling is one of the world's oldest sports, which originated in Scotland in the 16th century. Games were usually played on frozen ponds and lochs during winter. Since the sport comprises uncontrollable and constantly changing environmental conditions, it sets the perfect 'test bed' for an AI-driven robot.
Developers had multiple trials using used the deep reinforcement learning (DRL) techniques, which allowed Curly to follow through uncertainties in the game. They hope these techniques could be applied to other robots, so they can also adapt to changing conditions in the real world.
Curling as the perfect testbed
Brain engineer Dr. Dong-Ok Won incorporated an artificial intelligence system into Curly's brain to advance curling into the 21st century. Prof. Won described curling as a combination of chess and bowling. "Two teams play alternately on the ice sheet, requiring a high level of strategic thinking and performance," the professor said.
"The game of curling can be considered a good testbed for studying the interaction between AI and the real world," said Won adding that the environmental characteristics vary each time while each throw creates an impact on the match's results.
The AI captures information about changing conditions on the ice sheet based on the robot's previous mistakes. The researchers only made a couple of calibration moves so that Curly could detect changes in the ice. Meanwhile, curling does not offer time for relearning because of the game's timing rules.
Curling had become popular in South Korea, where its national curling women's team, the "Garlic Girls," won the silver medal when the country hosted the 2018 Winter Olympics. However, the three-wheeled robot has beaten the Olympic -winning team. In the past, other AI machines have also beaten world champions at games like Chess and Go.
"Here, we report a curling robot that can achieve human-level performance in the game of curling using an adaptive DRL framework," the researchers wrote in their study that was published in Science Robotics journal.
The researchers also experienced the "sim-to-real gap" phenomenon. Robots perform well in simulations but fail in the unpredictable and difficult-to-model conditions of the real world.
"Applying AI technologies to the real world is a challenging problem," Won explained, adding that the actual world offers various uncertainties that "may be too complex and ill-defined to be modeled with sufficient accuracy."
The researchers also noted that operating beyond the laboratory would subject the system to unknown factors that can have a huge impact on the AI performance.
Meanwhile, the researchers boast that the method is transferable to a wide range of complex applications, including having robots solve human-level responsibilities.
This is owned by Tech Times
Written by CJ Robles
