Korean Ministry of Science and ICT announced that Korean joint research team of 8 institutions developed a curling robot with “Artificial Intelligence Curling SW” to create optimal curling stone pitching strategy.
Picture – Left skipper bot and right thrower bot
The curling robot, ‘Curly’ is divided into three main parts. AI program CurlBrain, skipper bot and a thrower bot.
The skipper bot recognizes the game situation through the camera mounted on the head. Then the skipper robot transmits a game image recognized through the camera and A.I. software CurlBrain establishes an optimum pitch strategy based on this information.
The thrower bot located on the opposite side of the field controls the pitching power, pitching direction, and stone rotation required for pitching, and plays the role of pitching the stone to the target point.
‘Curly’ is a robot with software and hardware and it requires an advanced level of technology as the game has to be played in the icy environment where the ice keeps changing real time.
For instance, Google’s ‘Alpha Go’ is software only A.I. but ‘Curly’ is an artificial intelligence SW ‘CurlBrain’ with a hardware ‘ skipper/thrower bot’. It is also necessary to support the hardware technology such as very accurate pitching force and direction control.
The AI curling program is designed for self-learning strategies by applying “deep learning techniques” and “kernel regression techniques” to effectively find the optimal strategy. Based on this, about 160,000 pitching data were used for initial learning, and after that, about 4.5 million pitching data generated by self-reinforcement learning was used.
‘Curly’ has also won “International Digital Curling Competition” held in Japan.
The research team said, “We started to develop an AI program later than Japan but we are proud to prove that our A.I.’s performance is world class”