A newly engineered artificial intelligence-powered robot named Curly can improve its next stroke by learning from its mistakes have beaten one of the world's best curling teams.

Curling is a sport that involves continuously changing environments in which its players slide a stone on a sheet of ice towards the center of a circular target. This sport provides the best avenue for an AI-powered robot to be tested.

The Korean Olympic silver-medal curling women's team lost three rounds out of four from the match against Curly. The narrows the gap between computer simulators and the real world, says the researchers from Korea University who developed the robot.

They hope that the learning techniques Curly has learned can be applied to robots who work in the real world setting and adapt to changing conditions.

Curly Beats Korea's Curling Team

One of the oldest sports from the 16th century in Scotland, Curling is one of the games that were usually played during winter on frozen lochs and ponds.

Brain engineer Dr. Dong-Ok Won built an artificial intelligence (AI) system into Curly, allowing it to bring curling firmly into the 21st century. The researchers used deep reinforcement learning (DRL) techniques, which made Curly able to learn from trial and error to compensate for the uncertainties present in the game.

AI-powered robots have already beat the Chess and Go world champions before, which brings the robot to the world of physical sports. According to Professor Won, the game of curling is a good testbed for the robot because it is like a combination of chess and bowling played two teams alternately on the ice sheet, requiring high-level strategic thinking and performance.

Professor Won added that the environment in curling is inconsistent as it changes now and then, and every throw of the stone has an impact on the outcome of the match. There is no time for learning during the game because of its timing rules.

The AI-powered robot works by capturing information on the changing environment on the ice sheet through the mistakes the robot has committed from its previous throws. To detect the changes on the ice and win the match, Curly only needed to complete a couple of calibration movements.

Read Also: AI Wrote An Op-Ed Convincing Humans That Robots Will Not Replace Humans

Curly Closes the Sim-to-Real Gap

Published in the journal Science Robotics, the study describes how Curly closes the sim-to-real gap. A robot that performs simulations has failed in the unpredictable conditions of the real world. Indeed, applying AI technology to the real world is very challenging, and operating outside the laboratory setting could affect the performance of the AI system.

Moreover, the real world is full of uncertainties that are too complex to model with enough accuracy. It is necessary to incorporate these uncertainties in the modeling efforts and to measure and estimate the changes in real-world settings.

The researchers said the technique is transferrable to other complex applications like having robots solve tasks o a level at par with humans.

<iframe width="560" height="315" src="https://www.youtube.com/embed/z_-oyjQE2Jk" frameborder="0" allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture" allowfullscreen></iframe> 

Read More: New System Teaches Offshore Robots to Answer Questions

Check out more news and information on Robots and AI in Science Times.