Tweet
Computer Curling
September 29, 2020
In the past we have had discussions of computer chess, Go, checkers and poker.
https://forum.chesstalk.com/forum/ch...er-plays-poker
and
https://forum.chesstalk.com/forum/ch...light=computer
I have been astounded to read that now computers have beaten man at curling.
From The Times:
https://www.thetimes.co.uk/edition/news/computer-beats-top-team-at-curling-07drtcb02
Extracts from that article (which may be behind a paywall):
Computer beats top team at curling
They have bested humanity at the ancient games of Go and chess. Now a computer has achieved dominance in an icy new domain by thrashing one of the world’s best curling teams.
The computer, called Curly, won three out of four matches against one of South Korea’s top-ranked sides.
Its inventors believe that the feat carries significance far beyond the ice rink. Curling, they argue, unlike more purely cerebral games such as chess, is an excellent test bed for studying the interaction between artificial intelligence systems and the real world.
Curly used an AI technique called deep reinforcement learning, an approach where complex behaviours are learnt from trial and error. Crucially, having been trained on virtual simulations, it was able to thrive in a world full of hidden influences, randomness and changing conditions.
One accommodation was made for the computer: there were no “sweepers”, the players who usually accompany the stone as it skids down the rink, using a broom to sweep the ice in front of it. Sweeping can decrease friction, which makes the stone travel straighter and longer.
The computer used two robots to play. One carried a set of cameras to monitor the house portion of the rink. The second launched Curly’s stones while also applying a spin, which would have caused them to hook left or right.
Its victories differed from other storied conquests made by machines over humans in board games, the researchers say. “From an AI perspective, curling — often referred to as “chess on ice” — is different compared with board games such as chess or Go in multiple ways,” they write in the journal Science Robotics.
The machine used what it had gleaned from millions of simulated plays and merged this with a limited amount of data gathered from playing on a real ice rink, including two practice throws before the start of each game.
Johannes A Stork of Örebro University writes in a commentary in Science Robotics that Curly’s ability to react to past errors and successfully adapt to the current ice sheet conditions suggests that it had bridged what computer scientists call the “sim-to-real gap” — the chasm between the physical realm and digital simulations.
September 29, 2020
In the past we have had discussions of computer chess, Go, checkers and poker.
https://forum.chesstalk.com/forum/ch...er-plays-poker
and
https://forum.chesstalk.com/forum/ch...light=computer
I have been astounded to read that now computers have beaten man at curling.
From The Times:
https://www.thetimes.co.uk/edition/news/computer-beats-top-team-at-curling-07drtcb02
Extracts from that article (which may be behind a paywall):
Computer beats top team at curling
They have bested humanity at the ancient games of Go and chess. Now a computer has achieved dominance in an icy new domain by thrashing one of the world’s best curling teams.
The computer, called Curly, won three out of four matches against one of South Korea’s top-ranked sides.
Its inventors believe that the feat carries significance far beyond the ice rink. Curling, they argue, unlike more purely cerebral games such as chess, is an excellent test bed for studying the interaction between artificial intelligence systems and the real world.
Curly used an AI technique called deep reinforcement learning, an approach where complex behaviours are learnt from trial and error. Crucially, having been trained on virtual simulations, it was able to thrive in a world full of hidden influences, randomness and changing conditions.
One accommodation was made for the computer: there were no “sweepers”, the players who usually accompany the stone as it skids down the rink, using a broom to sweep the ice in front of it. Sweeping can decrease friction, which makes the stone travel straighter and longer.
The computer used two robots to play. One carried a set of cameras to monitor the house portion of the rink. The second launched Curly’s stones while also applying a spin, which would have caused them to hook left or right.
Its victories differed from other storied conquests made by machines over humans in board games, the researchers say. “From an AI perspective, curling — often referred to as “chess on ice” — is different compared with board games such as chess or Go in multiple ways,” they write in the journal Science Robotics.
The machine used what it had gleaned from millions of simulated plays and merged this with a limited amount of data gathered from playing on a real ice rink, including two practice throws before the start of each game.
Johannes A Stork of Örebro University writes in a commentary in Science Robotics that Curly’s ability to react to past errors and successfully adapt to the current ice sheet conditions suggests that it had bridged what computer scientists call the “sim-to-real gap” — the chasm between the physical realm and digital simulations.