First, how the robot won: by watching. There were three strategies that could produce a 100% win record, but they all basically boil down to using a high-speed camera and human-beating electronic reflexes to identify the oncoming shape of the opponent’s hand and play the corresponding move to beat it. Just the angle of the wrist or early movement of the fingers is enough to give away what move the human is headed for.The newest version of a robot from Japanese researchers can not only challenge the best human players in a game of Rock Paper Scissors, but it can beat them — 100% of the time. In reality, the robot uses a sophisticated form a cheating which both breaks the game itself (the robot didn’t “win” by the actual rules of the game) and shows the amazing potential of the human-machine interfaces of tomorrow.
The worst strategy produced a winning move about 0.02 seconds after the human’s move, the fastest almost instantaneously, but in all cases the robot is technically waiting to see the opponent’s move before deciding on its own — that’s cheating, bro! This is why the robot famously “Never loses. Ever.” — because it’s not really playing.
But the approach has implications for more than just children’s games. Basic interpersonal interaction, while trivially easy for a human being, is notoriously difficult for a robot; to complete a handshake the robot must simultaneously move its own arm, observe the human’s arm, and modify its movements in response to those observations. While this study’s sort of rock-means-paper response is of course very simple, it is a proof of concept for real-time robot response to human movement. This also applies to movement assistance, such as military and industrial exoskeletons, which look to identify and help with the user’s movements.
Rock Paper Scissors is sometimes known as “Janken,” which gave the robot its super-creative name: “The Janken (Rock Paper Scissors) Robot.”
Rock Paper Scissors has actually fascinated roboticists for some time; last year, researchers figured out that there are broadly reliable patterns in human RPS behavior, giving their robot an advantage through sheer insight into how humans play the game. In that case, a robot was able to learn and exploit the statistical oddities of how humans non-randomly select their next play, gaining an advantage in the aggregate. This newer cheating approach shows how robotic abilities offer avenues around traditional problems, often with far more impressive results.
It is telling, though, that to increase the response time a less and less real-world scenario was required. With the use of specialized backgrounds and lighting conditions, everything has to be just right for the robot to be able to correctly interpret the early movements of a human hand and arm. It shows how very hard it is going to be, to bring robots into the everyday. The ability to beat people at RPS basically boils down to the ability to see and react to movement fast enough to fool human perception — that’s a very important threshold.
This isn’t really there, just yet. As mentioned, the conditions have to be just right for its latency to get this low, and the rig is quite expensive. But with high-speed cameras and an ever-improving ability to let robots actually interpret what they see, we may not be far from a future with real robot integration into everyday life.
No comments:
Post a Comment