A array tennis-playing robot tin support up a rally against humans, but similar galore amateur players, it struggles erstwhile attempting fancier shots.
Yapeng Gao, Jonas Tebbe and Andreas Zell astatine the University of Tübingen successful Germany began by designing a machine simulation successful which a virtual robot limb equipped with a table tennis racket attempted to return ping pong balls crossed a virtual array tennis table.
The researchers ran this simulation truthful that a machine learning algorithm could larn however the velocity and predisposition of the racket affects the way the ball takes.
Once this algorithm, which learns by proceedings and error, could reliably instrumentality the ball, the researchers acceptable it up to power the question of a existent robot limb positioned adjacent to a existent array (pictured).
The strategy utilized 2 cameras to way the determination of the existent shot each 7 milliseconds, and the algorithm processed the signals and decided wherever to determination the robotic limb to deed and return the ball.
The signals that the algorithm sent allowed the robot limb to accurately play shots to wrong an mean of 24.9 centimetres of the intended location. This accuracy level was somewhat worse than erstwhile the algorithm was moving with a simulation – a communal occurrence, says Tebbe, arsenic machine simulations can’t accurately correspond everything successful existent life.
The robot limb goes up against a quality player Yapeng Gao, Jonas Tebbe, and Andreas Zell
The full process – including grooming successful the virtual simulation and successful the existent world – took conscionable 1.5 hours, demonstrating however rapidly algorithms tin larn to execute successful a caller situation.
However, though the robot performed good against quality players, it was tripped up by accelerated shots – and, surprisingly, by dilatory ones. “If a shot is slow, the robot needs to make much speed,” says Tebbe. Struggling to bash that, the shot often slumped disconnected the racket.
“By grooming the strategy for a relatively abbreviated play of clip the robot is capable to header good with differences successful serve, and susceptible of returning utilizing a random policy,” says Jonathan Aitken astatine the University of Sheffield successful the UK, who wasn’t progressive successful the study.
Aitken was amazed the algorithm flunked returning dilatory shots. He besides finds it absorbing that it sometimes struggled with making shots due to the fact that of the mechanical limitations of the robot system, rather than due to the fact that of shortcomings with the algorithm.
The robot limb has different limitations. For instance, it struggles to play backspin shots, says Zell, due to the fact that the robot limb is incapable to clasp the racket astatine the required space needed to execute specified shots. But contempt these issues, helium believes the robot is simply a bully player.
“It’s not worse than a regular quality player,” helium says. “It’s already connected par with me.”
Reference: arxiv.org/abs/2109.03100
More connected these topics:
English (US) ·