We learn by doing and some robots take the same approach. Researchers at the Technical University of Darmstadt have developed algorithms that prove that robot arms just need practice to learn complex activities. Practice makes perfect.
At first there is some hand-holding with a human to understand the basics of a ping-pong swing, then a TUD robot can gradually abstract those motions and return the ball in situations that go beyond the initial example. Turns out that the technique is effective. The test arm took just an hour of practice to successfully bounce back 88 percent of shots and compete with a human.
That’s pretty good for a new player. This science could lead to more robots that can learn quickly and perform well.