The AI Lab of Softbank Robotics Europe is doing some pretty cool stuff with the Pepper humanoid robot. In September 2016, we saw Pepper quickly master the “Ball in a Cup” game. After 100 attempts, Pepper finally catches the ball in the cup. In fact, it does so 10 straight times on camera. Watch the video below.
The AI Lab has turned its attention to another game: darts. OK, it’s velcro darts, but better to be safe than sorry during any trials. As it did with Ball in a Cup, Pepper uses trial-and-error learning to learn how to play darts. A human initially guides Pepper’s arm to teach it how to throw, but Pepper hits a bulls-eye after just 60 throws.
I have a dartboard in my basement and rarely hit the bulls-eye. Granted Pepper is standing a little close, and probably hasn’t been drinking, but it’s impressive nonetheless.
In the YouTube description, Softbank says the movement “is represented as a so-called dynamic movement primitive and optimized using an evolutionary algorithm” and that this technique was implemented using free software from GitHub.
Steve Carlin, VP & GM, SoftBank Robotics America, is on our “Improving Human-Robot Interaction” panel at the CES Robotics Conference, so we’ll have to pick his brain about Pepper’s new ability. But Softbank’s programmers need to tone down Pepper’s ability to hit the bulls-eye or darts will no longer be a go-to game at your local watering hole.