Robots that can hit a golf ball down the fairway aren’t exactly new but creating one that can play the subtle short game is a more difficult task. Golfi, a system that utilizes a neural network to figure out how to line up a putt and how hard to smash the ball to send it into the hole from anywhere on the green, was developed by researchers at Paderborn University in Germany.
The robot uses a Microsoft Kinect 3D camera to capture a photo of the green and then replicates hundreds of random images taken from various perspectives. It considers elements such as the rolling resistance of the grass, the weight of the ball, and the initial velocity. Annika Junker, a PhD student at Paderborn University, told IEEE Research that training Golfi on simulated golf strokes takes five minutes, as opposed to 30-40 hours if the team fed data from real-life shots into the system.
Once Golfi has determined which shot it should take, it rolls over to the ball and makes the putt using a belt-driven gear shaft and a putter attached. However, the robot does not always get the ball into the hole. Junker estimated that the robot hit the target 60-70 percent of the time. That’s still a higher accuracy rate than most amateur golfers, and if it misses, Golfi won’t go off the handle like Happy Gilmore.
Golfi, on the other hand, sometimes drove over the ball, moving it out of place. Because the researchers only tested the robot in the lab, real-world situations such as divot-filled greens or steep slopes may offer challenges for a system that depends on a bird’s-eye vision.
In any event, the researchers had no intention of creating a robot capable of competing with PGA Tour professionals. They hope that the strategies they developed for Golfi may be used to other robotics applications. “You can also extend it to other issues where you have some information about the system and could model sections of it to gain some data,” Niklas Fittkau, another Paderborn University doctorate student and co-lead author of a study on Golfi, told IEEE Research.
