Published May 9, 2022

Golf Ace

Learn to putt like your favorite pro golfer with machine learning.

IntermediateFull instructions provided8 hours1,857

Things used in this project

Hardware components

Adafruit PowerBoost 1000C

Hand tools and fabrication machines

Arduino nano board 33 IOT

Story

How It Works

When you are ready to putt, press the button on the handle of the putter. An accelerometer onboard the Arduino Nano 33 IoT will begin recording measurements during the swing. After the putt, an RGB LED on the head of the putter will give feedback -- a green light indicates that your form closely resembled that of the pro golfer that the putter was trained to recognize; red means that your technique was off.

In order for the putter to recognize proper putting technique, a K-means clustering algorithm (for anomaly detection) was trained on a set of accelerometer data collected from an instrumented putter. Ideally, this data would be collected from a professional golfer over time, such that there is enough data to perfectly characterize the putting technique that makes that golfer great. By loading a model trained with that data onto a Golf Ace putter, it would allow enthusiasts to buy a club that teaches them to putt like their favorite pro.

But, since there are no pro golfers that owe me a favor, for this prototype, I stood in as the "pro" that supplied the model training data. I am, after all, a winner of the GrandMaster Tournament of Champions Invitational minigolf tournament (which may or may not have been invented by myself and a few friends), so that should suffice for the occasion. To simplify building/training/deploying the anomaly detection model, I used Edge Impulse (public project here).

The Arduino sketch, which handles accelerometer data collection, machine learning model inferences, the RGB LED, and button presses can be found here.

One limitation of this device is that it is only capable of telling you if you have your technique right or not. It cannot tell you what you need to do differently. I envision a putter like this coming packaged with a video and/or written instructions from the golfer it was trained by to explain their technique. Then one could use that as the target they are working towards, with the putter offering feedback as to whether or not the target is being met.