Shortly after the COVID-19 pandemic began, Samuel Alexander and his housemates purchased a ping pong set and began to play — a lot. Becoming quite good at the game, Alexander realized that his style was not consistent with how more professional table tennis players hit the ball, as he simply taught himself without a coach. Because of this, he was inspired to create a smart paddle that uses an integrated IMU to intelligently classify which moves he makes and correct his form to improve it over time. 

Alexander went with the Nano 33 BLE Sense board due to its ease of use and tight integration with TensorFlow Lite Micro, not to mention the onboard 6DOF accelerometer/ gyroscope module. He began by designing a small cap that fits over the bottom of a paddle’s handle and contains all the electronics and battery circuitry. With the hardware completed, it was time to get started with the software.

The Tiny Motion Trainer by Google Creative Lab was employed to quickly capture data from the Arduino over Bluetooth® and store the samples for each motion. Once all of the movements had been gathered, Alexander trained the model for around 90 epochs and was able to achieve an impressive level of accuracy. His build log and demonstration video below shows how this smart paddle can be used to intelligently classify and coach a novice player into using better form while playing, and it will be fun to see just how good the model can get.