TinyML Smart Tennis Racket

Run Machine learning on a micro controller with an accelerometer sensor to classify different moves with the racket - Forehand, Backhand, Serve and Idle.

Data is collected using the tinyml-tennis-collector firmware, that sends data over BLE and there is a Web UI on web-bluetooth-bridge-ui folder that relays the data to Edge Impulse.

Demo video of the collecting data:

• Link on Twitter

The model was trained on Edge Impulse and exported to be used on the tinyml-tennis-classifier firmware, that them shows a different LED color depending on the class - idle (red), forehand(green), backhand(blue)

Demo video of the classification:

• Link on Twitter

️ ⚠️️️️THIS IS A WORK IN PROGRESS ⚠️

TODO

• Collect more data and with different people
• Collect Serve data
• Make data available for others to use

Bom - Bill of Material

• Particle Xenon and/or nRF52840 Dongle
  • I installed the Adafruit nRF52 Bootloader
  • I used the Particle Debugger and openocd to flash the bootloader.
• MPU 6500 Accelerometer Module

Upload firmware with PlatformIO

I recommend installing the Visual Studio Code (VSCode) IDE and the PlatformIO plugin to get started using it. Just follow the step on the link below:

https://platformio.org/platformio-ide

To deploy to the board, just open the tinyml-tennis-classifer or tinyml-tennis-collector folder and you can use the “Build” and “Upload” buttons on PlatformIO Toolbar. All libraries and dependencies will be downloaded.

Web Interface with Bluetooth

You need to generate an API Key/Secret pair to send data to Edge Impulse.

• Run on the command line:

cd web-bluetooth-bridge-ui
npm install
npm start

References

• https://www.tensorflow.org/lite/microcontrollers
• https://blog.arduino.cc/2019/10/15/get-started-with-machine-learning-on-arduino/
• https://medium.com/@devdevcharlie/play-street-fighter-with-body-movements-using-arduino-and-tensorflow-js-6b0e4734e118