ESP32 Axis Prediction Using TFLite

This example shows a working example of the ESP32 using tensorflow lite to predict yaw, pitch and roll movements.

Hardware Used

• FireBeetle 2 ESP32-E
• HMC5883L Compass

Instructions

• Gather your IMU training data and save to a csv file. imu.csv contains data formatted from a previous project.
• I prefer to load my environment using docker: docker run --gpus all -it --rm -v <this folder>:/tmp tensorflow/tensorflow:latest-gpu bash
• Run the training script: python3 train.py
• Convert the model to a byte array with:
  echo "const unsigned char model[] = {" > model.h
cat movement_model.tflite | xxd -i >> model.h
echo "};" >> model.h
• Include the generated model.h in to the sketch folder then compile and upload.

Result

Output confusion matrix showing actual vs predicted test

References

• https://github.com/arduino/ArduinoTensorFlowLiteTutorials/blob/master/GestureToEmoji/arduino_tinyml_workshop.ipynb
• https://docs.arduino.cc/tutorials/nano-33-ble-sense/get-started-with-machine-learning