Movuino Gesture Recognition (TinyML)

Table of Contents

• Instatllation Environments
• Instructions on magic wand example on Arduino Nano 33
• Instructions on deploying models on Movuino
• Instructions on running Dashboard
• Flowcharts
• Movuino Troubleshooting

Instatllation Environments

• Clone this repository with the following command

git clone https://github.com/soledadli/Movuino_gesture_recognition.git

Requirements includes:

Python == 3.8.3
Arduino == 1.8.12
Arduino_LSM9DS1 == 1.0.0
Arduino_TensorFlowLite == 1.14.0 - Alpha 

Instructions on magic wand example on Arduino Nano 33

Instructions on deploying models on Movuino

• Run the extract_imu_data.py file from the terminal

python3 extract_imu_data.py

• Upload the compiled data to Google Colab

• Open the Firmware_Peripheral_Classifier.ino

Replace your model.h file to the previous model.h

• Change your model name to g_model as the picture shows

• Upload the Arduino sketch with the new model to Movuino
• Upload the multiCentral_Classifier.ino to Adafruit Feather
• Open the Serial Monitor of the Adafruit Feather in Arduino IDE
• Turn on the Movuino watch
• Here are the predictions of your motion 🎈

Flowcharts

Use Movuino to Collect Gesture Data

Predict Motions & Plot out Real-time Predictions

Movuino Troubleshooting

Q1: The Arduino IDE cannot recognize the port for Movuino

• Check whether the board is correct. The correct version is Movuino Open Health Band v0.2.7.4
• Check whether the connection is good.
  • Check whether the red LED, charging indicator light, is on.
  • Check whether the board is on. If not, press the black switch for 2 seconds.
• Double-click the reset button on the Movuino board. (Red LED light will blink when you reset the board)
• Use the Debugging Board for testing codes

Q2: Having trouble compiling codes on Movuino (There is a higher chance that when compiling codes on Movuino, Arduino IDE may crush more often when there are bugs on the code or the model size is too big.)

• Check whether you use the right version of the TensorFlowLite Library.
  • Open Manage Libraries -> Search for Arduino_TensorFlowLite -> Check the version -> The one I tested and worked is 1.14.0 - Alpha
  • Do not choose versions with precompiled labels. They may not work.
• Verify the codes before you uploading them to Movuino
• Try uploading the imu data parts and test whether they work. If so, upload the tensorflow parts and the model part then.
• Unplug the board, replug it, and reupload the codes
• Bootloading your Movuino Board to clear out memories
  • Opening the Arduino IDE
  • For board, choose Adafruit Feather nRF52840 Feather
  • The bootloader option will be 0.3.3 SoftDevice s140 6.1.1
  • For programmer, choose Bootloader DFU for Bluefruit nRF52
  • Click Burn Bootloader
  • Wait the bootloading time, and reupdate your sketch after the process is finished.

Q3: Disk get ejected when after uploading the sketch

• Open the Serial Monitor, the sketch may be already successfully uploaded.
• Double-click the reset button, and replug it back to the computer, and run the sketch.

Q4: After uploading the sketch, there is no Serial.println information on Serial Monitor

• Parts of your code may crush. You may want to debug your code paragraphs by paragraphs
• If the Serial Monitor is still blank after after codes paragraphs by paragraphs.
  • Reset the board by double-clicking the button, until you see a message that similar to "board is ejected"
  • Unplug it and replug it
  • Test your board with a simple Serial.println("Test") not upload any codes related with model from void setup() and TensorFlow Lite Pipeline from void loop()
  • If the serial monitor prints out the information Test, then upload the codes with TensorFlow Lite and the model

Q5: The Arduio IDE gets stuck or it takes a while to load the information.

• Unplug your Movuino and replug it.
• Upload your sketch again. If your Arduino IDE still gets stuck, test your board with a simple Serial.println("Test") not upload any codes related with model from void setup() and TensorFlow Lite Pipeline from void loop()
  • If the serial monitor prints out the information Test, then upload the codes with TensorFlow Lite and the model