Nano 33 BLE Sense · Sensor Trainer

Collect sensor data, train, test, then bake the model into the Nano's firmware. One board, no SD card. Use Chrome or Edge.
Nano: not connected

0Connect

not connected
Both do exactly the same things. Bluetooth is wireless, USB is the most reliable (and the Arduino Serial Monitor must be closed).

1Live sensors — all 17 values the model uses

2Classes and samples

Name your classes (max 7, 12 characters). Click Record while doing that thing with the Nano. Each sample is a 1 second window of all sensors. Aim for 30+ samples per class, and vary how you do the action. The dot is the colour the Nano's RGB LED will show for that class.
Windows per click: Start delay:
Connect a Nano, then record.

Last captured window: all 17 senses over 1 s (the number is the window average). Click a chart to set how much that sense counts.

No sense selected.
Effect 0 = the model ignores that sense, 1 = full strength. Change effects, then train again. They are applied in training, in the browser test and on the Nano, and baked into the header.

3Train

Epochs: Learning rate:
Not trained yet.
Orange = loss, green = accuracy on the training samples.

4Test

"Test in browser" runs the model here on live windows from the Nano. "Send model to Nano" loads it into the Nano's RAM (lost at reboot), then "Test on Nano" shows what the Nano itself decides.
Nano model: unknown (press Check Nano)

5Bake the model into the Nano

Train a model first.
  1. Save myNanoModel.h in the sketch folder, next to nano33-ble-trainer.ino.
  2. In the Arduino IDE, open the sketch and press Upload. The file is detected automatically.
  3. Power the Nano from any USB supply, no computer needed. After a few seconds the RGB LED shows the class. Press Check Nano above to confirm it holds the baked model.
Log

By Jeremy Ellis · webmcu-ai · MIT license · use at your own risk