Pre-trained Models
What’s Next
Contact Us
First Use
Wake-word detection
A demo where Pico4ML-BLE provides always-on wake-word detection on whether
someone is saying yes or no, using its onboard microphone and pre-trained speech
detection model.
Magic Wand (Gesture Detection)
A demo where Pico4ML-BLE casts several types of spells in one of the following
three gestures: “Wing”, “Ring” and “Slope”, using its IMU and pre-trained gesture
detection model.
Person Detection
A demo where Pico4ML-BLE predicts the probabilities of the presence of a person
with a Himax HM01B0 camera module.
Wake-word Detection
1. Click on the corresponding uf2. file “micro_speech.uf2”
2. Click on the “Download” button. This file will be downloaded to your computer.
3. Go grab your Raspberry Pi or laptop, then press and hold the BOOTSEL button on
your Pico4ML-BLE while you plug the other end of the micro USB cable into the
board.
4. Release the button after the board is plugged in. A disk volume called RPI-RP2
should pop up on your desktop.
5. Double-click to open it, and then drag and drop the UF2 file into it. The volume
will automatically unmount and the screen should light up.
6. Hold your Pico4ML-BLE closer and say “yes” or “no”. The screen will display the
corresponding word.
Person Detection
Build a Magic Wand with Edge Impulse
Build models on your own
1. Click on the corresponding uf2. file “person_detection_int8.uf2”
2. Repeat the second to fifth steps mentioned in “Wake-word Detection Using” to
light up the screen with the .uf2 file for person detection.
3. Hold your Pico4ML-BLE to capture images. The screen will display the image and
the probabilities of the presence of a person.
The increased Bluetooth help to implement wireless data collection, training, and
model update by designing a web client based on WebBLE. The collected data is
converted to a format that can be read by Edge Impulse through our conversion
script, and then the transfer learning of the model is carried out. Please refer to the
Doc page to build your custom wand project:
https://www.arducam.com/docs/pico/arducam-pico4mltinyml-
devkit/how-to-build-a-magic-wand-with-edge-impulse-on-arducam-pico4ml-bl
e/#26-gesture-recording
If you are developing your own models on Pico4ML-BLE with the Raspberry Pi 4B or
Raspberry Pi 400, you can refer to:
Source file for the 3d-printable enclosure
If you’ve got a 3d printer, you can print your own enclosure for Pico4ML-BLE with
the source file in the link below:
Website: www.arducam.com
Skype: arducam
Doc: arducam.com/docs/pico/
https://github.com/ArduCAM/pico-tflmicro
https://www.arducam.com/downloads/UC-798-Pico4ML-BLE-CASE.zip
1. Click on the corresponding uf2. file “pico4ml_ble_magic_wand.uf2”
2. Repeat the second to fifth steps mentioned in “Wake-word Detection Using” to
Magic Wand (Gesture Detection)
Go to the https://github.com/ArduCAM/pico-tflmicro/tree/main/bin page, then
you will find the .uf2 files for the 3 pre-trained models.
light up the screen with the .uf2 file for magic wand.
3. Wave your Pico4ML-BLE quickly in a W (wing), O (ring), or L (slope) shape. The
screen will display the corresponding mark.
