pico-shape-detection

Running a shape detection machine learning model using accelerometer data on the Raspberry Pi Pico W.

---

Checklist of implementation

• Read accelerometer data
• Capture button presses
• Store accelerometer data on button press
• Make tflite-micro library interface
• Feed captured data into the model
• Display classification on LEDs
• Error Handling

---

Running the project

Prequisites:

• The pico-sdk library
• CMake & Make
• minicom
• The tflite-micro library for the Raspberry Pi Pico

For the pico-sdk, follow the guide in Raspberry Pi Pico C/C++ SDK for Windows and Debian-based¹ systems. Since I use Arch, I followed this guide to install the sdk: Arch Linux guide

ensure that the PICO_SDK_PATH refers to the actual path of pico-sdk

echo $PICO_SDK_PATH

The pin numbers have been chosen according to the following wiring:

These pin numbers can be changed as per your layout:

pico-shape-detection/src/main.cpp

Lines 16 to 27 in c49df2e

	enum Pin {
	// buttons
	b_send = 13u,
	b_record,
	// I2C pins
	mpu_sda = 16u,
	mpu_scl,
	// LEDs
	};

In the repository directory, create a build directory, cd into it, run cmake and then make

mkdir build && cd build
cmake ../
make -j4

Before uploading the program to your Raspberry Pi Pico W, check for existing ACM terminals connected to the PC

ls /dev | grep ttyACM

Connect the cable connected to the Raspberry Pi Pico W while pressing the BOOTSEL button and release once connected

Mount the RPI-RP2 and copy the compiled .uf2 file from the build directory to the Raspberry Pi Pico W

udisksctl mount -b /dev/disk/by-label/RPI-RP2
cp pico_acc.uf2 /run/media/<USER>/RPI-RP2/

Reading output from serial port

Find the newly connected serial terminal

ls /dev | grep ttyACM

using minicom, connect to the serial terminal

sudo minicom -b 115200 -o -D /dev/ttyACMn