noahwo / TinyML-MCU

MCU components for Tiny Machine Learning as a Service

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

TinyML-MCU

Code for TinyMLaaS MCU devices.

Bridge

Dependencies

The bridge has a dependency on usbutils. On Debian-based systems, it can be installed with

apt install usbutils

Running

Locally

Install the python dependecies

pip install -r requirements.txt

There are two different ways to run the bridge locally.

First, to run a development server, run

flask --app main.py --debug run

For production, this is not ideal. To use waitress as production server

waitress-serve main:app

Docker

The docker version of the bridge requires nestybox/sysbox.

The provided docker-compose file looks like this

version: '3'

services:
  bridge:
    build:
      context: .
    # If you have devices connected to the relay, add them here
    # devices:
      # - "/dev/ttyACM0:/dev/ttyACM0"
    volumes:
      - "/dev/bus/usb:/dev/bus/usb"
      - "/dev/serial:/dev/serial"
    runtime: sysbox-runc
    ports:
      - 5000:8080

Any microcontrollers you want to control with the bridge need to be added to the devices at this point. They also need to be connected to the computer.

If you want to add more microcontrollers to the bridge later, you can add them manually by adding them to the docker-compose file and restarting the container.

There is also a provided script to check for added Arduinos. The script docker-container-restarted automatically detects, when new arduinos are added and restart the container with the new devices. It can be run by making it executable and running it

chmod u+x docker-container-restarter.py
./docker-container-restarter.py

There might be problems with serial port permissions. As a solution, give permissions to the port for all users with

chmod 777 /path/to/port

Note that this will most likely require root priviledges.

Serving the application to internet

Webserver

If you already have a webserver setup, such as Apache2 or Nginx, with which you can serve the application to the internet, add a new proxy for the bridge according to the style of the webserver.

Port forwarding

Another way is to add a new port forwarding setting to the host machine to port 5000 in the router.

Ngrok

However, these might not be usable by everyone. With Ngrok, you can serve the bridge to the internet without having access to the local router.

  1. Install ngrok
  2. Serve the application with
ngrok http 5000

Microcontrollers

There are seperate instructions for microcontrollers.

Arduino

About

MCU components for Tiny Machine Learning as a Service


Languages

Language:C 69.4%Language:C++ 27.1%Language:Python 1.9%Language:CMake 1.3%Language:Dockerfile 0.4%