ECE3091/ ENG5105 is a design unit at Monash University, where students are required to design, build and program a mobile robot to solve some task. The robot is built around a differential drive platform, Raspberry Pi and camera module, but students are allowed to use whatever processor or custom accessories they like. 3D printed accessories and custom manipulators are encouraged. This repository has selection of code resources for mobile robot navigation and control, and a crash course in object detection.
- Robot navigation and control
- Getting frames from the Pi Camera
- Simple blob (colour/shape) detection strategy
- Simple CNN (2D regression) detection strategy
There are also some nice c++ programming resources if you prefer to program using this language.
If you would like to configure your raspberry pi to use jupyter follow the steps below.
I'll assume you have an operating system with a terminal, and that can ssh. On windows, check out the windows subsystem for linux, and putty. I will be using Ubuntu, which generally ships with terminal and ssh.
I installed Ubuntu 20.04 server on my raspberry pi, because I wanted a 64 bit operating system for PyTorch (see below). You could follow a similar process for the Raspbian Buster distribution pre-installed on your sd cards, but you won't be able to use PyTorch if you do this.
Let's log into our Ubuntu raspberry pi using ssh
ssh ubuntu@ip_address
Once you're in, it's time to install some useful software. tmux is a terminal multiplexer for running multiple programs in one terminal and remotely executing programs that will continue to run if the network drops. htop is a terminal process viewer that will be useful to monitor processor load/ memory use. vim is a text editor for people who hate emacs. Most of the rest are things we'll use in python.
sudo apt-get update
sudo apt-get upgrade
sudo apt-get install git python3-scipy jupyter libatlas-base-dev avahi-daemon tmux htop vim python3-pip python3-matplotlib libgl-dev net-tools
Since we've installed the avahi-daemon, it's a good idea to change the hostname to something unique, eg. ECE3091GROUPNO. Use the command
hostnamectl set-hostname ECE3091GROUPNO
to do this. You may need to reboot sudo reboot for this to take effect.
Now, assuming you're on the same network, logging in to the pi remotely will be a lot easier going forward, just using:
Ok, lets continue to install some more packages on the pi.
pip3 install --upgrade pip
pip3 install --user jupyter opencv-python torch
Next, we'll clone this repository
git clone https://github.com/mgb45/ECE3091_resources/
Now we'll run jupyter notebook in a headless fashion
jupyter notebook --no-browser --port 8888 --ip 0.0.0.0
You can now access the notebook running on the raspberry pi using any other computer on the network, by simply typing
http://raspberry_pi_ip:8888/
into a browser address bar, where raspberry_pi_ip is the ip address of your raspberry pi. Eg. mine is 192.168.20.14.
You can find out your raspberry pi ip address by typing ifconfig in a raspberry pi terminal, or going to your router settings pages and looking at the addresses of connected devices.
Use ctrl-c in the terminal to kill a running process (eg. if you want to close the notebook). Shutdown your pi cleanly with sudo shutdown -h now
Note that the raspberry pi is really not suited to deep learning (Consider the Nvidia Jetson if you want a better embedded computer for this). You can run models, but I wouldn't try to train bigger models on the pi.