youngsoul / eightbitdriving

Project for Machine Learning

8-Bit Autonmous Machine Learning Driving

Your assignment, if you should so choose to accept, is to collect training data from an 8-bit car driving simulator and train a machine learning model to drive the 8-bit car autonously.

Two scripts are provided for you:

• training.py

This script will run the test track for you to use the left/right arrow keys to simulate driving and collect training data.

• drive_by_model.py

This script will look for a file called: best_driving_model.sav which assumes you have saved a Scikit-Learn model training on the driving data.

The drive_by_model.py will use your model to drive the 8-bit car on a random road.

---

Setup your environment

This project uses a terminal window package called, curses which likely does not work on Windows.

Recommended Setup - Docker

if you have Docker installed on your machine the recommended setup is the following:

Build a local container that can execute the 8-Bit project

docker build --tag 8bitdrive .

docker run -it --rm 8bitdrive /bin/bash

docker run -it -v /Users/patryan/Development/mygithub/eightbitdriving:/code --rm 8bitdrive /bin/bash

After the run command, you should be in a linux terminal window with a prompt like:

root@f78503698765:/code#

Note that the Docker file does expose the current directory to the /code directory in the container so if you choose to use an IDE to edit/create the files locally they will be visible to the container.

MacOS

The curses library is available in MacOS. For Mac setups, you can use the following if you do not want to use Docker.

python3 -m venv venv

source venv/bin/activate

pip install -r requirements.txt

The rest of this setup assumes Docker.

Training Data

The training data is a 150 samples of what the road looked like and the driving instruction. For example the data for each Left, Right, and Straight would look like the following:

Left

Straight

Right

Collect the training data

When you start the training, the 'car' is at the bottom of the simulated road and should be in two different colors:

You use your left and right arrow keys to move the car left/right. No arrow press, means straight.

You have to allow for a bit of creative visualization in that the car never moves from the center of the area, it is the road the moves. (Its 8-bit, its simulated)

Assuming you have an open terminal in the running docker container ( see above ) and you are in the /code directory:

This will create a file called: training.csv

There are no column headers and the first column is the turning direction:

0 - Left

1 - Straight

2 - Right

Your Challenge

Create a machine learning model that uses the training.csv data to predict based on the road configuration.

Save your model to a file called, best_driving_model.sav

Test Your Model

To test your model use drive_by_model.py. This script assumes you have saved your models as described above.

Inside the docker container:

python3 drive_by_model.py

Select random and see how your model is behaving.

NOTE There is a file called, model_training.py which shows just one way to do this. If you get stuck feel free to use this to help guide your work but try to do it on your own.