This project explores the reinforcement learning technique of Q-learning to help an automated cab safely reach its destination in an appropriate amount of time. The environment the cab is in consists of a grid of streets and the cab must learn the best actions to take at a given intersection based on sensory data such as whether a stop light is green or red and what direction other cars are traveling at the intersection.
This project is using Python 2.7 and needs the following libraries:
- pygame
- jupyter notebook
Pygame can be a little tricky to get working in Python 2.7 but is not technically required to run the program. It allows a visualization of the decisions the Smartcab is making as it drives in the environment. To turn off the visualization set the display flag equal to False on line 197 of the smartcab/agent.py file. If you would like to install pygame this link might be helpful in getting it installed. I have found using pip install pygame and brew install sdl sdl_image sdl_mixer sdl_ttf smpeg portmidi (if you use Homebrew) to be successful in installing the package.
To launch the program and see the Smartcab start learning the best actions to take run the following from the root of the project:
python smartcab/agent.py
The analysis of how the Smartcab used Q-learning can be found in a Jupyter notebook and can be run using the following from a terminal:
jupyter notebook smartcab.ipynb
Here is a screenshot of the Smartcab driving through the city in pygame (Smartcab is white vehicle):
