Are you interested in self-driving cars? Do you want to learn RL? If so, look no further! In this series of tutorials, we'll teach you how to train an RL model to drive a simulated car in Metadrive, an environment for testing self-driving cars. We assume only knowledge of basic ML, and will teach you any RL concepts. To ensure your understanding, we provide coding exercises for you to fill out, as well as solutions to check your work against.
Here's a list of the topics you should be familiar with, alongside some sources that you can review them with:
- Partial Derivatives and Gradients
- Probability Distributions
- Backprop
- 3Blue1Brown's Backprop Video: https://www.youtube.com/watch?v=tIeHLnjs5U8
- Andrei Karpathy's Micrograd Video: https://www.youtube.com/watch?v=VMj-3S1tku0
- PyTorch
Note: Metadrive only officially supports Windows and Linux systems
The most up-to-date instructions for installing Metadrive can be found on the Metadrive repository
We reccomend directly git cloning the latest version of the repository, as it contains recent updates that allow you to use it with newer libraries:
git clone https://github.com/metadriverse/metadrive.git
cd metadrive
pip install -e .
The tutorials build off of each other, so we reccomend reading them in the order listed: