RL_Car_Racing

An introduction to common reinforcement methods (RL) leveraging Gymnasium (formerly OpenAI Gym), as a featured part of Carnegie Mellon University's MRSD Summer Software Bootcamp.

Learning Objectives:

1. Introduction to RL & Algorithsm
   1.1 Deep Q-Learning (DQN)
   1.2 Double DQN (DDQN)
   1.3 Proximal Policy Optimization (PPO) - COMING SOON
2. Getting started with and understanding Gymnasium (previously OpenAI Gym)
3. Getting exposure to some MLOps tools 3.1 Weights and Biases (WandB)
   3.2 Data Version Control (DVC) - COMING SOON

Assignment Output:

When completed, you should obtain an RL agent that learned to drive completely on its own accord. The agent below was trained using a DDQN model with nothing but images of the environment and a set of actions to choose from.

DDQN agent trained from scratch on the Gymnasium Car Racing Environment

Getting Started

Head over to docs to get started!

Note that since this is an assigment, there are solutions which can be found at the solutions branch. However, the solution to this assignment is not unique, there are likely countless ways to obtain a driving agent. I would urge you to attempt as much of this assignment on your own as possible and only resort to the solutions after countless attempts.

If you are looking for an even harder approach of this problem, head to the hard branch where I remove informative docstrings and typehinting which requires you to really know your stuff.