This project consists of a custom maze environment made using the base gym environment and ran using a coupled GUI made using pygame. The purpose is to train and test reinforcement learning agents on this environment.
The maze is generated procedurally and the location of the objectives and agent are also randomized.
The environment can be derived in order to be further customized. This feature is useful to test custom observation spaces of the environment.
The project contains APIs that makes agent creation and training easy, as the environment is gym based. Implementations of such agents are SARSA and Q-Learning. The performance was measured for each agent, an example of SARSA would be in the following image:
- Clone the repo
- Create Environment via Conda
conda env create -f environment.yml
conda activate RL-Labyrinth