Easy example Maze and learned Policy
This repository creates a random maze and attempts to solve it using Deep Q-learning. The agent may not hold any memory other than the experience replay buffer, may not import anything from the environment, and may not use any heuristics to solve the maze - this is a pure Deep Q-learning solution.
Creates a random maze to be solved, draws it and the policy, sets the rules
Trains the DQN for 10 minutes and then tests it
Chooses actions, contains DQN and experience replay buffer