Reinforcement Q-Learning Flagged Maze

Exploring reinforcement learning concepts and solve a maze problem. Discusses model state determination, techniques for state reduction, impact of learning rate (α), and impact of discount factor (γ) on decision-making and convergence. (GitHub)

Model State Determination and Reduction:

1. Number of model states depends on the environment size.
2. Reducing states through equivalence of certain positions.

Concepts and Components:

1. States: Correspond to agent positions in the environment.
2. Actions: Define agent movements – “up”, “down”, “left”, “right”.
3. Rewards: Define penalties and incentives.
4. Goal State: End point to reach, marked as “T”.

Learning Rate (α) Impact:

1. Speed of convergence and oscillation.
2. Exploration vs. exploitation balance.
3. Stabilization and solution accuracy.

Discount Factor (γ) Impact:

1. Long-term vs. short-term rewards.
2. Optimal policy and goal emphasis.
3. Convergence and temporal consistency.