Q-Learning Grid World Learning

Agent learns optimal path via Q-learning. Watch exploration become exploitation as the Q-table fills with experience.

Agent Goal Wall High Q-value

Parameters

Learning rate (alpha) 0.1 Discount factor (gamma) 0.95 Epsilon (exploration) 0.3 Speed (steps/frame) 5

Training

Display

Show Q-value heatmap Show policy arrows

Episode: 0
Steps this ep: 0
Total reward: 0
Best ep reward: -
Avg last 10: -