A reinforcement learning agent explores a gridworld maze, learning by trial and error — ε-greedy action selection balances exploration against Q-value exploitation, the Q-table heatmap gradually carves out a policy leading from start to reward.