Episode
0
of 30
Total Reward
0
cumulative
Epsilon (ε)
0.30
exploration rate
Step
0
in episode
Q-Learning (Off-Policy)
Q-Learning updates: Q(s,a) ← Q(s,a) + α[r + γ·max Q(s',a') − Q(s,a)]. It learns the optimal policy regardless of which actions are taken, using the greedy max Q of the next state.
Colors in cells: blue triangles = low Q-value, red = high Q-value. Arrow = greedy policy direction.