Train a Q-learning agent to find the reward, and watch its Q-table fill in.
Start in one corner, reward in the opposite one, nothing in between: any path of 8 moves is optimal.
Episode 1, move 0.
Press Step once to see a single Q update with its numbers.