• The agent starts in the top left corner.
• Each step gets an negative reward of -1
• If the agent steps on one of the magenta cliff tiles on the top, there is a negative reward of -100 for falling off the cliff and the agent has to start again from the top left corner.
• The goal is to reach the top right corner. When reaching the top right corner, everything starts all over again.

Sometimes it strikes me how dark, twisted and cynical this is, even though it’s just a toy example. Basically it states

• -1 for every step you make means nothing less than living is suffering
• Even if you reach your goal you don’t get any reward, just you’re suffering ends till it starts all over again. You can’t escape.
• Jumping off the cliff won’t help you. You just have to start over again. There’s no escape.

There’s also a much more positive life lesson one could learn from $\epsilon$-greedy exploration policies. I’ll definitely have to write about it soon, to balance out this post.