Sometimes the world of reinforcement learning is so dark and cynical.
Gridworlds like the one I plottet below are utilized to experiment and explore different core characteristics of learning agents.
Unfortunately they aren’t as popular as they have been as a research tool (this paper shows that gridworlds still are very relevant) they are still omnipresent in teaching because they have so small discrete state spaces and actual humans can actually understand the state representations (mostly just xy coordinates on a grid).
The example below has the following rules(taken from the excellent Sutton and Barto’s amazing book on RL).
