@user76284 Can you elaborate? What do you need the benchmarks for? What do you mean by "MDP as benchmark"? Furthermore, I think you should ask a related (but clearer) question on the main website!
@nbro I was just wondering whether there are particular examples of discrete MDPs that can be used as benchmarks for measuring the performance of various reinforcement learning algorithms (again in a discrete setting). For example, chess, go, othello, etc. are used as benchmarks to measure the performance of various algorithms for solving perfect information games. The same goes for poker for imperfect information games.
I think gridworld-like environments are often used as example MDPs, for instance.
I'm asking this as an alternative to simply sampling a random MDP from the space of all possible MDPs (with $n$ states and $k$ actions).
