: This research paper discusses using Deep RL to tackle the vast state spaces of NES titles, noting that in an average Mario level, a character can occupy thousands of different x-positions across multiple timesteps.
: Projects like ArvindSoma's A3C build upon the foundational paper "Asynchronous Methods for Deep Reinforcement Learning" to train agents specifically for the NES environment. Technical Context of the NES Original Super Mario Bros NES
The "depth" of the NES original is also frequently discussed in the context of its legacy: : This research paper discusses using Deep RL