Would it be possible to have a short example for a non-neural model? Is something equivalent to a character framebuffer exposed by the API? Thanks!
The env returns the observations that are already separated and encoded into a dictionary for convenience, this has nothing to do with neural net. Is there a particular format you need?