Could we make an RLlib environment which is easily customized via the config (rather than writing custom code every time)?
A couple ideas:
- Should we have an environment variable for loading additional state/action/reward builders?
- Should we be able to load/scale/combine reward builders?