Can environment feed model data in addition to observations?

Hey!

I would like to know if the obs returned by the env.step is the only input to the model or if there is a possibility that the environment adds additional info with the returned obs?

I am not sure if the Markov state is equal to the observation provided.

Adding part of the code I refer to:

obs = env.reset()
for i in range(10000):
  action, _state = model.predict(obs, deterministic=True)
  obs, reward, done, info = env.step(action)
  if done:
    env.render()
    obs = env.reset()
    break

Thank you for your time!

[araffin]

So I understand that it is safe to refer to obs as the state in MDP?

yes. obs is what is called state in MDP, but in practice, as the env is not always fully observable, observation is a more general term.

[Zymoo]

Appreciate that, thanks!