tgangwani / bmil Goto Github PK

I think I can't start learning because there is no expert data.
However, I do not know how to create expert data.
Please teach me how to create expert data.

Hi,

Could I ask some questions about the generation of the experts' demonstrations?

1. Are those experts' demonstrations trained directly on the POMDP environment, i.e. do experts have the access to all features?
2. Or the experts are trained in the normal traditional MuJoCo environments? And the states are masked according to POMDP when necessary?

By the way, would it be possible to release some naive experts' example files?

I am having trouble getting the repo to run. The problem is c = env.get_curr_ac() if hasattr(env, 'get_curr_ac') else None (in subproc_vec_env.py) will return None for most test environments. Do you use a custom Gym environment (implement a get_curr_ac method)?