Closed stevenh-tw closed 5 years ago
Hi @karta1297963
Thanks for bringing this up! There is currently an open PR here addressing this: https://github.com/Unity-Technologies/obstacle-tower-env/pull/92. We should hopefully get it in soon, and be able to resolve this issue.
This has been addressed in v2.1 release. Please re-open if you feel necessary.
The code in obstacle_tower_env.py line138-142
Seems like the observation space haven't included
current_floor
variable, which will make the _flatten_obs(a function in subproc_vec_env.py from OpenAI baseline) not wrappingcurrent_floor
during return of the step function._flatten_obs function: in line 198~199 it will only flatten the amount of
len(observation space)