Could not broadcast input array from shape (12,5) into shape (5,12) when using PPO2 and DummyVecEnv

[ryan-snyder]

I'm playing around with some baseline algorithms with a custom environment that I made.

The environment has an observation space with a shape of (5,12).

For some reason, when trying to use PPO2 and creating a DummyVecEnv for my custom environment, it flips the observation shape from (5,12) to (12,5).

The error is happening in

for k in self.keys:
     if k is None:
         self.buf_obs[k][e] = obs
     else:
         self.buf_obs[k][e] = obs[k]

Is this something I need to fix in my custom environment or an actual bug?