GAIL returning the same action after training

I am trying my hand at some Imitation Learning algorithms and how to apply them. I am using the Imitation Learning library built on top of Stable Baselines3.

I created a custom environment using Open AI Gym with an action space of type Discrete of x elements and an observation space of type Box with shape (1,) and bounds of low=0 and high=x+1.

I tried training GAIL (using the IL Library mentioned above) by creating a DummyVecEnv and training with several different time_steps (varying from 900 to 150000). Other parameters were left at default. However,the resulting model is always returning the same action which does not even correspond to the actions supplied by the expert.

Has anyone ever had this issue? Is it an issue of incorrect parameters or is something wrong in the make-up of the environment?

Source: Python Questions