MaxEnt estimates the initial state distribution rather than getting it from the env

[maxmdaniel]

MaxEnt IRL currently estimates the initial state distribution based on the provided expert trajectories:
for traj in self.expert_trajs:
mu[traj['states'][0], 0] += 1
mu[:, 0] = mu[:, 0] / len(self.expert_trajs)

Is there a way to instead get the exact initial state distribution from the gym.Env, similar to the way we extract the exact transition dynamics?