Modify code to experiment on Mujoco env (eg. Walker) in new version of Gym

[VuongLong]

Thank you so much for provide a Benchmark for Model-Based Reinforcement Learning.

I want to modify your code to new version of Gym, I have some question about your code:
in file mbbl/mbbl/env/gym_env/walker.py, in function reset(), after line 75 you do some work:

1. Get the Observation, store in self._old_ob
2. Call self._env.reset()
3. Set the self._old_ob back

As you comment in line 75: # the following is a hack, there is some precision issue in mujoco_py. I'm not familiar with old versions of Gym but in new version Gym (eg. 0.12.1), self._env.reset() return new observation.

I wonder if you want to recover the previous state after calling self._env.reset() or just there is an issue in old version of Gym.

For the case of there is an issue of old version:
Can I just get: self._old_ob = self._env.reset() then return it? (For current work, I skip the groundtruth-dynamic)

Thank you so much for your support!