Normalize observation space

[pkel]

The documentation of our RL training algorithms encourages us to normalize the observation spaces. This PR implements this recommendation.

https://stable-baselines3.readthedocs.io/en/master/guide/rl_tips.html