Action Invariant Dynamics Learning

This was the final version of our first attempt at creating an action invariant representation in the hopes that it could help improve sample complexity/generalization in RL

Current Version

V1.1

fix bug where I didn't update policies.py