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