Learning Action-based Representations Using Invariance

Max Rudolph, Caleb Chuck, Kevin Black +3

Robust reinforcement learning agents using high-dimensional observations must be able to identify relevant state features amidst many exogeneous distractors. A representation that…