Decomposing the Generalization Gap in Imitation Learning

for Visual Robotic Manipulation