Decomposing the Generalization Gap in Imitation Learning for Visual Robotic Manipulation

TL;DR

This paper investigates the factors affecting generalization in imitation learning for visual robotic manipulation, proposing a new benchmark and analyzing the relative difficulty of different environmental factors.

Contribution

It introduces a new simulated benchmark with multiple factors of variation and provides an analysis of their impact on generalization in imitation learning.

Findings

Certain environmental factors are more challenging for generalization.

The difficulty order of factors is consistent across simulation and real robots.

A new benchmark facilitates controlled evaluation of generalization.

Abstract

What makes generalization hard for imitation learning in visual robotic manipulation? This question is difficult to approach at face value, but the environment from the perspective of a robot can often be decomposed into enumerable factors of variation, such as the lighting conditions or the placement of the camera. Empirically, generalization to some of these factors have presented a greater obstacle than others, but existing work sheds little light on precisely how much each factor contributes to the generalization gap. Towards an answer to this question, we study imitation learning policies in simulation and on a real robot language-conditioned manipulation task to quantify the difficulty of generalization to different (sets of) factors. We also design a new simulated benchmark of 19 tasks with 11 factors of variation to facilitate more controlled evaluations of generalization. From our study, we determine an ordering of factors based on generalization difficulty, that is consistent across simulation and our real robot setup.