Object-Centric Representations Improve Policy Generalization in Robot Manipulation

TL;DR

This paper demonstrates that object-centric visual representations significantly enhance the generalization of robotic manipulation policies across diverse visual conditions, outperforming traditional dense and global features.

Contribution

It introduces the use of object-centric representations for robotic manipulation and benchmarks their effectiveness against other visual encodings in various tasks.

Findings

OCR-based policies outperform dense and global representations in generalization.

Object-centric representations maintain robustness under lighting, texture, and distractor variations.

OCR achieves these improvements without task-specific pretraining.

Abstract

Visual representations are central to the learning and generalization capabilities of robotic manipulation policies. While existing methods rely on global or dense features, such representations often entangle task-relevant and irrelevant scene information, limiting robustness under distribution shifts. In this work, we investigate object-centric representations (OCR) as a structured alternative that segments visual input into a finished set of entities, introducing inductive biases that align more naturally with manipulation tasks. We benchmark a range of visual encoders-object-centric, global and dense methods-across a suite of simulated and real-world manipulation tasks ranging from simple to complex, and evaluate their generalization under diverse visual conditions including changes in lighting, texture, and the presence of distractors. Our findings reveal that OCR-based policies outperform dense and global representations in generalization settings, even without task-specific pretraining. These insights suggest that OCR is a promising direction for designing visual systems that generalize effectively in dynamic, real-world robotic environments.