Learning Geometrically-Grounded 3D Visual Representations for View-Generalizable Robotic Manipulation

TL;DR

This paper introduces a unified framework for robotic manipulation that learns 3D geometric representations from single-view data, enabling better generalization across viewpoints and improving success rates in diverse tasks.

Contribution

It proposes a novel single-view 3D pretraining method combined with policy distillation, addressing limitations of previous multi-view dependent approaches.

Findings

Outperforms previous methods by 12.7% in success rate on RLBench tasks.

Achieves strong zero-shot view generalization with minimal success rate drops.

Demonstrates effective transfer of 3D understanding to manipulation skills.

Abstract

Real-world robotic manipulation demands visuomotor policies capable of robust spatial scene understanding and strong generalization across diverse camera viewpoints. While recent advances in 3D-aware visual representations have shown promise, they still suffer from several key limitations, including reliance on multi-view observations during inference which is impractical in single-view restricted scenarios, incomplete scene modeling that fails to capture holistic and fine-grained geometric structures essential for precise manipulation, and lack of effective policy training strategies to retain and exploit the acquired 3D knowledge. To address these challenges, we present MethodName, a unified representation-policy learning framework for view-generalizable robotic manipulation. MethodName introduces a single-view 3D pretraining paradigm that leverages point cloud reconstruction and feed-forward gaussian splatting under multi-view supervision to learn holistic geometric representations. During policy learning, MethodName performs multi-step distillation to preserve the pretrained geometric understanding and effectively transfer it to manipulation skills. We conduct experiments on 12 RLBench tasks, where our approach outperforms the previous state-of-the-art method by 12.7% in average success rate. Further evaluation on six representative tasks demonstrates strong zero-shot view generalization, with success rate drops of only 22.0% and 29.7% under moderate and large viewpoint shifts respectively, whereas the state-of-the-art method suffers larger decreases of 41.6% and 51.5%.