3D-MVP: 3D Multiview Pretraining for Robotic Manipulation

TL;DR

This paper introduces 3D-MVP, a novel 3D multi-view pretraining method using masked autoencoders and a multi-view transformer, significantly enhancing robotic manipulation performance by enabling better 3D scene understanding.

Contribution

It proposes a new 3D pretraining approach with masked autoencoders and a multi-view transformer, improving robotic manipulation generalization over 2D-only methods.

Findings

Improved performance on virtual robot manipulation tasks

Effective 3D scene understanding through pretraining

Enhanced generalization of vision-based policies

Abstract

Recent works have shown that visual pretraining on egocentric datasets using masked autoencoders (MAE) can improve generalization for downstream robotics tasks. However, these approaches pretrain only on 2D images, while many robotics applications require 3D scene understanding. In this work, we propose 3D-MVP, a novel approach for 3D Multi-View Pretraining using masked autoencoders. We leverage Robotic View Transformer (RVT), which uses a multi-view transformer to understand the 3D scene and predict gripper pose actions. We split RVT's multi-view transformer into visual encoder and action decoder, and pretrain its visual encoder using masked autoencoding on large-scale 3D datasets such as Objaverse. We evaluate 3D-MVP on a suite of virtual robot manipulation tasks and demonstrate improved performance over baselines. Our results suggest that 3D-aware pretraining is a promising approach to improve generalization of vision-based robotic manipulation policies. Project site: https://jasonqsy.github.io/3DMVP