Stable Object Reorientation using Contact Plane Registration

TL;DR

This paper introduces a system that predicts stable object orientations using contact plane registration, effectively handling multimodal rotation spaces and operating from noisy, partial observations, with strong simulation and real-world results.

Contribution

It proposes a conditional generative model for contact surface classification to improve stable reorientation predictions, advancing beyond existing methods.

Findings

Outperforms state-of-the-art in simulated stacking tasks

Achieves strong zero-shot sim2real transfer on real objects

Handles noisy, partial pointcloud observations effectively

Abstract

We present a system for accurately predicting stable orientations for diverse rigid objects. We propose to overcome the critical issue of modelling multimodality in the space of rotations by using a conditional generative model to accurately classify contact surfaces. Our system is capable of operating from noisy and partially-observed pointcloud observations captured by real world depth cameras. Our method substantially outperforms the current state-of-the-art systems on a simulated stacking task requiring highly accurate rotations, and demonstrates strong sim2real zero-shot transfer results across a variety of unseen objects on a real world reorientation task. Project website: \url{https://richardrl.github.io/stable-reorientation/}