CaltechAUTHORS
  A Caltech Library Service

Fast Uncertainty Quantification for Deep Object Pose Estimation

Shi, Guanya and Zhu, Yifeng and Tremblay, Jonathan and Birchfield, Stan and Ramos, Fabio and Anandkumar, Animashree and Zhu, Yuke (2020) Fast Uncertainty Quantification for Deep Object Pose Estimation. . (Unpublished) https://resolver.caltech.edu/CaltechAUTHORS:20210225-132731801

[img] PDF - Submitted Version
See Usage Policy.

3560Kb

Use this Persistent URL to link to this item: https://resolver.caltech.edu/CaltechAUTHORS:20210225-132731801

Abstract

Deep learning-based object pose estimators are often unreliable and overconfident especially when the input image is outside the training domain, for instance, with sim2real transfer. Efficient and robust uncertainty quantification (UQ) in pose estimators is critically needed in many robotic tasks. In this work, we propose a simple, efficient, and plug-and-play UQ method for 6-DoF object pose estimation. We ensemble 2-3 pre-trained models with different neural network architectures and/or training data sources, and compute their average pairwise disagreement against one another to obtain the uncertainty quantification. We propose four disagreement metrics, including a learned metric, and show that the average distance (ADD) is the best learning-free metric and it is only slightly worse than the learned metric, which requires labeled target data. Our method has several advantages compared to the prior art: 1) our method does not require any modification of the training process or the model inputs; and 2) it needs only one forward pass for each model. We evaluate the proposed UQ method on three tasks where our uncertainty quantification yields much stronger correlations with pose estimation errors than the baselines. Moreover, in a real robot grasping task, our method increases the grasping success rate from 35% to 90%.


Item Type:Report or Paper (Discussion Paper)
Related URLs:
URLURL TypeDescription
http://arxiv.org/abs/2011.07748arXivDiscussion Paper
ORCID:
AuthorORCID
Shi, Guanya0000-0002-9075-3705
Zhu, Yuke0000-0002-9198-2227
Additional Information:We would like to thank members of the NVIDIA AI Algorithms research team for their constructive feedback and Nathan V. Morrical for his help with the ViSII renderer.
Record Number:CaltechAUTHORS:20210225-132731801
Persistent URL:https://resolver.caltech.edu/CaltechAUTHORS:20210225-132731801
Usage Policy:No commercial reproduction, distribution, display or performance rights in this work are provided.
ID Code:108208
Collection:CaltechAUTHORS
Deposited By: George Porter
Deposited On:26 Feb 2021 15:09
Last Modified:26 Feb 2021 15:09

Repository Staff Only: item control page