Conditional Variational Autoencoders for Probabilistic Pose Regression

TL;DR

This paper introduces a probabilistic approach using Conditional Variational Autoencoders to improve robot pose estimation from images, especially in environments with repetitive structures, by modeling multiple pose hypotheses.

Contribution

It presents a novel training strategy for CVAEs that generates a generative model of camera poses, enabling sampling from the pose posterior distribution for better localization.

Findings

Outperforms existing methods in ambiguous environments

Supports multiple pose hypotheses through probabilistic modeling

Provides a theoretically grounded and streamlined approach

Abstract

Robots rely on visual relocalization to estimate their pose from camera images when they lose track. One of the challenges in visual relocalization is repetitive structures in the operation environment of the robot. This calls for probabilistic methods that support multiple hypotheses for robot's pose. We propose such a probabilistic method to predict the posterior distribution of camera poses given an observed image. Our proposed training strategy results in a generative model of camera poses given an image, which can be used to draw samples from the pose posterior distribution. Our method is streamlined and well-founded in theory and outperforms existing methods on localization in presence of ambiguities.