VOCAL: Visual Odometry via ContrAstive Learning

TL;DR

VOCAL introduces a novel contrastive learning framework for visual odometry that improves interpretability and robustness by framing VO as a label ranking problem and integrating Bayesian inference.

Contribution

It redefines visual odometry as a label ranking task using contrastive learning, enhancing interpretability and multimodal data compatibility within a data-driven framework.

Findings

Outperforms existing methods on KITTI dataset

Improves interpretability of visual odometry models

Enhances flexibility with multimodal data

Abstract

Breakthroughs in visual odometry (VO) have fundamentally reshaped the landscape of robotics, enabling ultra-precise camera state estimation that is crucial for modern autonomous systems. Despite these advances, many learning-based VO techniques rely on rigid geometric assumptions, which often fall short in interpretability and lack a solid theoretical basis within fully data-driven frameworks. To overcome these limitations, we introduce VOCAL (Visual Odometry via ContrAstive Learning), a novel framework that reimagines VO as a label ranking challenge. By integrating Bayesian inference with a representation learning framework, VOCAL organizes visual features to mirror camera states. The ranking mechanism compels similar camera states to converge into consistent and spatially coherent representations within the latent space. This strategic alignment not only bolsters the interpretability of the learned features but also ensures compatibility with multimodal data sources. Extensive evaluations on the KITTI dataset highlight VOCAL's enhanced interpretability and flexibility, pushing VO toward more general and explainable spatial intelligence.