EMA-VIO: Deep Visual-Inertial Odometry with External Memory Attention

TL;DR

This paper introduces EMA-VIO, a deep learning visual-inertial odometry model with external memory attention, achieving accurate and robust localization in challenging environments by effectively fusing multimodal sensor data.

Contribution

The paper presents a novel VIO framework utilizing external memory attention, improving data fusion efficiency and robustness over existing recurrent-based models.

Findings

Outperforms traditional and learning-based VIO methods in various scenes.

Maintains accuracy in challenging conditions like overcast days and water-filled terrains.

Demonstrates robustness and efficiency in pose estimation.

Abstract

Accurate and robust localization is a fundamental need for mobile agents. Visual-inertial odometry (VIO) algorithms exploit the information from camera and inertial sensors to estimate position and translation. Recent deep learning based VIO models attract attentions as they provide pose information in a data-driven way, without the need of designing hand-crafted algorithms. Existing learning based VIO models rely on recurrent models to fuse multimodal data and process sensor signal, which are hard to train and not efficient enough. We propose a novel learning based VIO framework with external memory attention that effectively and efficiently combines visual and inertial features for states estimation. Our proposed model is able to estimate pose accurately and robustly, even in challenging scenarios, e.g., on overcast days and water-filled ground , which are difficult for traditional VIO algorithms to extract visual features. Experiments validate that it outperforms both traditional and learning based VIO baselines in different scenes.