Deep-Learning-Based Pedestrian Inertial Navigation: Methods, Data Set, and On-Device Inference

Modern inertial measurements units (IMUs) are small, cheap, energy efficient, and widely employed in smart devices and mobile robots. Exploiting inertial data for accurate and reliable pedestrian navigation supports is a key component for emerging Internet of Things applications and services. Recently, there has been a growing interest in applying deep neural networks (DNNs) to motion sensing and location estimation. However, the lack of sufficient labelled data for training and evaluating architecture benchmarks has limited the adoption of DNNs in IMU-based tasks. In this article, we present and release the Oxford Inertial Odometry Data Set (OxIOD), a first-of-its-kind public data set for deep-learning-based inertial navigation research with fine-grained ground truth on all sequences. Furthermore, to enable more efficient inference at the edge, we propose a novel lightweight framework to learn and reconstruct pedestrian trajectories from raw IMU data. Extensive experiments show the effectiveness of our data set and methods in achieving accurate data-driven pedestrian inertial navigation on resource-constrained devices.

I. Introduction

Modern microelectromechanical system (MEMS) inertial measurement units (IMUs) are small (a few mm²), cheap (several dollars a piece), energy efficient, and pervasive. As a low-cost yet powerful sensing modality, they have received a large amount of research effort and deeply weave into a wide range of applications. For instance, today’s smartphones come with embedded IMUs while users can use them for different location-based services, e.g., indoor navigation, localization, and outdoor trajectory analysis [1]. Moreover, emerging cyber gadgets, such as wristbands and VR/AR headsets, also actively utilize IMUs to provide continuous health monitoring [2], accurate activity tagging [3], and immersive gaming experiences [4]. On the side of robots and autonomous systems, IMUs are a long-standing sensing solution to navigation and grasping tasks [5].