Fifteen Years of Progress at Zero Velocity: A Review

Abstract

Fifteen years have passed since the publication of Foxlin's seminal paper "Pedestrian tracking with shoe-mounted inertial sensors". In addition to popularizing the zero-velocity update, Foxlin also hinted that the optimal parameter tuning of the zero-velocity detector is dependent on, for example, the user's gait speed. As demonstrated by the recent influx of related studies, the question of how to properly design a robust zerovelocity detector is still an open research question. In this review, we first recount the history of foot-mounted inertial navigation and characterize the main sources of error, thereby motivating the need for a robust solution. Following this, we systematically analyze current approaches to robust zero-velocity detection, while categorizing public code and data. The article concludes with a discussion on commercialization along with guidance for future research.

Three Symmetries for Data-Driven Pedestrian Inertial Navigation

“…1 (a). However, inertial navigation systems of this form are often plagued by modeling errors and systematic sensor errors (including biases and scale factor errors) [20]. In addition, there may exist motion patterns that are not accounted for in the motion models in h(•).…”

“…Specifically, in our first example with data-driven zero-velocity detectors, the navigation system will, regardless of how good the detector is, still be limited by the shortcomings of the zero-velocity model (non-zero mean of ZUPT errors, temporal correlation of ZUPT errors, etc.) [20]. Similarly, the downside of replacing u k in (1a) with a function learned from data is that this is an unnatural approach for breaking the cubic position error drift resulting from navigation based only on (1a), or for learning motion models related to the navigation state x, for example, motion models that constrain the speed to be within a specific interval.…”

“…However, there is nothing within the concept of inertial navigation itself that dictates that machine learning algorithms should not be of use. Quite the contrary, as has been shown multiple times, sensor errors and modeling errors lead to systematic estimation errors that cannot be described by standard, classical estimation frameworks using the kinematic inertial navigation equations, the zero-velocity model, and/or heuristic models for step estimation [19], [20]. Thus, the limited gain of data-driven approaches should not be taken for granted.…”

Inertial Navigation Using an Inertial Sensor Array

“…for a matrix L. The matrix L can be designed as L = diag(l x , l y , l z ), and for sufficiently high l i , the poles in (7) can be placed in the left-hand plane [32]. Thus, by providing feedback from a gyroscope triad the potentially unstable ODE in (6a) can be stabilized.…”

“…For instance, such aided inertial navigation has been done using satellite [2], video [3], radio [4], LIDAR [5], and magnetic field data [6]. Otherwise, the position and orientation errors' growth rate can be reduced by decreasing the measurement errors of the inertial sensors or using additional motion information such as Zero-velocity Updates (ZUPTs) [7]. Improving the sensor hardware [8] reduces measurement errors, but this typically comes with increasing cost and sensor size.…”