IEEE ROBOTICS AND AUTOMATION LETTERS. PREPRINT VERSION. ACCEPTED JANUARY, 2020 1
A Lightweight and Accurate Localization Algorithm
Using Multiple Inertial Measurement Units
Ming Zhang, Xiangyu Xu, Yiming Chen, and Mingyang Li
Abstract—This paper proposes a novel inertial-aided local-
ization approach by fusing information from multiple inertial
measurement units (IMUs) and exteroceptive sensors. IMU is a
low-cost motion sensor which provides measurements on angular
velocity and gravity compensated linear acceleration of a moving
platform, and widely used in modern localization systems. To
date, most existing inertial-aided localization methods exploit
only one single IMU. While the single-IMU localization yields
acceptable accuracy and robustness for different use cases, the
overall performance can be further improved by using multiple
IMUs. To this end, we propose a lightweight and accurate
algorithm for fusing measurements from multiple IMUs and
exteroceptive sensors, which is able to obtain noticeable per-
formance gain without incurring additional computational cost.
To achieve this, we first probabilistically map measurements
from all IMUs onto a virtual IMU. This step is performed
by stochastic estimation with least-square estimators and prob-
abilistic marginalization of inter-IMU rotational accelerations.
Subsequently, the propagation model for both state and error
state of the virtual IMU is also derived, which enables the
use of the classical filter-based or optimization-based sensor
fusion algorithms for localization. Finally, results from both
simulation and real-world tests are provided, which demonstrate
that the proposed algorithm outperforms competing algorithms
by noticeable margins.
Index Terms—Sensor Fusion; Localization; SLAM.
I. INTRODUCTION
I
N recent years, commercial products which exploit inertial
measurement units (IMUs) have been under fast develop-
ment. This popular motion sensor can be found in robotics,
personal electronic devices, wearable devices, and so on [1].
On one hand, the maturity of MEMS manufacturing process
significantly reduces the size, price, and power consumption
of the IMU hardware. On the other hand, significant progress
has also been made in both algorithm and software design
for IMUs, ranging from sensor characterization and calibra-
tion [2]–[5], measurement integration [6]–[8], sensor fusion
[9]–[13], and so on.
In this work, we focus on the inertial-aided localization,
which is to estimate the 6D poses (3D position and 3D
orientation) of a moving platform. Since localization with
only IMU inevitably suffers from pose drift, measurements
from other sensors (i.e. aiding), e.g., RGB cameras, depth
Manuscript received: September, 10, 2019; Revised December, 17, 2019;
Accepted January, 13, 2020.
This paper was recommended for publication by Editor Eric Marchand upon
evaluation of the Associate Editor and Reviewers’ comments.
The authors are with Alibaba Group, Hangzhou,
China. {mingzhang, xiangyuxu, yimingchen,
mingyangli}@alibaba-inc.com.
Digital Object Identifier (DOI): see top of this page.
Fig. 1. The IMU array board used in this work, which contains nine ST
LSM6DSOX IMUs marked by red rectangles and a processor interface to
connect cameras. The IMUs are synchronized by an embedded processor.
cameras, or LiDARs (Light Detection And Ranging sensors),
are typically used in combination with IMUs to provide long-
term performance guarantees [12], [14], [15]. To perform
accurate pose estimation, the majority of existing works use
measurements from IMU for pose prediction, which is fol-
lowed by probabilistic refinement using measurements from
other sensors [7]–[12].
To date, most algorithms on inertial-aided localization are
designed based on a single IMU [7]–[13]. Although these
algorithms are successfully deployed in different applications,
using additional IMU sensors creates new possibilities for
further improving the system accuracy and robustness. Com-
pared to other popular sensors for localization (e.g., cameras
or LiDARs), IMUs especially the off-the-shelf MEMS ones
are priced only hundredths or thousandths, and of smaller
size as well as lower power consumption. In addition, as a
reliable proprioceptive sensor, IMU also poses less restrictions
on operating environments and hardware configurations (in
contrast, e.g. stereo cameras require enough spatial baseline
to achieve performance gain [16], [17], which might not be
feasible on various applications including mobile devices).
Most existing methods on using multiple IMUs focus on
processing IMU measurements only or integration with global
navigation satellite systems (GNSS) [10], [18], [19]. Fusing
measurements from multiple IMUs with exteroceptive sensors
for localization is a less-explored topic. To the best of the
authors’ knowledge, the only work in recent years in this
domain is [20], which proposed an approach for vision-aided
inertial navigation using measurements from multiple IMUs.
However, the proposed algorithm is of significantly increased
arXiv:1909.04869v2 [cs.RO] 17 Jan 2020