Article
RTLIO: Real-Time LiDAR-Inertial Odometry and Mapping
for UAVs
Jung-Cheng Yang, Chun-Jung Lin , Bing-Yuan You , Yin-Long Yan and Teng-Hu Cheng *
Citation: Yang, J.-C.; Lin, C.-J.; You,
B.-Y.; Yan, Y.-L.; Cheng, T.-H. RTLIO:
Real-Time LiDAR-Inertial Odometry
and Mapping for UAVs. Sensors 2021,
21, 3955. https://doi.org/10.3390/
s21123955
Academic Editors: Kamil Krasuski
and Damian Wierzbicki
Received: 9 May 2021
Accepted: 4 June 2021
Published: 8 June 2021
Publisher’s Note: MDPI stays neutral
with regard to jurisdictional claims in
published maps and institutional affil-
iations.
Copyright: © 2020 by the authors.
Licensee MDPI, Basel, Switzerland.
This article is an open access article
distributed under the terms and
conditions of the Creative Commons
Attribution (CC BY) license (https://
creativecommons.org/licenses/by/
4.0/).
Department of Mechanical Engineering, National Yang Ming Chiao Tung University, Hsinchu 30010, Taiwan;
johnsongash.gdr07g@nctu.edu.tw (J.-C.Y.); chadlin.gdr07g@nctu.edu.tw (C.-J.L.);
physical31031.c@nycu.edu.tw (B.-Y.Y.); yanlong658.gdr08g@nctu.edu.tw (Y.-L.Y.)
* Correspondence: tenghu@g2.nctu.edu.tw
Abstract:
Most UAVs rely on GPS for localization in an outdoor environment. However, in GPS-
denied environment, other sources of localization are required for UAVs to conduct feedback control
and navigation. LiDAR has been used for indoor localization, but the sampling rate is usually too
low for feedback control of UAVs. To compensate this drawback, IMU sensors are usually fused
to generate high-frequency odometry, with only few extra computation resources. To achieve this
goal, a real-time LiDAR inertial odometer system (RTLIO) is developed in this work to generate high-
precision and high-frequency odometry for the feedback control of UAVs in an indoor environment,
and this is achieved by solving cost functions that consist of the LiDAR and IMU residuals. Compared
to the traditional LIO approach, the initialization process of the developed RTLIO can be achieved,
even when the device is stationary. To further reduce the accumulated pose errors, loop closure and
pose-graph optimization are also developed in RTLIO. To demonstrate the efficacy of the developed
RTLIO, experiments with long-range trajectory are conducted, and the results indicate that the RTLIO
can outperform LIO with a smaller drift. Experiments with odometry benchmark dataset (i.e., KITTI)
are also conducted to compare the performance with other methods, and the results show that the
RTLIO can outperform ALOAM and LOAM in terms of exhibiting a smaller time delay and greater
position accuracy.
Keywords: LiDAR-inertial odometry; state estimation; sensor fusion; SLAM
1. Introduction
1.1. Background
Precise ego-motion estimation and active perception play important roles when per-
forming navigation tasks or exploring unknown environments in robotics applications,
and the potential of small unmanned airborne (S-UAS) platforms applied to collect remote
sensing data have been analyzed [
1
]. Unmanned aerial vehicles (UAVs) running simultane-
ous localization and mapping (SLAM) algorithms can also be used to perform numerous
tasks, including surveillance, rescue, and transportation in extreme environments [
2
–
4
]. In
the field of SLAM, the performance of state estimation is highly reliant on sensors, such as
cameras, LiDAR, and inertial measurement units (IMUs). However, there are limitations
associated with each type of sensor, such as minimum illumination requirements and the
presence of noise. To overcome these shortcomings of stand-alone sensors, multiple sensors
have been used to increase the reliability of estimation [
5
–
9
]. The methods utilizing multiple
sensors for state estimation are categorized into two types: loosely coupled (cf. [
5
,
6
]) and
tightly coupled (cf. [
7
–
9
]). The tightly coupled approach directly fuses LiDAR and inertial
measurements through a joint optimization that minimizes some residuals, whereas the
loosely coupled approach deals with the multiple sensors separately. The tightly coupled
method is less computationally efficient and more difficult to implement than the loosely
coupled approach, but it is more robust in its approach to noise and more accurate [8].
Sensors 2021, 21, 3955. https://doi.org/10.3390/s21123955 https://www.mdpi.com/journal/sensors
