A Localization and Navigation Method with ORB-SLAM
for Indoor Service Mobile Robots
Shirong Wang, Yuan Li, Yue Sun, Xiaobin Li, Ning Sun, Xuebo Zhang, Ningbo Yu*
Abstract— Autonomous mobile robots need to acquire envi-
ronment information for localization and navigation, and thus
are usually equipped with various sensors. Consequently, the
system is complex and expensive, bringing obstacles for general
home applications. In this paper, we present an efficient, yet
economic and simple solution for indoor autonomous robots,
consisting of a basic mobile platform, a Kinect V2 sensor
and a computing unit running Linux. Within the ROS en-
vironment, the ORB-SLAM algorithm, pointcloud processing
methods and a feedback controller have been developed and
implemented respectively for localization, obstacle detection and
avoidance, and navigation. Experimental results showed robust
localization, safe and smooth navigation, good motion control
accuracy and repeatability, demonstrating the efficacy of the
system architecture and algorithms.
I. INTRODUCTION
In recent years, along with the increasing number of aging
people and decrease of labor population, the demand for
service robots is sharply growing and autonomous mobile
robots for services at home, care-giving agencies, office
buildings, etc, are remarkably popular. Autonomous mobile
robots can assist human to complete various tasks, and they
need to acquire the current pose and environment information
for localization, motion planning and control [1], [2].
Autonomous mobile robots are usually equipped with
a variety of sensors, such as range sensors of laser or
ultrasound or infrared principles, GPS, IMU, camera, high
precision encoder and so on, to acquire information from
the environment in which they are deployed [3], [4], [5].
With these information, the mobile robot can localize itself
and make motion planning and control. More sensors can
improve the system performance, but will increase the com-
plication and expense of the system, making it difficult for
use by average people and hard for large-scale deployments.
As technologies are advanced, RGB-D sensors have been
developed, such as Microsoft Kinect, ASUS Xtion Pro Live,
Intel RealSense, etc. RGB-D sensors have the advantages
of rich environmental information, non-contact measurement
and low cost. With one RGB-D sensor, color and depth
This work is supported by the National Natural Science Foundation of
China (61403215), the Natural Science Foundation of Tianjin (13JCYB-
JC36600) and the Fundamental Research Funds for the Central Universities.
Corresponding author Assoc. Prof. Dr. Ningbo Yu is with the Institute
of Robotics and Automatic Information Systems, Nankai University, and
Tianjin Key Laboratory of Intelligent Robotics, Nankai University, Haihe
Education Park, Tianjin 300353, China. Phone: +86 (0)22 2350 3960 ext.
801, Email: nyu@nankai.edu.cn.
Mr. Shirong Wang, Mr. Yuan Li, Ms. Yue Sun, Mr. Xiaobin Li, Dr. Ning
Sun, and Assoc. Prof. Dr. Xuebo Zhang are with the Institute of Robotics
and Automatic Information Systems, Nankai University, and Tianjin Key
Laboratory of Intelligent Robotics, Nankai University, Haihe Education
Park, Tianjin 300353, China.
information can be obtained at the same time, and they have
been widely used in the field of robotics. Extensive research
has been focused on RGB-D based SLAM methods [6], [7],
[8]. Felix Endres et. al. from Freiburg University proposed
the RGB-D SLAM algorithm, using the Kinect sensor in
the ROS platform to realize autonomous localization and
mapping [9]. Daniel Maier et. al. used the NAO humanoid
robot together with ASUS Xtion Pro Live sensor to realize
autonomous localization, obstacle avoidance and motion
planning in indoor environment [10]. Washington University
and Microsoft lab developed a real-time visual SLAM system
based on graph optimization to build 3D maps [11]. Carnegie
Mellon University has developed a polygon reconstruction
and fusion algorithm to extract the planar features in the
original data [12]. Since the pointcloud map is not directly
used, the real-time performance of the algorithm can be
improved. Raul Mur-Artal et. al. proposed the ORB-SLAM
algorithm [13], taking advantages of the ORB features, and
the system operates well in real time even without GPU.
These work have significantly improved the accuracy and
real-time performance of RGB-D based SLAM techniques,
and thus promoted their application in mobile robots.
When autonomous mobile robots move in the environ-
ment, they need the ability to detect and avoid obstacle, in
addition to localization. Thus, various sensor technologies
have been taken to obtain the information of the obstacles
in the environment. Liu et. al. used raw point cloud from
3D laser to solve the 2.5D navigation problem [14]. In
recent years, with the growing application of RGB-D sensors,
studies on RGB-D based obstacle detection and avoidance
arose [15], [16].
In this work, we propose and establish a low-cost au-
tonomous mobile robot system, with a basic mobile platform
equipped with a general RGB-D sensor and a computing unit
running Linux. This mobile robot can autonomously localize
itself in the map, detect and avoid obstacles, and navigate in
the environment with its built in motion planning and control
algorithms. The robot is able to perform various service tasks
at different indoor environment such as home, care-giving
agencies, office buildings, etc, with its simple architecture
and minimized complexity. Furthermore, functionalities of
the system can be conveniently extended by software devel-
opment within the open ROS environment.
This paper is organized as following. The next section
describes in details the hardware and algorithms of the
mobile robot system. Section III presents the experiments
and results. Finally, section IV concludes the paper.
978-1-4673-8959-4/16/$31.00 © 2016 IEEE
443
Proceedings of The 2016 IEEE International
Conference on Real-time Computing and Robotics
June 6-9, 2016, Angkor Wat, Cambodia