Research on wide range localization for driverless vehicle
in outdoor environment based on particle filter
Jing Fu
1,2,3
,Tao Mei
3
,Hui Zhu
3
, Huawei Liang
3
, Biao Yu
3
1
Institute of Intelligent Machines, Chinese Academy of Sciences,
Hefei 230031, China
2
Department of Automation, University of Science and Technology of China,
Hefei 230027, China
3
Institute of Advanced Manufacturing Technology, Chinese Academy of Sciences,
Changzhou 213000, China
fujing2015@gmail.com
Abstract: The localization information is the premise of
robot autonomous navigation. In this paper, intelligent vehicle is
used as a platform for the present study. We perceive motion
information and environment signpost information through
wheel speed sensor and laser infrared radar SICK respectively,
and realize the global localization of the vehicle with the fusion of
the particle filter algorithm. The experimental result shows that
when the environment signpost position is known, the global
positioning error can be less than 1m.
Key words: localization, SICK, signpost, particle filter
I. Introduction
According to Leonard、Durrant-whyte [1], the major
research objects of autonomous navigation of mobile robot are
“Where am I?”, “What does the world look like?”, “Where do
I want to go?”, and “How do I get there?”, which can also be
interpreted as robot localization, map building, mission
planning and path planning. Among these research objects,
localization is not only the basic aspect of the research of
mobile robot, but also an indispensable part of the system,
since only by getting the precise position of the robot in the
map can effective planning decisions and efficient control be
made.
The robot localization can be divided into several types:
position tracking [2], global localization [3] and kidnapped
robot problem [4]. Position tracking is the simplest of these
localization issues. Given that the initial position of robot is
known, the task of position tracking is to eliminate the
accumulated error produced in the odometer track calculation
process, thus achieve accurate positioning. Kalman filter [5]
[6] and extended Kalman filter [7] methods are often used in
position tracking, since they can effectively track the position
of a robot. However, they must meet Gaussian distribution.
Compared with position tracking, global localization is much
more difficult. For this case, the initial position of the robot is
unknown, so it has to perceive the information of the
surrounding environment map to determine its own position.
The most difficult of all the localization types is the kidnapped
robot problem, where the robot in the moving process is
suddenly transferred to another position, yet it still memorize
the original position since it has not been informed.
Kidnapped robot problem is also often used in testing the
robot’s independent recovery ability in catastrophic
localization failure situation. As to global localization and
kidnapped robot problem, the methods often used are the grid
map method and Monte Carlo localization method [8] [9].
Monte Carlo localization method, which applies particle filter
to mobile robot localization [10], has the following advantages:
dealing with nonlinear and non-Gaussian systems, describing
multi-mode distribution, and realizing the global localization
of the robot. Of course there are some disadvantages of Monte
Carlo localization method, particle degradation and particle
poor problems are mainly included.
The present paper applies particle filter to vehicle
localization. We observe the signpost in the environment
978-1-4799-3977-0/14/$31.00 ©2014 IEEE
Proceedings of 2014 IEEE
International Conference on Mechatronics and Automation
August 3 - 6, Tianjin, China