Research article
Laser scan matching for self-localization of
a walking robot in man-made environments
Piotr Skrzypczynski
Poznan University of Technology, Poznan, Poland
Abstract
Purpose – The purpose of this paper is to describe a novel application of the well-established 2D laser scan-matching technique for self-localization of
a walking robot. The techniques described in this paper enable a walking robot with a 2D laser scanner to obtain precise maps of man-made
environments, which can be useful in search and reconnaissance missions, e.g. in warehouses, production plants, and other industrial areas.
Design/methodology/approach – The presented system combines two scan-matching algorithms (PSM and PLICP) to deal with low-quality range
data from a compact laser scanner and to provide robust self-localization in various types of man-made environments. Data from proprioceptive sensors
and simplifying assumptions holding in man-made environments are exploited to compensate for the varying attitude of the walking robot, particularly
in uneven terrain.
Findings – The experimental results suggest that neglecting either the poor initial pose guess obtained from the legged odometry, or the varying
attitude angles of a walking robot’s body, may lead to unacceptable results in self-localization and scan-based mapping. It is also demonstrated that
using the PSM algorithm to compute the initial pose estimate for the more precise PLICP scan-matching algorithm improves the results of self-
localization.
Research limitations/implications – So far, the presented self-localization system was tested in limited-scale indoor experiments. Experiments with
more extended and realistic scenarios are scheduled as further work.
Practical implications – Applying techniques described in this paper, the author was able to obtain the robot pose and precise maps of man-made
environments, which can be useful in USAR and reconnaissance missions, also in warehouses, production plants, and other industrial areas.
Originality/value – The scan-matching algorithms used in the presented research are not new, the contribution lies in combining them in order to
overcome issues specific to a small-size legged platform, using only common affordable hardware.
Keywords Programming and algorithm theory, Robots, Sensors, Lasers, Image scanners, Walking, Industrial robotics, Sensor review, Localization,
Languages, Scan matching
Paper type Research paper
1. Introduction
Legged robots are suitable for performing missions in
complex environments, because they are able to negotiate
many types of obstacles and climb stairs and the like, e.g. in
urban search and rescue (USAR) missions. Walking robots
may also find their way to commercial exploitation, as pointed
out in Rooks (2002), but research in perception and
navigation is still needed if real-life scenarios are to be tackled.
Due to the inherent properties of legged locomotion some
tasks that can be solved by standard algorithms in wheeled
robots are hard to solve if a legged robot is being used. One of
such tasks is self-localization, which is one of the central
problems for autonomous robots. This problem arises because
several implicit assumptions of the classic self-localization
algorithms are violated in the case of a legged robot.
The most common assumption is that the robot moves in a
flat world, and the sensory readouts (from range sensors in
particular) are always parallel to the ground plane.
Unfortunately, the legged locomotion is discrete, which often
causes uncontrolled oscillations of the attitude (i.e. pitch
w
and
roll
c
angles of the body). The vibratory motion of the walking
robot’s body makes it hard to keep the line of sight of any on-
board sensor perpendicular to the gravity vector all the time.
Another problem is the reliable dead reckoning
assumption – in most walking robots precise odometry is
not available (Gaßmann et al., 2005). Even if the robot has a
multi-sensory dead reckoning subsystem, e.g. employing
inertial sensors, rate-gyros, and a magnetic compass, the feet
slippages and inaccuracies in the complex mechanics make
the computed pose estimate very rough.
The situation is worsened by the limited payload of most
legged robots, which cannot carry large or heavy exteroceptive
sensors (Marques et al., 1999). Thus, in the case of range
sensors, miniature devices have to be used, such like the
Hokuyo URG-04LX, which yields low-quality range data
The current issue and full text archive of this journal is available at
www.emeraldinsight.com/0143-991X.htm
Industrial Robot: An International Journal
39/3 (2012) 242 –250
q Emerald Group Publishing Limited [ISSN 0143-991X]
[DOI 10.1108/01439911211217062]
Winning paper from CLAWAR 2011: Industrial Robot Journal Innovation
Award.
242