Page 199 - Autonomous Mobile Robots

Landmarks and Triangulation in Navigation 183

The target cylinders were located at (0, 1) and (−1, 0) because in these positions
◦
they can always be observed by the 180 scanner, allowing continuous position
updates. The localization error can easily be extracted from Figure 4.20 and is
of the order of 0.03 m. The position accuracy is better toward the origin of the
graph because the robot is nearer to the target positions of (0, 1) and (−1, 0).

4.6 CONCLUSIONS
This chapter addresses the problem of landmarks and triangulation in naviga-
tion of mobile robots. A novel landmark-based navigation system is proposed,
which consists of three types of landmarks (retro-reflective, digital, and geo-
metric) and three types of sensors (laser, vision, and odometry), as well as sonar
sensors. Somecorrespondingnavigationandtriangulationalgorithmshavebeen
developed so that the robot is able to estimate its position and update its internal
map continuously in a dynamic environment.
To improve the localization accuracy for mobile robots in continuous opera-
tion, the EKF algorithm has been adopted in the navigation process to integrate
odometry data and angle observations from the laser scanner in order to provide
a useful solution toward real-world applications. A triangulation module is
embedded into the proposed architecture to re-calibrate the robot’s position
when the robot is stationary or gets lost. The experimental results are presented
to show its applicability.
A digitallandmark-based localization algorithm formobile robotsisdemon-
strated, which uses a fast digits recognition method. The algorithm provides
an easy solution to landmark identification in complex environments, which
is robust to slope images. Some advantages of the algorithm are the flexible
extendibility of digital landmarks and the low computation cost of landmark
recognition. We are currently investigating the following four issues (i) other
information in the single landmark, for example, the edges of middle characters,
may be used; (ii) other data fusion methods to use pre-known position inform-
ation (from dead-reckoning or EKF); (iii) multiple landmarks may be seen in
some conditions, and triangulation methods may be used; and (iv) localization
algorithms based on texture landmarks.
The feasibility of cooperative localization based on one sensing robot and
two landmark robots is also investigated. It is implemented by a least squares
fitting approach optimized for the sequential natural of the range data and
the highly symmetric aspect of the circular geometric targets. This coloca-
tion scheme allows fast position and orientation determination with bounded
errors and reliability indicators in unknown indoor environments. The robust
localization algorithm lays the foundation for mapping featureless and highly
symmetric environments. Continuous localization was performed at 0.2 m/sec.
Continuous localization can be provided, however these scans should not be
incorporated into the global map, only the ones taken when stationary should

© 2006 by Taylor & Francis Group, LLC

FRANKL: “dk6033_c004” — 2006/3/31 — 16:42 — page 183 — #35