Page 111 - Autonomous Mobile Robots

94 Autonomous Mobile Robots

based on local signal-to-noise power estimates, found from several range
bins. The results show that the algorithm can detect features in the typically
cluttered outdoor environments tested, with a higher success rate compared to
the constant threshold and CFAR feature detection techniques.
A SLAM formulation using an augmented state vector which includes the
normalized RCS and absorption cross-sections of features, as well as the usual
feature Cartesian coordinates, was introduced. This is intended to aid the data
association process, so that features need not just be associated based on their
Cartesian coordinates, but account can be taken of their estimated normalized
reflection and absorption cross-sections also.
The final contribution is a predictive model of the form and magnitudes
of the power–range spectra from differing vehicle locations, for multiple line-
of-sight targets. This forms a predicted power–range observation, based on
estimates of the augmented SLAM state. The formulation of power returns
from multiple objects down-range is explained and predicted RADAR range
spectra are compared with real spectra, recorded outdoors.
This work is a step toward building reliable maps and localizing a vehicle
to be used in mobile robot navigation. Further methods of including the target
presence probability of feature estimates into SLAM are being investigated.

APPENDIX
The binary hypothesis testing problem is a special case of decision problems.
The decision space consists of target presence and target absence represented
by δ 0 and δ 1 , respectively. There is a hypothesis corresponding to each decision.
H 0 is called null hypothesis (hypothesis accepted by choosing decision δ 0 ) and
H 1 is called the alternative hypothesis. The binary hypothesis problem has four
possible outcomes:
• H 0 was true, δ 0 is chosen : correct decision.
• H 1 was true, δ 1 is chosen : correct decision.
• H 0 was true, δ 1 is chosen : False alarm, also known as a type I error.
• H 1 was true, δ 0 is chosen : missed detection also known as a type II
error.

ACKNOWLEDGMENTS
This work was funded under the first author’s AcRF Grant, RG 10/01,
Singapore. We gratefully acknowledge John Mullane for providing some of
the outdoor RADAR scans and Javier Ibanez-Guzman, SIMTech Institute of
Manufacturing Technology, Singapore, for use of the utility vehicle. We further
acknowledge the valuable advice from Graham Brooker (Australian Centre for
Field Robotics) and Steve Clark (Navtech Electronics, UK).

© 2006 by Taylor & Francis Group, LLC

FRANKL: “dk6033_c002” — 2006/3/31 — 17:29 — page 94 — #54