Page 117 - Autonomous Mobile Robots

100 Autonomous Mobile Robots

Acknowledgments ............................................................ 143
References .................................................................... 144
Biographies ................................................................... 146

3.1 INTRODUCTION
Data fusion is the process of combining sensory information from different
sources into one representational data format. The source of information may
come from different sensors that provide information about completely different
aspects of the system and its environment; or that provide information about
the same aspect of the system and its environment, but with different signal
quality or frequency. A group of sensors may provide redundant information,
in this case, the fusion or integration of the data from different sensors enables
the system to reduce sensor noise, to infer information that is observable but
not directly sensed, and to recognize and possibly recover from sensor failure.
If a group of sensors provides complementary information, data fusion makes
it possible for the system to perform functions that none of the sensors could
accomplish independently. In some cases data fusion makes it possible for
the system to use lower cost sensors while still achieving the performance
specification.
Data fusion is a large research area with various applications and methods
[1–3]. In addition to having a thorough understanding of various data fusion
methods, it is useful to understand which methods most appropriately fit the
corresponding aspects of a particular application. In many autonomous vehicle
applications it is useful to dichotomize the overall set of application information
into (internal) vehicle information and (external) environmental information.
One portion of the vehicle information is the vehicle state vector. Accurate
estimation of the vehicle state is a small portion of the data fusion problem
that must be solved onboard an autonomous vehicle to enable complex mis-
sions; however, accurate estimation of the vehicle state is critical to successful
planning, guidance, and control. When it is possible to analytically model the
vehicle state dynamics and the relation between the vehicle state and the sensor
measurements, the Kalman filter (KF) and the extended Kalman filter (EKF)
are often useful tools for accurately fusing the sensor data into an accurate
state estimate. In fact, when certain assumptions are satisfied, the KF is the
optimal state estimation algorithm. The KF and its properties are reviewed in
Section 3.2.
This chapter has two goals. The first is to review the theory and application
of the KF as a method to solve data fusion problems. The second is to discuss
the use of the EKF for fusing information from the global positioning system
(GPS) with inertial measurements to solve navigation problems for autonomous
vehicles. Variousfusionparadigmshavebeensuggestedintheliterature—GPS,

© 2006 by Taylor & Francis Group, LLC

FRANKL: “dk6033_c003” — 2006/3/31 — 16:42 — page 100 — #2