Page 118 - Autonomous Mobile Robots

Data Fusion via Kalman Filter 101

inertial navigation system (INS) only, INS with GPS resetting, INS with GPS
position aiding (i.e., loose coupling), and INS with GPS range aiding (i.e., tight
coupling). This chapter presents each approach and discusses the issues that
are expected to affect performance. Discussion of latency, asynchronous, and
nonlinear measurements are also included.

3.1.1 Data Fusion — GPS and INS
For planning, guidance, and control applications it is critical that the state of
the vehicle be accurately estimated. For these applications, the state vector of
the vehicle includes the three-dimensional (3D) position, velocity, and attitude.
Often, it is also possible to estimate the acceleration and angular rate. Various
sensor suites and data fusion methods have been considered in the literature
[4–8]. This chapter focuses on one of the most common sensor suites [9–11]
— fusion of data from an inertial measurement unit (IMU) and a GPS receiver.
The chapter considers the positive and negative aspects of various methods that
have been proposed for developing an integrated system.
An IMU provides high sample rate measurements of the vehicle acceler-
ation and angular rate. An INS integrates the IMU measurements to produce
position, velocity, and attitude estimates. INSs are self-contained and are not
sensitive to external signals. Since an INS is an integrative process, meas-
urement errors within the IMU can result in navigation errors that will grow
without bound. The rate of growth of the INS errors can be decreased through
the use of higher fidelity sensors or through sensor calibration. In addition,
the INS errors (and calibrations) can be corrected through the use of external
sensors. With a well-designed data fusion procedure, even an inexpensive INS
can provide high frequency precise navigation information [12]. The rate of
growth of the INS error will depend on the IMU characteristics and data fusion
approach.
A GPS receiver measures information that can be processed to directly
estimate the position and velocity of the receiver antenna. More advanced multi-
antenna GPS approaches can also estimate the vehicle attitude [13–15]. The
accuracy of the vehicle state estimate attained by GPS methods depends on the
receiver technology and the processing method. Civilian nondifferential GPS
users can attain position estimates accurate to tens of meters. Differential
GPS users can attain position estimates accurate to a few meters. Differen-
tial GPS users capable of resolving carrier phase integer ambiguities can attain
position estimates accurate to a few centimeters. The main disadvantage of state
estimates determined using GPS is that the estimates are dependent on reception
of at least four GPS satellite signals by the receiver. Satellite signal reception
requires direct line of sight between the receiver and the satellite. While this
line of sight is obstructed for a sufficiently large number of satellites, the GPS
solution will not be available.

© 2006 by Taylor & Francis Group, LLC

FRANKL: “dk6033_c003” — 2006/3/31 — 16:42 — page 101 — #3