Aided inertial navigation system
Lead Scientist/Project Manager: Michael B. Matthews
Lead Engineer: Paul McGill
The term sensor fusion describes the process of assimilating
measurements from a suite of sensors with widely varying sampling
resolutions and latencies. By assimilation we mean combining the
observational information into an optimal state estimate of the observed
system. The sensors may exhibit timevarying behavior or statedependent
performance. They may be dynamically qualified within complex operating
envelopes, or they may be dynamically removed from operation altogether as
a result of component failure or scheduled servicing. New sensors may be
dynamically added to the system in the absence of complete configuration
information. Given such a dynamic suite of sensors, how do we determine in
real time the optimal state estimate of the system?
Vehicle navigation is an extreme example of the dynamic sensor fusion
problem. In the navigation problem, we wish to estimate the position and
velocity of the vehicle given measurements from a suite of inertial,
geomagnetic, and acoustic sensors. Each sensor is highly state dependent,
each operates only in a small region of the state space, and each provides
measurements of only a subset of the system states. Furthermore, the
sampling rates and measurement latencies of the various sensors differ by
two orders of magnitude. Given this dynamic situation, how do we estimate
the position and velocity of the vehicle?
In the Kalman filter approach to the state estimation problem, the
characteristics of each sensor are incorporated into the filter by
augmenting the system model with a model of the sensor. The Kalman gain is
computed from the augmented system model, and the state estimate is
computed from the Kalman gain as a linear combination of both model
predictions and sensor measurements. In the dynamic sensor problem
described above, the situation becomes very complex as the system model is
dynamically augmented and reduced with changing sensor complement.
Dynamic sensor fusion is a very general problem in the sense that basic
theoretical results are generally applicable to a broad class of
estimation problems including the multiresolution field estimation
problem currently under study. The practical implementation of these
results may greatly influence the realization of the recommended MBARI
ocean observing systems (MOOS) concept.
We propose (1) to study the aforementioned problem of dynamic sensor
fusion from within the context of vehicle navigation, and (2) to design,
build, and test a ringlasergyrobased navigation sensor data
concentrator for ROV Tiburon.
