This application of SRI's evidential
reasoning technology uses models of antiair threat and emitter
operations such as SA-6 and PATHAND, in combination with continuously
valued environmental constraints such as clarity, illumination, and range,
to probabilistically identify antiair threats and their modes of operation
based upon sensor reports.
The problem is that of a penetrating aircraft flying over hostile
territory. Any number of different surface-to-air missile systems and
antiaircraft artillery might fire on the aircraft. For the aircraft to
avoid or neutralize these threats, it must first detect and identify them.
To do so, it has a number of different sensors that can be tuned to detect
and identify various threats in different modes of operation.
As an aircraft approaches one of these threats, it goes through a number of
operational steps, each with characteristically different observables.
Typically, a threat begins in surveillance mode and transitions to
acquisition mode, which is followed by target tracking, launch, and
guidance modes. The pattern of emissions through these steps aids the
identification process. A complicating factor is that many of these threats
have both electro-optical and radar guidance modes that they select based
upon environmental conditions. The problem is to identify the threats and
their modes of operation as quickly as possible.
The development of this application began with the construction of two
frames of discernment: one models the possible threats and
their modes of operation, and the other does the same for the component
emitters. The illustration (upper right) includes the model for an SA-6
(a.k.a. GAINFUL). It has five generic modes of operation, each with a
specific electro-optical and radar option. A compatibility
relation captures the allowable transitions among the modes of
operation, while another compatibility relation constrains the modes by the
range to the target aircraft. Other frames and relations further constrain
the modes according to the clarity and illumination in the environment. The
threat models are hierarchically related to the emitter models.
Collectively, these frames and relations constitute Gister's
gallery that models the performance envelopes of the
anticipated antiair threats.
A Gister analysis specifies how sensory data are
processed to arrive at a consensus interpretation. In this case, the
analysis (lower right) is a circuit that continually assesses the evolving
situation based upon both newly acquired sensory data and the previous
conclusions. When a sensor detects a threat or emitter, a probabilistic
identification is entered at one of the sensory input nodes along with a
probabilistic estimate of the range. Through the analysis circuit, the
immediately prior conclusions are projected forward to the current
time, according to the possible state transitions in the gallery, and then
this updated historic information is fused with the newly acquired data
and the translated
constraints imposed by the current environmental conditions. As a result,
probabilistic estimates of the threat's type and mode of operation are
produced (upper left). In turn, these estimates are fed forward, helping to
constrain the conclusions during the next iteration.
This application of evidential reasoning illustrates the use of an analysis
circuit to make continual updates to conclusions in an evolving situation.
It also makes use of continuously valued environmental measures as a means
of constraining its conclusions; probabilistic information is produced and
consumed at two levels of abstraction (i.e., threats and their component
emitters); both the identity and activity of the target systems are
predicted. These predictions are used to better interpret future sensory
information and to perform sensor tasking. The demonstration scenario
includes reports from radar warning receivers and imaging electro-optical
John D. Lowrance,
Artificial Intelligence Center
Gister and Gister-CL are trademarks of SRI International.
Copyright © 1995 SRI International, 333 Ravenswood Ave., Menlo Park, CA 94025 USA.
All rights reserved.