This application of SRI's evidential
reasoning technology highlights its ability to simultaneously
reason with hard constrains, such as island barriers and maximum speed of
vehicles, in combination with diverse sources of soft constraints of
varying credibility, such as reports from stationary sensors, mobile
sensors, direct sightings, and indirect sightings (e.g., bubbles), to
establish the likely paths of underwater vehicles.
Minisubmarines that use tank-like treads to crawl through shallow waters
were reportedly used by the Soviets off the coast of Sweden (When Soviet
Subs Lurk Near Sweden's Coast, Jack Anderson and Dale Van Atta, United
Feature Syndicate). These waters are characterized by thousands of small
islands that inhibit the use of conventional submarines. Sightings of
bubbles by fishermen and periscopes by sunbathers, along with reported
sensor contacts by Swedish military vessels and sensor nets, support this
conjecture. However, such reports are scarce and difficult to interpret.
The problem is to reconstruct what might have happened after the fact. How
many subs must there have been to explain the data? Where were they?
The first step in building this application was to develop a frame
of discernment for the possible locations of these subs. We
overlaid a grid on a map of the area of interest, determined which grid
cells contained water at a sufficient depth to support these subs, and
enumerated these cells as elements of the frame (upper left). Next, the
cells were linked in a compatibility relation to capture
how these subs could move from one cell to another with the passage of an
hour (upper right). Since islands prevent direct transit between
neighboring cells, they are reflected in the compatibility relation by the
absence of connecting arcs. This frame and compatibility relation
constitute Gister's gallery that delimits the range of
possibilities for how these subs might move.
Each sub contact is represented in a Gister analysis as
a report. Each indicates an area within which a sub was located, the time
of the contact, and a degree of confidence that captures the credibility of
the source. Utilizing Gister's projection capabilities, the
reported probabilistic locations of subs can be moved forward or back in
time via the compatibility relation. When multiple reports are
synchronized, they are fused to form a probabilistic
consensus. A byproduct of fusion is a numerical degree of
conflict. Here it is used to cluster compatible reports
into a single track, while incompatible reports are assumed to be referring
to different subs. Given a cluster of compatible reports, an analysis
(lower left) is constructed that calculates the tightest possible bounds on
that sub's location over time. Each report probabilistically constrains the
sub's location during each hour of the analysis.
This application of evidential reasoning highlights a number of
capabilities: natural constraints, such as islands and maximum speed of
travel, are captured as logical (certain) constraints within the gallery;
uncertain constraints, corresponding to negative sensor net reports, are
easily incorporated in the analyses; other diverse sources of probabilistic
information are incorporated as reports with varying degrees of
credibility. We demonstrate a semiautomated approach to determining the
number of minisubs that best explain the data and a fully automated
approach to subsequently determining their probable locations over time.
The demonstration scenario includes two minisubs that give rise to five
reported contacts from disparate sources spaced over a ten-hour period.
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.