Environmental Engineering Reference
In-Depth Information
view of the world. This is often done by having each sensor perform its sensor-
specific processing and then merge the sensor output [
91
,
192
]. The techniques
used vary between domains and the types of sensors being used.
One advantage of data fusion is that it can supply a more accurate under-
standing of the environment. This is because the data missing from one sensor
can be filled in by another. Also, when multiple potential interpretations are
possible in the data from a single sensor, information supplied from another
sensor can help remove ambiguities.
Data fusion is a complex subject and many design decisions must be made
when designing a system. Some of the important issues are:
•
How is the sensed data represented? (pixels, numbers, vectors, symbols,
etc.)
•
How are the different sensor representations merged into the common
view?
•
How is conflicting information handled? Does one viewpoint win or are all
views represented?
•
Can uncertainty be represented with appropriate weightings?
•
Can information from one sensor be used to fill in holes in the information
from another?
•
What is the level of granularity in the data (pixels, symbols, etc.) and how
will the discrepancies between data granularity and model granularity be
handled?
5.7 Testing Technologies
The development of robust agent systems requires testing. Testing supports
the detection of errors in implementation where the system implementers over-
looked or misunderstood a requirement, or just made a mistake. But another
major purpose of testing is to uncover requirements that are missing in the
original specifications when the system was designed.
A complete testing plan requires testing at each stage of the development
effort and involves numerous strategies. Ultimately, the actual system (hard-
ware and software) should be tested in a realistic environment that is capable
of exercising both the nominal situations and many error cases. While this
level of testing is important, it can be very expensive. As the cost of compu-
tation declined, it became possible to develop very realistic testing environ-
ments that exist solely in a computer without physical test hardware. This
section will limit itself to software testing relative to cooperative autonomy.
5.7.1 Software Simulation Environments
Software simulation environments are based on the idea that it is possible
to use computerized simulations to model accurately not only the system
being tested, but also its operating environment. While they cannot replace
