Page 352 - Introduction to AI Robotics

335
9.4 Associative Methods

Figure 9.8 Image measurements and the appropriate response from two different
views.

is in the neighborhood of the location, then the image measurements should
be approximately the same pattern as the image signature, only the pattern
may be offset due to the not-quite-at-the-right-place viewpoint.
If the robot can identify the image signature, or portion of it, in the current
image, it will then know whether to turn left or right to localize itself relative
to the location. The use of image signatures to direct a robot to a specific
location is called visual homing. The inspiration for visual homing came
from Nelson’s research into how bees navigate. It is easy to speculate that
baby bees are learning the image signatures of their hive as they execute the
zoom in/out behavior described in Ch. 3. In that case, the compound eyes
serve as de facto partitions of what to humans would be a single image.

9.4.2 QualNav
Levitt and Lawton took the ideas of neighborhoods and visual homing to an
extreme for outdoor navigation over large distances as part of the Defense
AUTONOMOUS LAND Advance Research Projects Agency (DARPA) Autonomous Land Vehicle (ALV)
VEHICLE (ALV) project in the late 1980’s. 85 At that time, topological navigation using rela-
tional graphs appeared promising for indoor environments, but seemed to
resist application to outdoor terrains. Part of the challenge was the notion of