Page 338 - Introduction to AI Robotics

321
Part II
digm. Most of the techniques presented in Part II will go into the deliberative
component of Hybrid architectures.
One important observation is that the four questions of navigation largely
ignore an implicit fifth question: how am I going to get there? Based on Part I,
the obvious answer is “by using reactive behaviors.” But navigation is delib-
erative, and the issue of integrating deliberation and reaction for navigation
in a Hybrid architecture is still largely open. Work addressing this issue of
interleaving planning and execution is presented in Ch. 9.

Spatial Memory

The answer to what’s the best way there? depends on the representation of
the world that the robot is using. The world representation will be called
SPATIAL MEMORY the robot’s spatial memory. 63 Spatial memory is the heart of the cartographer
object class (or its equivalent) in a Hybrid architecture, as described in Ch. 7.
Spatial memory should provide methods and data structures for process-
ing and storing output from current sensory inputs. For example, suppose
a robot is directed to “go down the hall to the third red door on the right.”
Even for the coordination and control of reactive behaviors, the robot needs
to operationalize concepts such as “hall,” “red,” “door” into features to look
for with a perceptual schema. It also needs to remember how many red doors
it has gone past (and not count the same door twice!). It would also be ad-
vantageous if the robot sensed a barrier or dead-end and updated its map of
the world.
Spatial memory should also be organized to support methods which can
extract the relevant expectations about a navigational task. Suppose a robot
is directed this time to to “go down the hall to the third door on the right.”
It could consult its spatial memory and notice that odd numbered doors are
red, and even numbered are yellow. By looking for “red” and “yellow” in
addition to other perceptual features of a door, the robot can more reliably
identify doors, either by focus of attention (the robot only runs door detec-
tion on red and yellow areas, not every image) or by sensor fusion (more
sources of data means a more certain percept).
Spatial memory supports four basic functions:

ATTENTION 1. Attention. What features, landmarks to look for next?

REASONING 2. Reasoning. Can that surface support my weight?

PATH PLANNING 3. Path planning. What is the best way through this space?