Page 316 - Introduction to Autonomous Mobile Robots

301
Planning and Navigation
and so this method is not useful as a general solution to the navigation problem. However
such robotic systems do exist, and this method can be useful in two cases:

Static route-based applications. In mobile robot applications where the robot operates
in a completely static environment using a route-based navigation system, it is conceivable
that the number of discrete goal positions is so small that the environmental representation
can directly contain paths to all desired goal points. For example, in factory or warehouse
settings, a robot may travel a single looping route by following a buried guidewire. In such
industrial applications, path-planning systems are sometimes altogether unnecessary when
a precompiled set of route-based solutions can be easily generated by the robot program-
mers. The Chips mobile robot is an example of a museum robot that also uses this architec-
ture (118). Chips operates in a unidirectional looping track defined by its colored
landmarks. Furthermore, it has only twelve discrete locations at which it is allowed to stop.
Due to the simplicity of this environmental model, Chips contains an executive layer that
directly caches the path required to reach each goal location rather than a generic map with
which a path planner could search for solution paths.

Extreme reliability demands. Not surprisingly, another reason to avoid on-line planning
is to maximize system reliability. Since planning software can be the most sophisticated
portion of a mobile robot’s software system, and since in theory at least planning can take
time exponential to the complexity of the problem, imposing hard temporal constraints on
successful planning is difficult if not impossible. By computing all possible solutions off-
line, the industrial mobile robot can trade versatility for effective constant-time planning
(while sacrificing significant memory of course). A real-world example of off-line plan-
ning for this reason can be seen in the contingency plans designed for space shuttle flights.
Instead of requiring astronauts to problem-solve on-line, thousands of conceivable issues
are postulated on Earth, and complete conditional plans are designed and published in
advance of the Shuttle flights. The fundamental goal is to provide an absolute upper limit
on the amount of time that passes before the astronauts begin resolving the problem, sacri-
ficing a great deal of ground time and paperwork to achieve this performance guarantee.

6.3.4.2 Episodic planning
The fundamental information-theoretic disadvantage of planning off-line is that, during
run-time, the robot is sure to encounter perceptual inputs that provide information, and it
would be rational to take this additional information into account during subsequent exe-
cution. Episodic planning is the most popular method in mobile robot navigation today
because it solves this problem in a computationally tractable manner.
As shown in figure 6.25, the structure is three-tiered as in the general architecture of
figure 6.23. The intuition behind the role of the planner is as follows. Planning is compu-