Page 78 - Introduction to AI Robotics

61
2.5 Advantages and Disadvantages
could be reused for a submarine, especially since RCS is not object-oriented.
In terms of robustness, RCS does attempt to provide some explicit mech-
anisms. In particular, it assumes the Value Judgment module simulates a
plan to confirm that it should be successful when deployed. The use of sim-
ulation is common for operating equipment in a well-known environment
where every piece of equipment is known. The most notable example is a
nuclear processing cell. With such detailed information, it is fairly straight-
forward (although computationally expensive) to simulate whether a par-
ticular course for a robot would collide with equipment and cause a spill.
This is a very limited form of robustness. The disadvantage is the time delay
caused by the robot mentally rehearsing its actions prior to executing them.
Simulation may not be appropriate for all actions; if a piece of the ceiling is
falling on the robot, it needs to get out of the way immediately or risk coming
up with the best place to move too late to avoid being crushed.

2.5 Advantages and Disadvantages

Robots built in the time period before 1990 typically had a Hierarchical style
of software organization. They were generally developed for a specific ap-
plication rather than to serve as a generic architecture for future applications.
The robots are interesting because they illustrate the diversity and scope of
applications being considered for mobile robots as far back as 15 or 20 years
ago.
The primary advantage of the Hierarchical Paradigm was that it provides
an ordering of the relationship between sensing, planning, and acting. The
primary disadvantage was planning. Every update cycle, the robot had to
update a global world model and then do some type of planning. The sens-
ing and planning algorithms of the day were extremely slow (and many still
are), so this introduced a significant bottleneck. Notice also that sensing and
acting are always disconnected. This effectively eliminated any stimulus-
response types of actions (“a rock is crashing down on me, I should move
anywhere”) that are seen in nature.
The dependence on a global world model is related to the frame problem.
In Strips, in order to do something as simple as opening a door, the robot had
to reason over all sorts of details that were irrelevant (like other rooms, other
doors). NHC and RCS represent attempts to divide up the world model into
pieces best suited for the type of actions; for example, consider the roles of
the Mission Planner, Navigator, and Pilot. Unfortunately, these decomposi-