Page 79 - Introduction to AI Robotics

62
2 The Hierarchical Paradigm
tions appear to be dependent on a particular application. As a result, robotics
gained a reputation as being more of an art than a science.
Another issue that was never really handled by architectures in the Hier-
archical Paradigm was uncertainty. Uncertainty comes in many forms, such
as semantic (how close does NEXTTO mean anyway?), sensor noise, and ac-
tuator errors. Another important aspect of uncertainty is action completion:
did the robot actually accomplish the action? One robotics researcher said
that their manipulator was only able to pick up a cup 60% of the attempts;
therefore they had to write a program to check to see if it was holding a cup
and then restart the action if it wasn’t. Because Shakey essentially closed its
eyes during planning and acting, it was vulnerable to uncertainty in action
completion.

2.6 Programming Considerations

It is interesting to note that the use of predicate logic and recursion by Strips
favors languages like Lisp and PROLOG. These languages were developed
by AI researchers specifically for expressing logical operations. These lan-
guages do not necessarily have good real-time control properties like C or
C++. However, during the 1960’s the dominant scientific and engineering
language was FORTRAN IV which did not support recursion. Therefore, re-
searchers in AI robotics often chose the lesser of two evils and programmed
in Lisp. The use of special AI languages for robotics may have aided the split
between the engineering and AI approaches to robotics, as well as slowed
down the infusion of ideas from the the two communities. It certainly dis-
couraged non-AI researchers from becoming involved in AI robotics.
The Hierarchical Paradigm tends to encourage monolithic programming,
rather than object-oriented styles. Although the NHC decomposes the plan-
ning portion of intelligence, the decomposition is strictly functional. In par-
ticular, NHC and RCS don’t provide much guidance on how to build modu-
lar, reusable components.

2.7 Summary

The Hierarchical Paradigm uses a SENSE then PLAN then ACT (S,P,A).It
organizes sensing into a global data structure usually called a world model
that may have an associated knowledge base to contain a priori maps or
knowledge relevant to a task. Global data structures often flag that an ar-