Advantages and disadvantages

Advantages and disadvantages

Potential field styles of architectures have many advantages. The potential
field is a continuous representation that is easy to visualize over a large region
of space. As a result, it is easier for the designer to visualize the robot’s
overall behavior. It is also easy to combine fields, and languages such as C++
support making behavioral libraries. The potential fields can be parameterized:
their range of influence can be limited and any continuous function
can express the change in magnitude over distance (linear, exponential, etc.).
Furthermore, a two-dimensional field can usually be extended into a threedimensional
field, and so behaviors developed for 2D will work for 3D.
Building a reactive system with potential fields is not without disadvantages.
The most commonly cited problem with potential fields is that multiple
fields can sum to a vector with 0 magnitude; this is called the local
minima problem. Return to Fig. 4.19, the box canyon. If the robot was being
attracted to a point behind the box canyon, the attractive vector would cancel
the repulsive vector and the robot would remain stationary because all forces
would cancel out. The box canyon problem is an example of reaching a local
minima. In practice, there are many elegant solutions to this problem. One
of the earliest was to always have a motor schema producing vectors with a
smallmagnitude fromrandom noise.12 The noise in themotor schema would
serve to bump the robot off the local minima.
Another solution is that of NAVIGATION navigation templates (NaTs), as implemented
TEMPLATES by Marc Slack for JPL. The motivation is that the local minima problem
most often arises because of interactions between the avoid behavior’s repulsive
field and other behaviors, such asmove-to-goal’s attractive field. The
minima problem would go away if the avoid potential field was somehow
smarter. In NaTs, the avoid behavior receives as input the vector summed
fromthe other behaviors. This vector represents the direction the robot would
go if there were no obstacles nearby. For the purposes of this book, this will
be referred to as the strategic vector the robot wants to go. If the robot has a
strategic vector, that vector gives a clue as to whether an obstacle should be
passed on the right or the left. For example, if the robot is crossing a bridge
(see Fig. 4.27), it will want to pass to the left of obstacles on its right in order
to stay in the middle. Note that the strategic vector defines what is left and
what is right.
NaTs implement this simple heuristic in the potential field for RUNAWAY,
promoting it to a true AVOID. The repulsion field is now supplemented with
a tangential orbit field. The direction of the orbit (clockwise or counter

haviors. The coordinated control program is not specified; the designer
can use logic, finite state machines, whatever is deemed appropriate. Sequencing
is usually controlled by perceived cues or affordances in the environment,
which are releasers.
 Although all behaviors are treated equally, behaviors may make varying
contributions to the overall action of the robot. A behavior can change
the gains on another behavior, thereby reducing or increasing the magnitude
of its output. This means that behaviors can inhibit or excite other
behaviors, although this is rarely used in practice.
 Perception is usually handled by direct perception or affordances.
 Perception can be shared by multiple behaviors. A priori knowledge can
be supplied to the perceptual schemas, to emulate a specialized sensor
being more receptive to events such as hall boundary spacing.