Coordination and Control of Behaviors

Coordination and Control of Behaviors

KONRAD LORENZ Konrad Lorenz and Niko Tinbergen were the founding fathers of ethology.
NIKO TINBERGEN Eachman independently became fascinated not only with individual behaviors
of animals, but how animals acquired behaviors and selected or coordinated
sets of behaviors. Their work provides some insight into four different
ways an animal might acquire and organize behaviors. Lorenz and Tinbergen’s
work also helps with a computational theory Level 2 understanding of
how to make a process out of behaviors.
The four ways to acquire a behavior are:
INNATE 1. to be born with a behavior (innate). An example is the feeding behavior in
baby arctic terns. Arctic terns, as the name implies, live in the Arctic where
the terrain is largely shades of black and white. However, the Arctic tern
has a bright reddish beak. When babies are hatched and are hungry, they
peck at the beak of their parents. The pecking triggers a regurgitation
reflex in the parent, who literally coughs up food for the babies to eat. It
turns out that the babies do not recognize their parents, per se. Instead,
they are born with a behavior that says: if hungry, peck at the largest red
blob you see. Notice that the only red blobs in the field of vision should
be the beaks of adult Arctic terns. The largest blob should be the nearest
parent (the closer objects are, the bigger they appear). This is a simple,
effective, and computationally inexpensive strategy.
SEQUENCE OF INNATE 2. to be born with a sequence of innate behaviors. The animal is born with a
BEHAVIORS sequence of behaviors. An example is the mating cycle in digger wasps.

A female digger wasp mates with a male, then builds a nest. Once it sees
the nest, the female lays eggs. The sequence is logical, but the important
point is the role of stimulus in triggering the next step. The nest isn’t built
until the female mates; that is a change in internal state. The eggs aren’t
laid until the nest is built; the nest is a visual stimulus releasing the next
step. Notice that the wasp doesn’t have to “know” or understand the
sequence. Each step is triggered by the combination of internal state and
the environment. This is very similar to Finite State Machines in computer
science programming, and will be discussed later in Ch. 5.
3. to be INNATE WITH MEMORY born with behaviors that need some initialization (innate with memory).
An animal can be born with innate behaviors that need customizing
based on the situation the animal is born in. An example of this is bees.
Bees are born in hives. The location of a hive is something that isn’t innate;
a baby bee has to learn what its hive looks like and how to navigate
to and from it. It is believed that the curious behavior exhibited by baby
bees (which is innate) allows them to learn this critical information. A
new bee will fly out of the hive for a short distance, then turn around and
come back. This will get repeated, with the bee going a bit farther along
the straight line each time. After a time, the bee will repeat the behavior
but at an angle from the opening to the hive. Eventually, the bee will have
circumnavigated the hive. Why? Well, the conjecture is that the bee is
learning what the hive looks like from all possible approach angles. Furthermore,
the bee can associate a view of the hive with a motor command
(“fly left and down”) to get the bee to the opening. The behavior of zooming
around the hive is innate; what is learned about the appearance of the
hive and where the opening is requires memory.
LEARN 4. to learn a set of behaviors. Behaviors are not necessarily innate. In mammals
and especially primates, babies must spend a great deal of time
learning. An example of learned behaviors is hunting in lions. Lion
cubs are not born with any hunting behaviors. If they are not taught
by their mothers over a period of years, they show no ability to fend
for themselves. At first it might seem strange that something as fundamental
as hunting for food would be learned, not innate. However, consider
the complexity of hunting for food. Hunting is composed of many
sub-behaviors, such as searching for food, stalking, chasing, and so on.
Hunting may also require teamwork with other members of the pride. It
requires great sensitivity to the type of the animal being hunted and the
terrain. Imagine trying to write a program to cover all the possibilities!

While the learned behaviors are very complex, they can still be represented
by innate releasing mechanisms. It is just that the releasers and
actions are learned; the animal creates the program itself.
Note that the number of categories suggests that a roboticist will have a spectrum
of choices as to how a robot can acquire one or more behaviors: from
being pre-programmed with behaviors (innate) to somehow learning them
(learned). It also suggests that behaviors can be innate but require memory.
The lesson here is that while S-R types of behaviors are simple to preprogram
or hardwire, robot designers certainly shouldn’t exclude the use
of memory. But as will be seen in Chapter 4, this is a common constraint
placed on many robot systems. This is especially true in a popular style of
hobby robot building called BEAM robotics (biology, electronics, aesthetics,
andmechanics), espoused by Mark Tilden. Numerous BEAM robotweb sites
guide adherents through construction of circuits which duplicate memoryless
innate reflexes and taxes.
An important lesson that can be extracted from Lorenz and Tinbergen’s
work is that the internal state and/or INTERNAL STATE motivation of an agent may play a role
MOTIVATION in releasing a behavior. Being hungry is a stimulus, equivalent to the pain
introduced by a sharp object in the robot’s environment. Another way of
looking at it is that motivation serves as a stimulus for behavior. Motivations
can stem from survival conditions (like being hungry) or more abstract
goals (e.g., need to check the mail). One of the most exciting insights is that
behaviors can be sequenced to create complex behaviors. Something as complicated
as mating and building a nest can be decomposed into primitives
or certainly more simple behaviors. This has an appeal to the software engineering
side of robotics.