Page 322 - Introduction to AI Robotics

8.6 Emergent Social Behavior
no awareness of the goals of the other team members. 305
Now consider what happens when a robot ant encounters an asteroid it
can’t move. The robot stays there pushing. Eventually another robot will
come along because the asteroid is not moving. As it is attracted to the “dark
side” of the asteroid, it will come into range of the first robot. What hap-
pens? The avoid-robot behavior should be instantiated, causing the first ro-
bot to move over a bit. The second robot will also feel a repulsive force and
slow down. As the first robot moves out of the way, the angle of repulsion
changes, forcing the second robot to move sideways as well, as it continues
to move to the asteroid. Together, the interaction between the two robots
should cause them to naturally balance themselves behind the asteroid and
push together. The point is that the robots were not explicitly directed to all
work on the same NEO; they were each directed to find their own NEO, but
circumstances led them to the same one.

8.6 Emergent Social Behavior

The examples of heterogeneity, cooperation, control, and goals give some
hint of how an overall social behavior emerges from the actions of autono-
mous robots. The robot teams often are the result of extensive design efforts,
where the teams aren’t too large to interfere with each other, and are opti-
mally sized for the particular task, etc. Many researchers are exploring the
issues of what happens when the designer doesn’t have a choice about the
size of the robot population. How do social behaviors emerge in those cases?
And how can social rules or conventions be established to make the team
self-regulating and productive? This section summarizes two approaches:
creating social rules for the robots to follow, and allowing internal motiva-
tion to cause the robots to adapt their behavior to problems.

8.6.1 Societal rules

Maja Mataric has focused her researchon how group dynamics might emerge
in herds of multiple agents operating under fully distributed control. She ex-
plored the impact of density and the impact of societal rules on overall team
performance. 90 Each IS Robotics R2 robot was programmed with behaviors
using the Subsumption architecture. She set up a scenario where up to 20
identical robots (now known as “The Nerd Herd”) were given the same lo-
cation as a goal. The goal, however, was on the other side of a partition with
a narrow door, permitting only one robot to pass through the partition at a