Page 313 - Introduction to AI Robotics
P. 313
296
8
Multi-agents
or characteristics, in describing a team. For the purposes of this chapter,
heterogeneity, control, cooperation, and goals will be used as the dimensions. 117
8.2 Heterogeneity
HETEROGENEITY Heterogeneity refers to the degree of similarity between individual robots that
are within a collection. Collections of robots are characterized as being ei-
HETEROGENEOUS ther heterogeneous or homogeneous. Heterogeneous teams have at least two
TEAMS members with different hardware or software capabilities, while in homoge-
HOMOGENEOUS TEAMS
neous teams the members are all identical. To make matter more confusing,
members can be homogeneous for one portion of a task by running identi-
cal behaviors, then become heterogeneous if the team members change the
behavioral mix or tasks.
8.2.1 Homogeneous teams and swarms
Most multi-agent teams are homogeneous swarms. Each robot is identical,
which simplifies both the manufacturing cost and the programming. The
biological model for these teams are often ants or other insects which have
large numbers of identical members. As such, swarms favor a purely reactive
approach, where each robot operates under the Reactive Paradigm. Insect
swarms have been modeled and mimicked since the 1980’s. The proceedings
of the annual conference on the Simulation of Adaptive Behavior (also called
“From Animals to Animats”) is an excellent starting point.
An example of a successful team of homogeneous robots is Ganymede, Io,
and Callisto fielded by Georgia Tech. These three robots won first place in the
“Pick Up the Trash” event of the 1994 AAAI Mobile Robot Competition, 129
also discussed in Ch. 5. Recall that the objective of that event was to pick up
the most trash (coca-cola cans) and deposit it in a refuse area. The majority
of the entries used a single agent, concentrating on model-based vision for
recognizing trash, cans, and bins and on complex grippers.
The three identical robots entered by Georgia Tech were simple, both phys-
ically and computationally, and are described in detail in a 1995 AI Magazine
article. 19 The robots are shown in Fig. 8.1, and were constructed from an In-
tel 386 PC motherboard mounted on a radio-controlled toy tracked vehicle.
The robots had a miniature wide-angle video camera and framegrabber. The
flapper-style grippers had an IR to indicate when something was in the grip-
per. The robots also had a bump sensor in front for collisions. The robots
were painted fluorescent green.