Page 311 - Introduction to AI Robotics

294
Multi-agents
8
of robots working on a single task. Another motivation of multiple robots is
redundancy: if one robot fails or is destroyed, the other robots can continue
and complete the job, though perhaps not as quickly or as efficiently. Rodney
Brooks at MIT first proposed to NASA that teams of hundreds of inexpensive
ant-like reactive robots be sent to Mars in a technical report entitled “Fast,
Cheap and Out of Control” 30 in part because having many robots meant that
several robots could be destroyed in transit or during landing without a real
impact on the overall mission.
Multi-agent teams are becoming quite popular in robot competitions, espe-
cially two international robot soccer competitions: RoboCup and MIROSOT.
In these competitions, teams of real or simulated robots play soccer against
other teams. The soccer task explicitly requires multiple robots that must
cooperate with each other, yet react as individuals.
Readers with a strong background in artificial intelligence may notice sim-
ilarities between teams of mobile robots and teams of software agents (“we-
bots” which search the web and “knowbots” which do data mining). Those
similarities are not accidental; software and physical agents fall into a re-
DISTRIBUTED search area in Artificial Intelligence often referred to as Distributed Artificial
ARTIFICIAL Intelligence (DAI). Most of the issues in organizing teams of robots apply to
INTELLIGENCE (DAI) 10 119 26
software agents as well. Arkin, Bond and Gasser, Brooks, and Oliveira
et al. 113 all cite the problems with teams of multiple agents, condensed here
as:

Designing teams is hard. How does a designer recognize the characteristics
of a problem that make it suitable for multi-agents? How does the de-
signer (or the agents themselves) divide up the task? Are there any tools
to predict and verify the social behavior?
There is a “too many cooks spoil the broth” effect. Having more robots work-
ing on a task or in a team increases the possibility that individual robots
INTERFERENCE with unintentionally interfere with each other, lowering the overall pro-
ductivity.

It is hard for a team to recognize when it, or members, are unproductive. One
solution to the “too many cooks spoil the broth” problem is to try engi-
neering the team so that interference cannot happen. But this may not be
possible for every type of team or the vagaries of the open world may un-
dermine that engineering. To defend itself, the team should be capable of
monitoring itself to make sure it is productive. This in turn returns to the
issue of communication.