Coordination in Multi-Agent Systems

Coordination, as an art of managing
interdependency among activities, will be extensively
studied in this book under the multi-agent system paradigm.
To model the information essential to agent coordination,
this book proposes a Fuzzy Subjective Task Structure (FSTS)
model, through which agent coordination is viewed as a
Decision-Theoretic Planning problem, to which reinforcement
learning can be applied. Two learning algorithms,
"coarse-grained" and "fine-grained" are presented to address
agent coordination at two different levels. The
"coarse-grained" algorithm operates at one level and tackles
hard system constraints, while the "fine-grained" at another
level and for soft constraints. Besides reinforcement
learning, this book also proposes a bio-inspired approach to
agent coordination. A dynamic coordination model inspired by
biological metabolic system is presented. Agent coordination
is achieved as every agent performs iteratively a dynamic
optimization process, which utilizes explicitly the global
dynamics captured through the metabolic model. All research
results presented in this book are experimentally evaluated
to be effective and useful in practice.