Abstract
M.Sc. (Information Technology)
The purpose of the research presented in this dissertation is to provide a method for
enabling dynamic scalability and control in multi-agent systems. Traditional multi-agent
systems are not easily scalable, this being especially true for open, heterogeneous systems.
Adding or removing an agent from a multi-agent environment causes the perception of the
environment to change, with regard to the remaining agents in the environment. Agents are
required to update their plans and beliefs in order to compensate for the agent that has just
left or arrived.
The dissertation investigates multiple methods of achieving scalability in multi-agent
systems, such as the use of multi-agent platforms, organizational multi-agent systems and
meta-level control. The main focus, however, is drawn to the meta-level control approach
because of the flexibility in implementation and the level of abstraction it provides in the
multi-agent architecture.
The dissertation presents a Meta-level Controlled Multi-agent model (MCM) that is built on a
novel design based on the introduction of a centralized meta-level control layer in a multiagent
environment, capable of interacting with the entities in the environment, adapting to
changes in the environment, and most importantly, providing a mechanism for dynamic
scalability in the multi-agent environment. The model provides a blueprint for building multiagent
systems that are able to dynamically scale up and down without impeding the normal
operation of the system. Furthermore, the model allows for a modular architecture, such that
the control layer can be extended to address domain-specific issues without affecting other
modules. The model supports heterogeneity, since agents can be developed using multiple
languages or platforms, the only requirement being that every agent in the system needs to
subscribe to a standardized communication protocol.
The chosen prototype for the MCM model is an autonomous traffic system in a simulated
smart city environment, whereby vehicles are required to enter and leave the smart city
dynamically and without prior instruction. The prototype provides a good testbed to assess
the validity and success of the MCM model in achieving dynamic scalability, whilst ensuring
that the performance of the system, namely the average travel time (aTT) of the system, is
not negatively affected.