Multi-agent Communication Protocols with Emergent Behaviour
The emergent behaviour of a multiagent system depends on the component agents and how they interact. A critical part of interaction between agents is communication. This thesis presents a multi-agent system communication model for physical moving agents. The work presented in this thesis provides all the tools to create a physical multi-agent communication system. The model integrates different agent technologies at both the micro and macro level. The micro structure involves the architecture of the individual components in the system whilst the macro structure involves the interaction relationships between these individual components in the system. Regarding the micro structure of the system, the model provides the description of a novel hybrid BDI-Blackboard architectured agent that builds-in a hybrid of reactive and deliberative agent. The macro structure of the system, provided by this model, provides the operational specifications of the communication protocols. The thesis presents a theory of communication that integrates an animal intelligence technique together with a cognitive intelligence one. This results in a local co-ordination of movements, and global task coordination. Accordingly, agents are designed to communicate with other agents in order to coordinate their movements via a set of behavioural rules. These behavioural rules allow a simple directed flocking behaviour to emerge. A flocking algorithm is used because it satisfies a major objective, i.e. it has a real time response to local environmental changes and minimises the cost of path planning. A higher level communication mechanism is implemented for task distribution that is carried out via a blackboard conversation and ii negotiation process with a ground based controller. All the tasks are distributed as team tasks. A novel utilization of speech acts as communication utterances through a blackboard negotiation process is proposed. In order to implement the proposed communication model, a virtual environment is built that satisfies the realism of representing the agents, environment, and the sensors as well as representing the actions. The virtual environment used in the work is built as a semi-immersive full-scale environment and provides the visualisation tools required to test, modify, compare and evaluate different behaviours under different conditions. The visualization tools allow the user to visualize agents negotiations and interacting with them. The 3D visualisation and simulation tools allow the communication protocol to be tested and the emergent behaviour to be seen in an easy and understandable manner. The developed virtual environment can be used as a toolkit to test different communication protocols and different agent’s architecture in real time.
- PhD