Extending agent languages for multiagent domains

One of the most widely studied agent models is based on the notions of beliefs, desires and intentions (or BDI) as mental attitudes that guide the selection of courses of actions. However, BDI agent languages have been used mostly in the context of single agents based on a plan library of behaviours invoked reactively and, though they provide a theoretically sound basis for agent development, they offer limited support for multiagent systems with dynamic plan libraries.

In particular, when new plans not foreseeable at initial design time are required, the agent must be redesigned. Moreover, when designing multiagent systems, agent languages provide at most a communication language with no other consideration of interaction.

This thesis aims to address these limitations by introducing a new agent language and architecture that includes a mechanism for processing goals in a manner that decouples goal achievement from plan execution, as well as generating new plans to cope with unforeseen situations at design time. It bridges the gap between agent languages and multiagent systems by introducing a simple cooperation mechanism together with a norm processing
mechanism aimed to providing some degree of societal control.