Fri 28 Nov
Phan Minh Dung
(Asian Institute of Technology, Bangkok)
Agent's Capability and the Correctness of Reactive Plans in a Dynamic
In recent years there is a shift in AI from the classical paradigm of
rational agents to the notion of reactive agents that have an
intelligent ongoing interaction with dynamic and uncertain
environments. The emphasis on an agent's interaction to its
environment is an important change from the traditional theories on
representation and control.
The correctness of an agent's behavior depends on its capability to
deliver timely responses to relevant changes in its environment. A
common view on capability in the literature is that an agent is said
to be capable to execute an action A if it is physically possible for
her to do it provided that the environment is kind enough not to
interfere. It is clear that this viewpoint cannot model the agent's
capability in an possibly uncooperative or hostile environment.
In this paper we will develop a branching time logic for specifying
agent's capability in a possibly uncooperative or hostile multiagent
environment. We then develop a method for verification of the
correctness of reactive plans. The semantics of our logic is
grounded in terms of states and actions of automata-like agents.