Using LTLf/LDLf Synthesis Algorithms to Handle Goal Change in Autonomous Agents -- Implementation and Experiments

Abstract

In this research, we propose a formal definition for an intention management system (IMS) which an agent can use within a FOND domain for managing its intentions. These intentions can be expressed as any LTLf expression, in which temporally extended goals can also be provided to the IMS. The IMS provides an agent with several query- and update operations, by which an agent is given as much freedom as possible for reaching and adjusting its intentions throughout a run, all while maintaining consistency among this changing set of intentions. The definition of this IMS is used for developing a proof of concept in the form of software, which has been evaluated on several problem scenarios within FOND domains, expressed in LTLf. The IMS has proven to be effective for an agent in order to manage its dynamically changing intentions, complying to all requirements as described in the definition of the IMS. The definition of an IMS is meant as a proof of concept, proving its effectiveness in practice. This definition can be taken as a foundation for further research, where it could also be applied to LDLf expressions, handling beliefs and desires, and several other possible extensions on our research.