Alternatives to Threshold-Based Desire Selection in Bayesian BDI Agents

Bayesian BDI agents employ bayesian networks to represent uncertain knowledge within an agent’s beliefs. Although such models allow a richer belief representation, current models of bayesian BDI agents employ a rather limited strategy for desire selection, namely one based on threshold values on belief probability. Consequently, such an approach precludes an agent from selecting desires conditioned on beliefs with probabilities below a certain threshold, even if those desires could be achieved if they had been selected. To address this limitation, we develop three alternative approaches to desire selection under uncertainty. We show how these approaches allow an agent to sometimes select desires whose belief conditions have very low probabilities and discuss experimental scenarios