Situational preferences for BDI plans

In BDI programs it is quite common to find context conditions of plans which are over-constrained in order to ensure that the most preferred plan is selected for use. This is undesirable for at least two reasons. It makes the plan not available for use at all in situations where it could be of value as a back-up plan, and also it requires incorporation of information that conceptually belongs with the preferred plan. The ability to specify directly in a plan specification, aspects of the situation which would make the plan more or less desirable, enables a dynamically calculated preference ordering which removes the need to over-constrain applicability to obtain the desired plan selection. This paper addresses the issue of dynamically assigning a value to a plan instance, based on the current state and the particulars of the plan instance under consideration. The framework uses specifications based on logical formulae which are evaluated dynamically, using the current state and variable bindings provided via the plan’s context condition. These provide a simple mechanism for locally specifying the value of plan instances. This can be regarded as providing a degree of applicability for a plan, rather than simply a boolean value.