Using CSP Meta Variables in AI Planning

ABSTRACT. Reformulating Artificial Intelligence Planning problems as Constraint Satisfaction Problems(CSPs) brings a number of benefits. In this reformulation process however, the structure of the original planning problem is lost. Therefore, no planning problem specific guidance may be used during the assignment of values to the CSP variables. Extending work, in which we implemented a planning-specific CSP variable and value selection heuristic, the work described here aims to make better use of the propagation inherent in a given CSP solver by using CSP meta variables. Such meta variables are used here for goal-locking and for the assignment of resources to tasks within the problem solution process, with the aim being to reduce the search space, and to better guide the search within it.