Planning with Numeric Varibles in Multiobjective Planning

The purely propositional representation traditionally used to express AI planning problems is not adequate to express numeric variables when modeling real-world continuous resources. This paper presents a heuristic planning approach that uses a richer representation with capabilities for numeric variables, including durations of actions, and multiobjective optimisation. The approach consists of two stages. First, a spike construction process estimates the values of the variables associated with propositions/actions, without relaxing numeric effects in the calculus of the estimation. Second, a heuristic search process generates a relaxed plan according to the estimations of the first stage, and then performs search in a plan space. The relaxed plan and the heuristic estimations help the process find a plan while trying to optimise the multiobjective criterion.