True Pareto Fronts for Multi-objective AI Planning Instances

Multi-objective AI planning suffers from a lack of benchmarks with known Pareto Fronts. A tunable benchmark generator is proposed, together with a specific solver that provably computes the true Pareto Front of the resulting instances. A wide range of Pareto Front shapes of various difficulty can be obtained by varying the parameters of the generator. The experimental performances of an actual implementation of the exact solver are demonstrated, and some large instances with remarkable Pareto Front shapes are proposed, that will hopefully become standard benchmarks of the AI planning domain.