Evolving agent behavior in multiobjective domains using fitness-based shaping