Monolithic vs. hybrid controller for multi-objective Sim-to-Real learning