Reinforcement Learning to Adjust Parametrized Motor Primitives to New Situations