Annotating Motion Primitives for Simplifying Action Search in Reinforcement Learning