Stability maps are a novel approach of teaching dynamic motor skills to humanoid robots through imitation. Essentially, the algorithm calculates a low-dimensional map of a shown motion storing the stability information of each pose. Additionally, variations of such are calculated and integrated into the map as well. In a physics simulation the stability of shown and calculated poses included in the map is determined resulting in a space with stable und unstable postures. Since the trained motion can be described as a trajectory within this low-dimensional space it can be transformed with the result that only stable poses are include in the final stabilized robot motion.