The study of how infants strapped in a Jolly Jumper learn to bounce can help clarify how they explore different ways of exploiting the dynamics of their movements. In this paper, we describe and discuss a set of preliminary experiments performed with a bouncing humanoid robot and aimed at instantiating a few computational principles thought to underlie the development of motor skills. Our experiments show that a suitable choice of the coupling constants between hip, knee, and ankle joints, as well as of the strength of the sensory feedback, induces a reduction of movement variability, and leads to an increase in bouncing amplitude and movement stability. This result is attributed to the synergy between neural and body-environment dynamics.