Revised preprint of our work on Zeroth-order Deterministic Policy Gradient (ZDPG) just uploaded; this also contains a more detailed numerical results section.