In several areas of speech research, articulatory models able to produce a wide variety of speech sounds, not specific to any language, are needed as a starting point. Such research fields include the studies of sound system emergence in populations, infant speech acquisition research, and speech inversion research. Here we approach the problem of exploring the possible acoustic outcomes of a dynamic articulatory model efficiently, and provide an entropy based measure for the diversity of the explored articulations. Our exploration algorithm incrementally clusters produced babble into a number of target articulations, aiming to produce maximally interesting acoustic outcomes. Consonant gestures are defined as a subset of articulatory parameters and are thus superposed on vowel context, to provide a coarticulation effect. We show that the proposed algorithm explores the acoustic domain more efficiently than random target selection, and clusters the articulatory domain into a number of usable articulatory targets.