D-20-5 MODELING HUMAN MOTIONS SUCCESSIVELY IN MULTIPLE LATENT SPACES