To reduce burdens on obstetricians and pregnant women, we developed a robotic fetal ultrasonography system, which can safely and remotely perform ultrasound scans on pregnant women's abdomen. However, obstetricians must still control the robot in real time and learning the remote operation is time consuming. To solve this problem, the robot system must be improved to perform automatic ultrasound examination. A technical difficulty is that the robot must recognize the position and angle of the fetus in the uterus. In this study, we propose a fetal position estimation method based on existence probabilities of fetal body parts. The proposed method comprises 1) the classification of different fetal body parts based on the acquired ultrasound images and 2) the estimation of the appropriate position and angel of the probe to allow the robot to automatically collect the three-dimensional volume data of the fetus.