The mark test is a popular test for self-recognition. Although the extent to which self-recognition can be assessed remains controversial, the test elicits visually guided, self-oriented, and spontaneous reaching movements. In this study, we demonstrated that this self-oriented reaching is suitable for estimating sensorimotor body representation in toddlers. We developed a non-verbal task (Bodytoypo) to assess the localization of body parts by gamifying the mark test and conducted it with thirty 2- and 3-year old children. Specifically, we detected the children's skeletal data in real-time, displayed virtual marks on various parts of their body, and estimated their reaction time and accuracy of body part localization. Subsequently, a statistical-based automated analysis using 2-D image processing and conventional frame-by-frame coding were performed. The results revealed developmental changes in the children's reaching strategies. A few errors were observed around the face. A reduction in the error rate for joint and movable areas was observed in children between the ages of 2 and 3 years. An analysis of movement trajectories using a combination of image processing and machine learning algorithms showed that 2-year-olds acquired visually guided reaching (feedback control) from ballistic exploratory reaching and 3-year-olds acquired rapid and predictive reaching (feedforward control) from visually guided cautious reaching. It was also found that the accuracy of localization could be predicted by examining the coordination of body parts. Evaluation of the developmental changes in self-oriented reaching reveals new possibilities for the mark test and development of body representation.