Clustering Latent Sensor Distribution on Body Map . .. 5
Mori et al. provided a constructive model of a human fetus that forms behaviors
and understanding regarding its body through general movements in the womb of
its mother [8]. Sugiura et al. solved a similar problem by using the time difference
of arrival of signals and multidimensional scaling [9]. Kuniyoshi et al. generated
topographic somatosensory cluster map that is plotted the fetus’s sensor points on a
2 dimensional space as a result of a self-organizing process [10].
2.2 Body Schema
Even though the human body has multiple degrees of freedom, human beings can
control their bodies without considering the angle of joints and the length of links.
It is assumed that a body schema enables human beings to unconsciously control
their bodies. Humans have a multilink system in their brains for understanding their
bodies. Henry et al. define body schema as a unconscious process that dynamically
works when human beings control their motion and pose [11].
In a previous study of body schema, Martinez et al. estimated a multilink sys-
tem using an active-learning method based on vision information [12]. Nabeshima
et al. proposed a model that enables robots to extend their body schema, when robots
use tools, based on multimodal information. It is difficult to apply these methods to
our study which to estimate body map and body schema based on tactile information
because these methods require multimodal information including vision information.
3 Proposed Method
3.1 Overview
Figure 2 shows an approach of this study used to generate a body map and a body
schema. The proposed method consists of the following three steps. First, a topolog-
ical map is generated based on tactile information obtained from receptor cells on
the embodied system, as shown in Fig. 2a. Second, nodes of body schema are esti-
mated by clustering the generated topological map, as shown in Fig. 2b. The third
step entails generating a tree structure based on tactile information and the estimated
nodes of the body schema, as shown in Fig. 2c. As a preliminary stage to generate a
body schema, our approach generates a tree structure instead of a multilink system.
In this paper, we propose a method to generate a body map and a method to estimate
the nodes of the body schema, as shown in Fig. 2a and b, respectively.
我的内容管理
展开
