Affective Body Movements (for Robots) Across Cultures

Abstract

Humans are very good in expressing and interpreting emotions from a variety of different sources like voice, facial expression, or body movements. In this chapter, we concentrate on body movements and show that those are not only a source of affective information but might also have a different interpretation in different cultures. To cope with these multiple viewpoints in generating and interpreting body movements in robots, we suggest a methodological approach that takes the cultural background of the developer and the user into account during the development process. We exemplify this approach with a study on creating an affective knocking movement for a humanoid robot and give details about a co-creation experiment for collecting a cross-cultural database on affective body movements and about the probabilistic model derived from this data.

Appendix Recognition (Recall), Accuracy, and Precision

For calculating the three measures, the following notions are needed: true positive (tp), false positive (fp), true negative (tn), false negative (fn). The color codes in Table 8.5 exemplify to which part of the confusion matrix they refer to if the target emotion is anger. The following codes are used: tp (green), fp (yellow), tn (grey), fn (red).

Table 8.5 Confusion matrix summarizing results for all cultures (condition 1)

Table 8.6 Confusion matrix summarizing results for all cultures (condition 2)

Table 8.7 Confusion matrix for Danish culture (condition 1)

Table 8.8 Confusion matrix for Danish culture (condition 2)

Table 8.9 Confusion matrix for German culture (condition 1)

Table 8.10 Confusion matrix for German culture (condition 2)

Table 8.11 Confusion matrix for Japanese culture (condition 1)

Table 8.12 Confusion matrix for Japanese culture (condition 2)

$${\text{recognition}}\,({\text{recall}}) = \frac{tp}{tp + fn}$$

$${\text{accuracy}} = \frac{tp + tn}{all}$$

$${\text{precision}} = \frac{tp}{tp + fp}$$