The Effect of Robot Posture and Idle Motion on Spontaneous Emotional Contagion during Robot-Human Interactions

Humans tend to engage in spontaneous unconscious movements during social human-human interactions. When we are on the receiver side of the interaction, we tend to nod our head or mimic the changes in postures of the speaker. These gross behaviors are social indicators that we are engaging attention in the exchange. However, there are also micro changes in body sway, with slow shifts in body postures at frequencies that echo the emotional ambiance of the social exchange. Such idle motions have been defined as active states of communication [4] and can offer effortless collaborative situations.

Idle motions can influence the perception of a person’s emotional state and thus, modulate acceptance in social situations. They are related to the intensity of energy of the speaker with frequencies that are greater in high-energy emotions (anger and joy) than in low-energy emotions (sadness and contentment). Applying such idle movements to artificial agents may influence acceptance as human individuals perceive the robotic system as more animated and alive [5][1][9].

In the present study, we wanted to test whether manipulating the frequency of the idle motion would modify how human participants perceived a robot that was telling a sad story about its past life experiences. We controlled the frequency of the idle motion and measured the spontaneous body oscillations and shifts in body posture of the human participant while she/he was attentive to the negative content of the stories. We hypothesized that more emotion contagion would be observed in human participants when the robot oscillated at a low frequency as the expressed state of emotion was sad (low energy). We also hypothesized that a frizzing state of emotion, i.e., inhibition of spontaneous oscillations, should be observed in the human participants when the robot moved at high idle frequencies because such situations should be perceived as a threat from an angry robot (high energy). A total of three stories were used to give time for the emotional contagion to take place.

The motion capture data was recorded at a sampling frequency of 200 Hz. Marker velocity was extracted to compute 3D acceleration vectors. A filtering procedure was applied to extract all absolute acceleration peaks above 1000 $mm/s^2$; the remaining data series were interpolated before further analyses.

We computed the torso inclination as a function of idle frequency (high; medium; low) for each trial and participant using the shoulder markers.

A time-series spectral analysis was then calculated by applying wavelet transforms on the chest marker to extract body motion. As shown in Figure 2, we grouped the graphs of the power spectrum of each story (presented in rows) as a function of idle frequency (presented in columns), averaged across participants. The y-axis represents the period of the center marker in seconds as a function of the trial time, for the total story duration, in seconds. The average power spectral density (PSD) is represented with a color bar with brighter zones illustrating the highest power.

Results showed that participants looked more to the ground at the end than at the start of a trial (M=1.8 deg).

Participants did not engage in visible spontaneous torso oscillations with Buddy (see the absence of frequencies at 4, 8, and 12s in Figure 2). Nevertheless, the spectral analysis revealed a greater power density in the low idle conditions and an absence of movement in the high idle condition. This pattern of results was observed across all three stories, even if only stories 1 and 3 are presented in Figure 2. These findings suggest that participants tended to relax and engage in spontaneous oscillations when Buddy moved slowly; when idle was at high frequency, participants tended to freeze and inhibit spontaneous oscillations.

Figure 3 confirms that at high idle frequency, participants moved much less than for trials at low idle frequency. Nevertheless, in the high idle condition, movements at high frequency were more frequent than those at low frequency, confirming the presence of emotional contagion, even if to a lesser degree. This phenomenon resembles the posture changes reported when humans experience sadness or angry states during human-human interactions [2].

The Perceived Safety category of the GODSPEED was taken out as Buddy is small and harmless in appearance. For brevity, we only report the scores that obtained significant statistical differences, i.e., the Anthropomorphism and the Perceived intelligence categories (see Figure 4). Results are shown as a function of idle frequency (y-axis) and the mean score (x-axis).

An ANOVA on the GOODSPEED individual scores revealed a main idle effect for Anthropomorphism $(p<0.001)$ and Perceived Intelligence $(p<0.001)$ [1]. Participants perceived Buddy to be more anthropomorphic and more intelligent when it moved slowly (see Figure 4a). The analysis showed no effects of idle frequency on Likeability and Animacy. Such findings corroborate results reported previously in studies assessing emotional alignment during human-robot interactions [9].

In the not-so-far future, Robots will become more present in our environment. However, we must scrutinize how the interaction with these systems will influence us, humans [6]. In this study, we questioned whether the oscillatory movement of a robot’s head could modulate the spontaneous engagement of a human during face-to-face interaction. More specifically, we report how the idle motion of Buddy’s head can modulate emotional contagion and thus, impact a person’s affective state. In addition, our results suggest that Buddy is perceived as more anthropomorphous and intelligent when moving slowly. Future studies will need to test whether such findings are also found when using positive induction methods during human-robot interactions.