Robot Conflict Expression Influences Human Courage Self-Evaluations

Researchers from The University of Osaka, Kyoto Tachibana University, and Otemon Gakuin University have developed a controlled experimental paradigm to investigate how observing a robot express internal motivational conflict affects human self-evaluations of personal courage. Published in Frontiers in Robotics and AI, the study utilized a cylindrical robot to display approach-avoidance conflict through projected speech bubbles, allowing the team to systematically manipulate and observe pre-action motives that are typically difficult to assess in human models. The research comprised two distinct studies designed to first establish whether a robot expressing conflict is perceived as courageous, and second, to examine how such expressions influence observers’ own perceptions of their personal courage based on their preexisting tendencies.

The central premise of the work relies on the concept that courage is not the absence of fear, but rather the pursuit of a valued action despite fear, hesitation, or competing motives. In human interactions, the internal conflict preceding a courageous act is covert and difficult to manipulate systematically. By using a robot, the researchers could hold the agent’s appearance and overt behavior constant while explicitly displaying attributed motives through visual and verbal cues. This methodological approach allows for a reproducible examination of how observers interpret pre-action difficulty and how that interpretation relates to their own self-assessments of courage.

Study 1 focused on determining whether a robot displaying approach-avoidance conflict would be perceived as more courageous than a robot not displaying such conflict. The team employed a within-participant design involving 131 participants, aged 18 to 29, who viewed video stimuli of the robot in a scenario where it observed a person littering in a park. The robot’s motives were presented in speech bubbles projected near its head. In the conflict conditions, the robot displayed both approach motives, such as the park becoming cleaner, and avoidance motives, such as the risk of being yelled at. In the no-conflict conditions, only approach motives were presented. All conditions concluded with the robot performing the admonishing action, saying, “Excuse me, this is not a place to throw away trash.”

The results of Study 1 indicated that the robot in the conflict condition received significantly higher courage ratings than the robot in the no-conflict condition. The main effect of conflict was statistically significant, with a partial eta squared of 0.086. This finding supports the hypothesis that displaying pre-action conflict, even when the final behavior is identical, leads observers to interpret the behavior as more courageous. Additionally, conflict ratings, which measured the extent to which the robot appeared to waver between motives, were also higher in the conflict condition. The study also compared sequential and simultaneous presentation methods for the motives. While simultaneous presentation yielded higher conflict ratings, it did not significantly affect courage ratings, suggesting that the perception of courage was driven by the presence of conflict rather than the specific temporal format of its display.

Study 2 expanded the investigation to examine how these robot expressions influenced observers’ post-stimulus self-evaluations of their own personal courage. The researchers hypothesized a three-way interaction among preexisting courage tendency, conflict expression, and the robot’s final action. They predicted that participants with a low preexisting courage tendency would show the highest self-evaluations when observing the robot express conflict and then take action. This hypothesis was grounded in observational learning theories, which suggest that seeing a model cope with difficulty can enhance an observer’s sense of self-efficacy and feasibility of action.

The study involved 126 participants, also aged 18 to 29, who were classified into low-courage and high-courage groups based on their pre-stimulus scores on the Japanese version of the Courage Measure (CM-J). The cutoff for classification was a score of 4 on the 7-point scale, with scores below 4 designated as the low-courage group. Participants viewed four video conditions that crossed conflict (present or absent) with action (admonition or no admonition). In the no-conflict conditions, the motives were congruent with the final behavior; approach motives preceded action, while avoidance motives preceded non-action. In the conflict conditions, both approach and avoidance motives were presented simultaneously, regardless of the final action.

Contrary to the primary hypothesis, the predicted three-way interaction was not supported by the data. The main effect of conflict on self-evaluations was not significant, nor was the main effect of action. However, the researchers observed a significant two-way interaction between preexisting courage tendency group and conflict expression. This interaction was not specified in the primary hypothesis but emerged as a secondary finding. In the low-courage group, self-evaluation scores tended to be higher in the conflict condition, with a mean difference of 0.122. Conversely, in the high-courage group, self-evaluation scores tended to be lower in the conflict condition, with a mean difference of -0.083.

It is crucial to note that neither of these within-group differences reached statistical significance after correction for multiple comparisons, with both p-values at 0.064 following Holm correction. Therefore, the researchers characterize this contrasting pattern as an exploratory finding rather than a definitive causal effect. The data suggest that robot-expressed conflict may be received differently depending on the observer’s preexisting courage tendency, but this interpretation requires confirmatory testing in future studies. The lack of significance in the primary three-way interaction indicates that the specific combination of conflict and action did not produce the expected differential effect on self-evaluations across the groups.

The study provides a controlled Human-Robot Interaction (HRI) paradigm for examining how robot-expressed motives are perceived. By using speech bubbles to represent internal states, the team extended previous methods that used speech bubbles to represent attention or mental focus. This approach allows for the manipulation of motivational structures in a way that is difficult to achieve with human models, where internal states are not directly observable or controllable. The manipulation checks in both studies confirmed that participants perceived the intended conflict expression, with conflict ratings significantly higher in the conflict conditions. This validates the use of the robot as a medium for presenting representations of pre-action motives under controlled conditions.

Several limitations constrain the interpretation of these findings. First, the study measured self-evaluations of courage but did not measure actual courageous behavior. The CM-J is a trait-level measure, and its sensitivity to state changes following brief stimuli has not been established. Therefore, the results should be interpreted as condition differences in post-stimulus self-evaluations, not as changes in actual behavior or durable trait changes. Second, the design of Study 2 involved a confounding between motive valence and final action in the no-conflict conditions, meaning the factors were not entirely orthogonal. Third, the study used only video stimuli of a robot and did not compare other presentation media, such as physical robots, humans, or avatars. Thus, it is not possible to determine whether the results are unique to robots or attributable to the explicit presentation of conflict through speech bubbles.

The scenario used in the study was limited to a robot admonishing a person littering in a park. While this scenario involves both valued action and interpersonal risk, it remains unclear whether the findings generalize to other types of courageous acts, such as speaking up in interpersonal contexts or seeking help. The researchers acknowledge that the stimuli did not reproduce the gradual coping sequence often found in the observational learning literature, and the outcome was self-evaluation rather than task-specific self-efficacy. These factors limit the ability to draw direct conclusions about the mechanisms of observational learning in the context of courage.

Future research should focus on clarifying the psychological processes underlying the observed interaction between preexisting courage tendency and conflict expression. The team suggests measuring perceived similarity to the robot, perceived action feasibility, and self-efficacy to better understand how observers process the robot’s motives. Additionally, future studies should compare robots with other presentation media to isolate the role of embodiment and co-presence. Applied research should also examine how personalized expressions, adjusted for user characteristics, might influence actual behavior in contexts such as education, counseling, or everyday prosocial behavior. The current study provides a foundational step toward understanding how robots can function as social models by representing otherwise covert pre-action motives.

Subject of Research: The impact of robot-expressed approach-avoidance conflict on human self-evaluations of personal courage

Article Title: Effects of observing a robot expressing approach-avoidance conflict on observers’ self-evaluations of personal courage

Article References: Shimizu, Y., Ban, M., Takahashi, H., & Ishiguro, H. (2026). Effects of observing a robot expressing approach-avoidance conflict on observers’ self-evaluations of personal courage. Frontiers in Robotics and AI, 13, Article 1988670. https://doi.org/10.3389/frobt.2026.1988670

Image Credits: AI Generated

DOI: 10.3389/frobt.2026.1988670

Keywords: human-robot interaction, personal courage, approach-avoidance conflict, observational learning, robot expression, psychology, self-efficacy, social modeling, Effects, observing, robot, expressing