Movement Predictability Outweighs Robot Appearance in Fear-Relevant Responding During First Encounters Between Older Adults and Social Robots

<title>Abstract</title> <p>This study investigated whether robot embodiment, movement predictability, or postural configuration is the principal determinant of fear-relevant responding during first encounters between older adults and social robots. It addressed the research question: Which robot design characteristics most strongly influence subjective and physiological fear-relevant responses during initial human-robot interaction? A controlled within-subjects experiment was conducted with 40 community-dwelling older adults (65–73 years) in Doha, Qatar. Participants interacted with a humanoid robot (Pepper) and a non-humanoid robot (Temi) while movement predictability (smooth versus jerky) and postural configuration (open versus closed) were systematically manipulated. Fear-relevant responding was assessed using a purpose-developed 16-item Perceived Fear Scale together with baseline-corrected heart rate (ΔHR), skin conductance level (ΔSCL), and heart rate variability (ΔlnRMSSD). Data were analyzed using Holm-corrected linear mixed-effects models within a multimodal late-fusion framework. Robot embodiment did not produce significant standalone effects after correction for multiple comparisons. In contrast, movement predictability produced robust effects across all physiological outcomes (d = 1.05–2.94; all Holm-adjusted p &lt; .001), whereas postural configuration yielded smaller but consistent effects (d = 0.90–2.28). Subjective appraisal and physiological responses were weakly associated (all |r| ≤ .26), indicating complementary rather than redundant information. Behavioral predictability exerted a stronger influence on fear-relevant responding than anthropomorphic appearance under the experimental conditions investigated. These findings indicate that movement predictability was the strongest determinant of fear-related responses within the commercial robot platforms examined, whereas platform-level embodiment differences produced more limited and context-dependent effects. Because embodiment was operationalized through comparison of two commercially available robots rather than systematic manipulation of anthropomorphic realism, these findings should not be interpreted as a direct test of the Uncanny Valley hypothesis or generalized to all dimensions of robot appearance.</p>