Drawing on the uncanny valley framework, trust theory, and similarity attraction theory, this study examines how customers’ multidimensional perceptions of humanoid service robots shape their approach and avoidance behaviors through two relational states: trust and rapport. Subsequently, structural equation modeling and mediation analysis were employed for testing. The results indicate that customers’ overall perceptions of service robots not only encourage approach behaviors but may simultaneously intensify avoidance tendencies, reflecting the ambivalent nature of human–robot interactions. We interpret this dual activation through the uncanny valley framework, in which humanlike robots simultaneously elicit attraction and aversion. Trust and rapport play critical mediating roles in this process, effectively reducing avoidance responses while strengthening customers’ approaches. Further analyses reveal that different perceptual dimensions operate through distinct mechanisms in the formation of trust and rapport. This study aims to deepen the comprehension of customer response mechanisms to humanoid service robots through a relational perspective, and offers practical insights for hotels seeking to balance operational efficiency with emotional experience in robot design and management.