While service robots in exhibition spaces are evolving into social agents that mediate visitor experiences beyond mere information delivery, the service paradox remains a persistent challenge where unsolicited proactive interventions diminish visitor satisfaction. This study proposes the concept of psychological boundaries as a regulatory mechanism for visitors' openness to robotic intervention and develops a boundary-aware interaction policy. Empirical analysis of 164 visitors revealed that service preferences are structured into three latent dimensions: proactive support, adaptive responsiveness, and autonomy-oriented minimal service. Furthermore, the study identifies that hedonic and epistemic visit motivations are key determinants of the permeability of these psychological boundaries. Based on these findings, we established a cost-sensitive decision model optimized for the asymmetric social costs of misclassification. The results demonstrate the efficacy of this policy in accurately classifying visitor types at an early stage and selecting appropriate service intensities with high accuracy. This research suggests that robotic intelligence resides not merely in increasing service quantity but in situational restraint, which involves determining when to withhold intervention to respect visitor autonomy, thereby providing a novel framework for the design of non-intrusive service robots.