With the increasing deployment of service robots in urban public spaces, this study examines how robot proportion (RP) and composite interaction scenarios affect pedestrian circulation performance. An AnyLogic agent-based model of Quancheng Square, Jinan, was developed using a fixed total population of 10,000 agents. Three composite interaction scenarios (CIS-1–CIS-3) jointly vary interaction distance and duration. Human–robot mixed density (HRMD) and average pedestrian speed (APS) were evaluated over a common post-transient interval. To address population composition confounding and stochastic variability, an additional fixed-pedestrian-demand validation under CIS-3 used 10 independent random-seed runs per condition. Across the original runs, HRMD generally increased with RP, whereas APS showed a non-monotonic pattern. In the validation experiment, APS decreased monotonically as robot number increased when pedestrian demand was held constant, while the fixed-total design showed a high robot-level recovery. This contrast indicates that the apparent APS recovery was partly associated with reduced pedestrian demand. Overall, pedestrian circulation patterns were sensitive to both robot deployment and population composition.