Agent speed explains more tracking error variance than aggregation method in simulated crowd-sourced continuous control

Abstract Crowd-sourced control systems—in which distributed participants collectively steer a shared agent through real-time voting—are formalized in robotics as Multi-Operator Single-Robot (MOSR) teleoperation and also used on audience-participation platforms. Design effort has concentrated on how participant inputs are aggregated, while an equally consequential variable has received less direct attention: the speed at which the agent operates. Across 8,010 simulation runs, a two-way analysis of variance (ANOVA) on the 2,160-run factorial shows that agent speed accounts for 46–97% of tracking-error variance versus 2–29% for aggregation method—a ratio of 1.6–46 × (median 6.9 ×). This dominance holds at 5%, 20%, and 40% adversarial ratios and persists without reversal across 27 sensitivity configurations spanning the state-observation delay (0–867 ms), participant noise, smoothing, and disruptor behavior (11,340 additional runs); method differences at matched operating speeds are substantially smaller than speed effects. A trajectory-dependent optimal speed (2.0 m/s for circular trajectories at a 433 ms state-observation delay, Fixed policy) is invariant across the tested adversarial ratios, and crowd-size effects are benchmark-specific (Fixed policy, v = 5.0 m/s). A four-parameter multiplicative model captures the Fixed-policy circle response ( R 2 = 0.988; 4.4% held-out error). These results identify agent speed calibration as the higher-priority design consideration within the tested factorial.