Simulation-Guided Subset Aggregation for Large-Scale Tacton Similarity Ratings

Exploring perceptual dissimilarity spaces of large-scale Tactons (i.e., Tactile icons) can inform the design of distinguishable haptic feedback. Yet, collecting pairwise similarity ratings for entire Tacton sets becomes costly as set size increases, prompting the need for alternative methods like subset aggregation. Despite previous efforts, little systematic investigation exists on efficient subset size or participant number needed to estimate large-scale Tacton perceptual spaces within a bounded error threshold. We address this gap by introducing a model that simulates between-subject variability in similarity perception. The model explores various distributions under different conditions, including total Tacton numbers and subset-to-total ratios, to guide user studies. Guided by these simulations, we evaluated subset aggregation with three small-scale Tacton sets (12 or 14 patterns) and one large-scale set (48 patterns). Study 1 revealed that initial simulations underestimated real-world variability. We refined the model, ran simulations for larger-scale conditions, and validated them in subsequent studies. The updated model closely matched reality, showing that designers can use our subset aggregation method to prototype perceptual spaces for large-scale Tacton sets. Notably, 4–7 observations were sufficient to achieve $\rho\geq 0.6$, compared to the typical 12 required for generalization. We discuss the efficacy of subset aggregation and future research directions.