Modelling Physical Heterogeneity of Agents by Means of Correlated Distributions

Agent heterogeneity is a concept that can significantly influence the dynamics of simulated evacuations. This study investigates how the total evacuation time (TET) is influenced by sampling the agent free-flow velocity and diameter from a correlated bivariate Gaussian distribution. Using the Pathfinder simulator, we model a corridor geometry with a bottleneck in the middle, testing three bottleneck widths, two corridor lengths, and both unidirectional and bidirectional flow. Two ways of distribution truncation are considered: symmetrical, with parameters truncated at the same quantile, and asymmetrical. In highly constrained scenarios with bidirectional flow, a strong negative correlation causes a “sorting effect”, where larger and slower agents accumulate at the back of the crowd, leading to uneven bottleneck load and a TET increase of up to 5%. In less constrained geometries, correlation has little impact, with the velocity of the slowest agent emerging as a strong predictor of TET. Asymmetrical truncation is shown to influence the edges of the marginal distributions, masking the effects of correlation on TET. The results are compared to alternative physical heterogeneity models, emphasising the importance of carefully calibrating parameter distributions, particularly in lower-density scenarios.