Development of torsional social force model for pedestrian dynamics in corner navigation: Insights from empirical observations

Abstract

Empirical observations show that both corridor width and departure position—defined by the distance from the inner corner—significantly affect pedestrian turning dynamics, such as turning position, speed, and angular speed. However, these factors are often overlooked in existing models. To address this gap, we developed an improved Torsional Social Force Model (TSFM) based on controlled experiments. Experiments involved individual and group navigation in angled corridor with widths of 1.5, 2.0, 2.5, and 3.0 m, with participants departing from various positions. Data from individual navigation in a 2.0 m corridor were used for model development, while data from other widths and group navigation validated the model. The TSFM represents pedestrians as spherocylinders, where movement is driven by force components and turning by torque components. Unlike traditional models with fixed turning axes and constant angular speeds, the TSFM integrates spatially variable turning start and end positions, angular speed and speed profiles, and turning axis distributions into its forces and torque calculations. Validation results show the TSFM significantly outperforms classic social force models (SFMs). It reduces the trajectory mean squared error from 0.45 m² to 0.03 m², and decreases travel time errors from >15 % to <5 %. Averaged speed error decreases from 25.9 % to 8.0 %, and angular speed error from 11.2 % to 4.1 %. The TSFM demonstrates superior accuracy in capturing turning dynamics during corner navigation.