Time-to-Collision Based Social Force Model for Intelligent Agents on Shared Public Spaces

Abstract

Intelligent transportation modes such as autonomous robots and electric scooters with ride assistance are gaining popularity, but their integration into public spaces poses challenges to pedestrian safety and comfort. Nevertheless, the attempts to address the problem are scattered and sometimes contradictory. Models describing the behavior of heterogeneous crowds are necessary for solution evaluation before implementation. Moreover, autonomous agents benefit from these models, aiming to operate more efficiently while prioritizing pedestrian safety. The novelty of the proposed model is integrating time-to-collision, an indicator of road users’ subjective safety, into the social force model, the primary tool for pedestrian movement predictions. Moreover, the model considers the cumulative effects of anticipating other agents’ trajectories and the incurred time-to-collisions within a specific time horizon. We conduct controlled experiments using electric scooters to calibrate the model, discuss the distribution of parameter sets, and present pooled parameter population properties. Furthermore, we validate the model’s performance for electric scooters in complex scenarios and compare it with previous models using behavior naturalness metrics. Lastly, we compare the model’s accuracy and computation resource intensity to existing models. The model is computationally cheap and better equipped to estimate nearby people’s comfort level, making it a better candidate for intelligent agents’ path-planning algorithms in shared spaces.