A modified social force model considering collision avoidance based on empirical studies

Abstract Avoiding collisions between pedestrians is not based solely on geometric approaches, but also involves human social conventions. Previous collision avoidance models on pedestrians often overlooked the significance of personal space and intrusion variations of intruders such as intrusion angles, intrusion extents and danger levels. The avoidance behavior of pedestrians is affected by the relative position, movement direction and distance from their initial position to the path intersection point with the intruders. To build and calibrate a pedestrian avoidance model, virtual reality and realistic experiments with dynamic and static intruders were conducted under different conditions. The critical avoidance boundary, avoidance process function and probability of avoidance side are analyzed from the experiments. Through a comparative analysis, the differences between personal and geometric space required for avoidance were identified. Moreover, an avoidance model that calculates the steering angle based on the kinematic constraints and relative position of intruders is incorporated into the social force model in this study. It successfully replicates pedestrian avoidance behavior when faced with both static and dynamic intruders, and offers a valuable tool for addressing complex pedestrian movements in highly competitive spatial environments.