Revisiting the Theoretical Basis of Agent-Based Models for Pedestrian Dynamics

Robust agent-based models for pedestrian dynamics, which can predict the motion of pedestrians in various situations without specific adjustment of the model or its parameters, are highly desirable. But the modeller’s task is challenging, in part because it mingles different types of processes (cognitive and mechanical ones) and different levels of description (global path planning and local navigation). We argue that the articulations between these processers or levels are not given sufficient attention in many current modelling frameworks and that this deficiency hampers the effectiveness of these models. Conversely, if a decision-making layer and a mechanical one are adequately distinguished, the former controlling the desired velocity that enters the latter, and if local navigation is not guided solely by intermediate way-points towards the target, but by broader spatial information (e.g., a floor field), the greater robustness can be achieved. This is illustrated with the ANDA model, recently proposed based on such considerations, which was found to reproduce a remarkably wide range of crowd scenarios with a single set of intrinsic parameters.