Complex Heterogeneous Crowding Phenomena: Multi-Agent Modeling, Simulation, Empirical Evidences and the Case of Elderly Pedestrians

The simulation of complex systems is nowadays one of the major applications of the multi-agent paradigm and it is widely applied in many fields. In the current scenario of global urbanization, the research for the development of intelligent transportation systems has gain much interests in the last decades, and it has produced relevant efforts on the simulation field as well. Collective forms of transport are one of the most sensible solutions for mitigating traffic congestion and pollution, and they imply a growing importance of pedestrians in the planning and design activities. The Thesis focusses on these activities, proposing discrete approaches for the microscopic simulation of pedestrian traffic with several innovative components. The literature of pedestrian modeling, presented in Chapter 2, can be classified with a three layers scheme that defines the behavioral levels of pedestrian choices. Most of the literature is aimed at the simulation of the so-called "operational" level that describes, in the case of pedestrian, the pure walking behavior from sources to destinations. At this bottom-line level, in fact, the consistent empirical knowledge on the physical properties of the pedestrian traffic allows a sufficient validation of the simulation models, which are then usable for a dynamical analysis of pedestrian flows in arbitrary environments. Nonetheless, the consideration of only the operational level lacks the route choice activity of pedestrians, which resides at the tactical level and that becomes fundamental if the simulation scenario represents a complex environment, with also different possible intermediate points (e.g. ticket windows). Available knowledge on pedestrian wayfinding is scarce, but designing simulation models for this activity is a way of defining requirements for further studies on this topic. Chapter 3 discusses the first microscopic model proposed for pedestrian traffic. The model proposes a hybrid agents architecture describing two synchronized and communicating components to deal with both tactical and operational level of pedestrian behavior. At the lowest level the model extends the well-known floor field model, proposing innovative extensions as the consideration of pedestrian groups and an approach to manage different speeds of people. The component dedicated to the route choice of the agents describes an adaptive behavior aimed at minimizing the individual traveling time towards the final destination, considering the static configuration of the environment and the perceived state of the system. Chapter 4 presents the results of the simulation model by firstly presenting quantitative results on the operational level, which are compared with state-of-art empirical data. Experiments about the tactical level explore the overall behavior of the simulated pedestrians in presence of different paths towards a destination. Chapter 5 analyzes a slightly different microscopic model defined for the walking behavior of pedestrian, whose aim is an integration with the MATSim simulation system, based on a queue model and mainly used for vehicular traffic simulation in metropolitan areas. The definition of this model finds its purposes in the approach to manage the route choice inherited from MATSim, which follows an iterative learning algorithm implying numerous simulations of the same scenario. Starting from a systematic choice of the shortest path at the first iteration, the agents are able to adjust their routes, for the following iteration, based on the traveling time experienced at the current iteration. The simulation campaign can converge to an equilibrium, of different type (Nash equilibrium or system optimum) according to the perceived travel time. This alternative microscopic model for pedestrian traffic is very simple and optimized, yet is able to be perfectly calibrated to fit the empirical data for the validation, as it will be shown in the results section.