Microsimulation Approaches to Pedestrian Route Assignment Modelling



Microsimulation Approaches to Pedestrian Route Assignment Modelling

Authors

J Amos, B Kohn, V Zachariadis, Legion Limited, UK

Description

We propose an original approach to pedestrian route assignment, with real-life applications in mind. The movement network is defined using visibility attributes and the route costs are updated by a feedback-based self-learning method.

Abstract

Despite the increasing number of pedestrian simulation applications in the market, and the evident trend towards individual level models (agent-based, microsimulation, cellular automata), the modelled aspects of pedestrian behaviour are, almost exclusively, related to microscopic movement (i.e. obstacle avoidance, collision avoidance and crowd dynamics).

Strategic pedestrian movement behaviour and especially route choice and way-finding modelling are either approached in non-vigorous ways, like using simple static metrics (e.g. shortest distance paths) to evaluate alternative route options, or, in most cases, are ignored. In the later case, pedestrian simulation demands external user-defined route assignment inputs.

At Legion, over the last 20 months, we have been researching the behavioural assumptions and computational implications of several traffic assignment approaches and, based on the findings and their extensibility to pedestrians, we are developing solutions that will expand considerably the simulation capacities of our products.

There are two reasons that make pedestrian route assignment modelling especially challenging: the definition of the movement network (i.e. the choice-set) and the calculation of the estimated utility of each option.

Contrary to most transport systems, where route choices are generally constrained in predefined formal networks, pedestrians are free to move in continuous space (infinite number of trajectories). In order to model route assignment space needs to be discretised, either by using uniform grids (non-robust discretisation) or by using behavioural assumptions to formulate graphs representing the flow patterns that are expected to develop (robust discretisation). Given the size and complexity of many modelling cases and the accuracy levels that are needed, we argue that graph discretisation is the only appropriate method. In the paper we propose a vigorous and innovative method to extract, relatively simple - invariably of the configurational complexity of the modelled space, movement networks.

Traffic assignment methods have been, traditionally, based on aggregate utility and flow metrics. Although recent advances in microsimulation have made possible the validation of volume delay functions, the fact is that due to extensive field work, little improvement was needed. On the other hand, the kinetic characteristics of pedestrians (rapid acceleration and low turning radius), the configurational sensitivity of experienced costs and the multi-directionality of flow patterns make the formulation of aggregate cost functions highly unlikely. Therefore, accurate route assignment simulations demand disaggregate pedestrian movement approaches. We suggest that the computational price of performing pedestrian microsimulation means that formal traffic assignment iterative algorithms (i.e. Successive averages, Frank - Wolfe) prove impractical and may lead to behaviourally unrealistic distributions. We go on to describe a solution based on system-learning and real-time feedback (to simulate accumulated experience) and on-the-fly judgement (to simulate prevailing conditions).

The paper will discuss the aforementioned issues in depth, focusing on pedestrian route assignment and will follow the development process of the proposed approach. We will examine the advantages and disadvantages of the method and will present the outcomes of several tests. The scope of the application of route assignment will also be discussed in the light of the special behavioural characteristics of pedestrians. A section of the paper will focus on estimating experienced route costs based on dynamic, static and configurational attributes of alternative routes.

In general, we propose an original approach to route assignment, designed around the special characteristics of pedestrians and with real-life applications in mind. The innovative elements of the method are related to the two most challenging aspects of pedestrian route assignment: the definition of the movement network (i.e. the discretisation of space) where we employ a isovist-based, configuration-sensitive solution and the calculation and updating process of the dynamic elements of the route costs where we employ pedestrian microsimulation and a feedback-based self-learning method.

Publisher

Association for European Transport