Emergency Evacuation Simulator (EES) - a Tool for Planning Community Evacuations in Australia

1 Introduction

For the past few years we have been working with emergency services in Australia on a simulation tool-Emergency Evacuation Simulator (EES)-that can be used to build community evacuation scenarios to aid understanding and planning. Figure 1 shows an example view of an evacuation produced by our tool. The large red patch shows a progressing fire front to the northwest of town. Evacuating vehicles are colour-coded to indicate speeds-from green (free flowing) through to red (slow) giving instant visual feedback on traffic conditions. Together with analysis of egress times the tool can provide valuable understanding of likely and worst case scenarios.

The development work-flow has typically involved an initial meeting with agencies to develop the scenario details, coding the scenario, and then meeting again to discuss results and refine further in an iterative way. However, naturally this process was dependent on availability of resources and personnel, and a single iteration could take weeks. More recently we have built an industry grade web-based interface for the simulator, that allows working groups to visually construct an evacuation scenario and obtain simulation results very quickly-within minutes. This has greatly aided usability.

2 BDI-ABM Integration Framework

Technically, EES uses our general-purpose BDI-ABM infrastructure [Singh et al., 2016] that allows agent-based simulations to be built where conceptually the "brains" of some agents exist in a Belief-Desire-Intention (BDI) reasoning system, e.g., JACK [Busetta et al., 1998], while their "bodies" live in the agent-based model (ABM), e.g., MATSim [Horni et al., 2016] traffic simulator (Figure 2). We use BDI [Rao and Georgeff, 1991] for cognitive modelling as it provides an established and robust computational model, captures richness of behaviour in a compact visually-presentable hierarchical representation of goals and plans, and allows behaviours to be readily understood and (informally) validated by domain experts. The setup still conforms to the standard paradigm of situated agents that continuously perceive changes in the environment, and perform actions that cause further changes to it.

Architecturally, the framework consists of three distinct layers. A generic layer realises the conceptual model shown in Figure 2 and provides the functionality for packaging information about actions and percepts passed between the ABM and BDI systems at each time-step. The system layer provides the code necessary for linking a particular BDI system or ABM system into the generic layer. Here, the BDI System Integration has three main functions, to (a) provide a transparent programming model for durative actions, (b) collect actions from, and distribute percepts to, individual agents (c) detect the end of a BDI reasoning cycle. The ABM System Integration on the other hand serves to (a) update actions status', (b) perform an ABM time step, and (c) distribute actions to, and collect percepts from, individual agents. Finally, the application layer provides the domain specific code to form an executable program.

So far we have developed BDI integrations for JACK [Busetta et al., 1998], Jadex [Braubach et al., 2005], and GORITE [Jarvis et al., 2013], and ABM integrations for Repast [North et al., 2007], GAMS [GAMS Development Corporation, 2015], MATSim [Padgham and Singh, 2016], and a Python-based simulator [Geard et al., 2013]. As well, we have presented example uses of the BDI-ABM framework for community evacuations [Padgham et al., 2011; Padgham et al., 2014; Singh and Padgham, 2015], parental decision-making around child vaccinations [Geard et al., 2013], and reverse-auction incentive schemes for conservation targets [Singh et al., 2016].

3 Simulating Community Evacuation Scenarios

In this demo, users will get hands-on experience in the use of the web-based EES tool for building community evacuation scenarios. We will provide a short written description of the emergency scenario. Users will then be asked to visually construct the scenario within EES. Specifically, they will:

• select a bushfire model within the tool and understand the likely impact of the progressing fire front on a town;
• choose sensible settings for the timing of the evacuation alert (as well as the peak in departure times of vehicles) with respect to the fire, based on their estimate of how long it would take to establish the severity of the fire and community members to respond;
• specify the number of vehicles, their locations, destinations they would evacuate to, and safe lines that should be monitored in the simulation for egress times; and
• run the simulation and interpret results.

The user sessions are expected to take 10-15 minutes each, and also offer opportunities for observing attendees to participate along with the user attempting the task. At the end, we will ask users for quick feedback on various aspects of the tool, including usability, and value of provided results and analysis from various perspectives. We hope to extract valuable suggestions and insights from an international audience, given that so far the tool has only been trialled in Australia.

Finally, we will have at hand information about the underlying technologies including the BDI-ABM framework, for attendees who may be interested in applying cognitive reasoning to simulations in their own work.

Acknowledgments

We thank Telematics Trust and Mount Alexander Shire Council for funding parts of this work, and Mount Alexander Shire Fire Management Committee for helping specify, test, and evaluate the web-based EES version.