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

Dhirendra Singh, Lin PadghamIJCAI 2017 · DOI

TL;DR

This paper introduces the Emergency Evacuation Simulator (EES), a tool that integrates Belief-Desire-Intention (BDI) cognitive agents with agent-based models (ABM) to simulate complex community evacuations. EES empowers emergency services to plan and understand evacuation scenarios more effectively, moving beyond simple rule-based simulations to incorporate more realistic human decision-making.

What Problem Does This Paper Address?

Planning effective community evacuations, especially in the face of natural disasters like bushfires, is a complex challenge. Traditional simulation tools often rely on simplified agent behaviors, failing to capture the nuanced cognitive processes that drive human decision-making during emergencies. This can lead to unrealistic simulations and suboptimal evacuation plans. The authors address this by developing an infrastructure that bridges the gap between sophisticated BDI cognitive models, which represent agent intentions and beliefs, and ABM systems, which simulate the physical environment and agent actions. This integration allows for more realistic modeling of how people perceive threats, form intentions, and act during evacuations.

Methods

The core of the EES is a BDI-ABM integration framework. This framework allows for the "brains" of some agents (their cognitive reasoning) to reside in a BDI system (like JACK or Jadex), while their "bodies" (their presence and actions in the simulation) are managed by an ABM (like MATSim). The architecture is structured into three layers:

  • Generic Layer: Handles the transparent packaging and transfer of information (percepts and actions) between the ABM and BDI systems at each simulation timestep.
  • System Layer: Provides the specific code to link a particular BDI system or ABM system into the generic layer. This involves functions for managing durative actions, distributing percepts, and detecting the end of BDI reasoning cycles.
  • Application Layer: Contains the domain-specific code to create an executable program, in this case, for evacuation scenarios.

The paper highlights the development of an industry-grade web-based interface for EES, which significantly improves usability by allowing users to visually construct and run evacuation scenarios quickly, often within minutes.

Datasets

Not discussed in the paper. The paper focuses on the simulation tool itself and its application to planning, rather than on training or evaluating against specific datasets of past evacuations.

Key Insights and Ideas

  • Bridging Cognitive and Physical Simulation: The most significant insight is the successful integration of BDI cognitive models with ABM. This allows for simulations that go beyond simple stimulus-response, incorporating elements of planning, goal-driven behavior, and intention formation, which are crucial for realistic human behavior modeling.
  • Enhanced Usability for Emergency Services: The development of a web-based interface dramatically improves the accessibility and utility of the simulation tool for emergency planners. This allows for rapid scenario construction and iteration, a critical factor in operational planning.
  • Understanding Complex Dynamics: By simulating realistic agent behavior, EES can provide valuable insights into likely and worst-case evacuation scenarios, helping authorities understand egress times, traffic conditions, and the impact of various planning decisions.
  • Iterative Refinement: The iterative development workflow, facilitated by the web interface, allows for quick feedback loops between simulation developers and emergency agencies, leading to more refined and effective planning tools.

Key Results

The paper presents EES as a demonstration tool, highlighting its capability to:

  • Visually construct evacuation scenarios, including defining fire fronts and their progression.
  • Specify evacuation parameters such as alert timing, vehicle departure patterns, and safe egress routes.
  • Run simulations that provide visual feedback on evacuation progress, including vehicle speeds and traffic congestion (as shown in Figure 1).
  • Analyze egress times to understand potential evacuation bottlenecks and risks.

The authors emphasize the improved usability and speed of obtaining simulation results due to the web-based interface, enabling quick iterations and analysis.

Limitations and Future Work

  • Limited to Australia: The tool has primarily been trialled in Australia, and the authors express a desire to gather insights and suggestions from an international audience.
  • Need for Validation: While the paper mentions informal validation by domain experts, more formal validation against real-world evacuation data would strengthen the tool's credibility.
  • Future Work: The authors suggest that the underlying BDI-ABM framework is general-purpose and can be applied to other domains requiring cognitive reasoning in simulations, such as parental decision-making around vaccinations or reverse-auction incentive schemes. The focus for EES itself would likely be on further refinement and broader adoption by emergency services.

Summary updated

BibTeX
@inproceedings{singh2017emergency,
  author = {Dhirendra Singh and Lin Padgham},
  title = {{Emergency Evacuation Simulator (EES) - a Tool for Planning Community Evacuations in Australia}},
  booktitle = {International Joint Conference on Artificial Intelligence},
  year = {2017},
  doi = {10.24963/ijcai.2017/780},
}
Figure