ESCAPES - Evacuation Simulation with
Children, Authorities, Parents, Emotions, and Social comparison
Motivation
Conducting live exercises for evacuating thousands of people is generally impossible. A staged evacuation would necessarily miss crucial aspects of the required response (e.g. fear, confusion) on the part of both emergency personnel and evacuees, or the exercise would be considered unethical.
Description
Agent-level modeling is essential to creating realistic behaviors that a trainee needs to respond to in evacuation scenarios. We use BDI-style agents as has been done in existing literature, but augment this with features specific to an airport evacuation domain. Specifically, we include Children, Parents, and Authority Figures. In general, a number of features are salient for any evacuation simulation, such as realistic Spread of Knowledge, Emotions, and Social Comparison. Finally, we use Massive software to create compelling visualizations that will enhance training and policy-evaluation for evacuation authorities.
Children/Parents: The large presence of families makes airport domains distinctly different from other commonly evaluated scenarios such as office buildings or subway stations during rush hour, where the population is largely working-age men and women. The dynamics of a family such as seeking each other out and not escaping on their own and children dragging parents into interesting stores are extraordinarily different from the rest of the population, necessitating their inclusion in any accurate simulation of an airport evacuation.  |
ESCAPES simulation screenshot showing parent/child pair running against the crowd perhaps to seek out other family members prior to exiting or moving towards a suboptimal exit due to imperfect environment knowledge. |
Authority Figures: Authority figures are unusually pervasive in airport situations, unlike in subways, malls, or office buildings. Their reactions and directions have an enormous impact on the swiftness and safety of any evacuation. The fact that they have prior training, greater familiarity with the airport, and are expected to take leadership roles makes their inclusion as a unique agent-type an obvious feature to have.
Spread of Knowledge: The realistic spread of knowledge through a population is particularly important in an evacuation simulation where people may be first-time visitors unfamiliar with the domain. In ESCAPES, we model imperfect information about the environment (e.g., exit locations) as well as the need to evacuate, recreating real-life situations where people may proceed to suboptimal exits or linger for far longer than they should.
Social comparison: Social comparison is the well-documented phenomenon that people compare themselves with surrounding people and seek to minimize differences they perceive with individuals that are seen as similar. In the absence of knowledge about exits or the need to evacuate and in the presence of large crowds, social comparison becomes particularly relevant and our agents use this method to augment the individualistic decision-making that is often the sole mechanic in other evacuation simulations.
Emotions and Emotional Contagion:
In any evacuation situation, emotions are heightened and their impact on decision-making has been well studied in psychology.
We include models of individual emotions that incorporate their generation, dissemination to others, control by authority figures, as well as impact on decision-making and locomotion.
EMOTIONAL CONTAGION:
We have extended this module in subsequent work including a workshop paper at AAMAS 2011 and a full paper at IVA 2011. These works have looked at evaluating existing computational models of emotional contagion using sensitivity analyses as well as comparison with live video footage of crowd dispersion.
We analyzed three models (the ASCRIBE Model, the Durupinar Model, and the ESCAPES simulation environment) as well as numerous variations, like the effect of fear on speed (binary/graduated), exogenous decay of fear (including/excluding), and the proximity of effect (including/excluding). We were able to conclude that the Durupinar Model appears to consistently underperform the other models tested and that including the proximity of effect greatly increases the accuracy of each model. We show that the accuracy with which ESCAPES can reproduce a given scene varies depending on the emotional contagion model being used, motivating the need for a model that accurately describes the phenomenon but is simple enough that it only uses realistic data inputs for calibration and validation.
There are extremely few known video clips of crowd evacuations with which to evaluate emotional contagion in this context. The most usable one found was identified and used by researchers at VU University to perform a similar evaluation (including Tibor Bosse, Mark Hoogendoorn, Michel C. A. Klein, Jan Treur, and Natalie van der Wal). It occurred on May 4th in Amsterdam at the country's National Day celebration and can be seen in various YouTube clips such as this one. We also analyzed a video from a protest in Greece in which police fired tear gas at the base of the stairs and the crowd fans out as a result. The YouTube link for that video can be found here.
We intend to continue expanding our understanding of emotional contagion from a quantifiable, computational perspective to further improve the agent models of ESCAPES as well as other application domains. We hope to develop and validate an emotional contagion model using controlled experiments for inclusion in the ESCAPES system to better describe the spread of emotions in an evacuation. The experimental techniques and development of such a data-driven contagion model would have a major impact on our understanding of emotional contagion as well as of belief, social, and idea contagion.
Collaboration
In addition to a cross-institutional team with members from the University of Southern California, the University of the Pacific, and Bar-Ilan University in Israel, we have also established collaboration with Massive Software. Massive has been used in dozens of commercial films over the past few years, beginning with the creation of the Lord of the Rings trilogy's epic-scale battle scenes as well as more recent uses in such films as The Curious Case of Benjamin Button, The Dark Knight, X-men Origins: Wolverine, Ratatouille, John Woo's Red Cliff, to name but a few.
Publications:
Evaluating Computational Models of Emotional Contagion
To appear: Intelligent Virtual Agents, 2011.
Modeling Emotional Contagion
In AAMAS 2011 Workshop on Multi-Agent Based Simulations (MABS), 2011.
ESCAPES - Evacuation Simulation with Children, Authorities, Parents, Emotions, and Social comparison
In Proceedings of International Conference on Autonomous Agents and Multiagent Systems (AAMAS), 2011.
Agent-based Evacuation Modeling: Simulating the Los Angeles International Airport
In Proceedings of the Workshop on Emergency Management: Incident, Resource, and Supply Chain Management (EMWS09), Irvine, CA, November 2009.
