Disasters are inevitable, but we can always be more prepared for when they might strike. In addition to natural catastrophes, we see an increasing number of human-instigated disasters (e.g., fires, acts of terrorism) impacting our built infrastructure. Today, in order to estimate the time it takes to evacuate in an emergency situation, the most common methods of performing engineering safety analyses rely on engineering calculations and/or computational tools. These methods usually attempt to simulate actions people take and the time it takes to perform each action.
However, most evacuation models rely on oversimplified, inconsistent, and in some cases even inaccurate assumptions of human behaviour. Individual responses tend to differ drastically in disaster scenarios and this array of possibilities needs to be accounted for in the ways in which we prepare ourselves and design our buildings.
For example, research to date shows that people engage in a variety of other activities during these events (some try to fight the fire, some attempt to rescue others, some move towards unsafe conditions while others evacuate). Scientists also note that people might make poor judgements about extreme events or the impact of these events.
"There is little insight into how people behave when exposed to extreme conditions"
In addition, behaviour also depends on people’s cognition (e.g., risk perception), responsibilities (e.g., a parent with a child vs. an adult) and their physical conditions. There is lack of available data and theory about occupant behaviour under uncertainty. Most theories and models assume behaviour is similar under different conditions. There is little insight into how people behave when exposed to these extreme conditions: how behaviour is influenced by the building’s design, building type, a person’s past experiences, people around them, people they are responsible for and so on.
Yet, it is not trivial to measure these behaviours and their impact. Work to date typically has used traditional research methods, such as unannounced fire drills, post-surveys, video-recordings and so on. However, none of these methods are reliable as they do not trigger natural human response (trauma, panic, etc.), they lack realistic scenario features (fire, smoke, explosion) or provide opportunities for controlled experiments. On the other hand, we cannot expose people to unsafe conditions for moral and legal reasons. We therefore decided to study people's behaviour under extreme events using immersive virtual environments.
Header image credit Arif Gardner
Virtual reality disasters: Experiment design
To explore, we designed a virtual reality (VR) experiment that was run simultaneously at The Alan Turing Institute (where I am a Rutherford Visiting Fellow) and Tsinghua University in Beijing during the summer of 2018. We also collected data from subjects at the University of Southern California in Los Angeles during the autumn of 2018. We focused on two factors: visibility and crowd movement.
In the visibility condition, we manipulated the visibility of exit routes by replacing walls with glass; making stairs, ticket booths, and exits more visible; reducing the number of columns and partitions, and adding cues about exits (see Fig. 1). In the crowd condition, we manipulated how the crowd moves by modelling non-player characters (virtual avatars), who were assigned to different evacuation routes to examine if participants would follow or avoid the crowd (see – Fig. 2).
The virtual environment was modelled based on a real metro station with multiple evacuation routes and decision points. A train with a compartment on fire approached the metro station at the beginning of the experiment. In the VR environment, once the participants perceived the fire, they evacuated with the crowd around them.
Over 100 individuals participated in the experiment in London and in total, we collected data from over 300 people. The directional information provided by the crowd was critical in participants’ evacuation processes; participants who followed the crowd at the beginning would likely continue to follow the crowd during the successive decision points, indicating a consistency of the influence of crowd dynamics over the course of evacuation. The effects of crowd movement existed in all three locations (London, Los Angeles, Beijing), and the effect of location on evacuation behaviour was found to be insignificant.
Visual accessibility attracted people to go to more visible directions during their evacuation, especially the participants in London, potentially due to their regular use of the Underground. The more visibly a direction led to exits, the more likely it attracted people to take the direction. Furthermore, improving the level of visual accessibility persuaded people to choose visible routes over the ones taken by the crowd, although such effect may vary in different cultures (i.e., Chinese participants were found to have higher following tendency than the others). The results also revealed that increasing the level of visual accessibility improved people’s evacuation performance during emergencies (i.e., shortened the evacuation distance and time and increased the speed).
Watch one person's actions whilst participating in the VR experiment:
The outcomes of this work will inform computational models that attempt to simulate evacuation behaviour (by predicting actions) and the time it takes to evacuate more accurately. With more accurate, data-driven models, engineers can develop safer and more secure data-driven building designs and operational procedures. Our goal is to inform building design and operations of physical facilities to provide safer and more secure environments during regular, as well as emergency, conditions.
For more information on how data-driven technology is impacting all aspects of engineering and industry, see our Data-Centric Engineering programme.