Hello, everyone. I am Sunny Singh and I will be presenting our research on the topic experimental study to compare factors influencing exit choice behavior in emergency evacuation situations using virtual reality techniques. Before we start, I want you all to quickly think

about places you've visited in past week. Given the extraordinary circumstances we're living in, it is entirely possible that you might not have visited a lot of places in past months, let alone the last week itself. But just to humor me, if you think about your average week under sort of regular circumstances, you'd realize that we visit a lot of places

and buildings like schools, workplaces, gyms, grocery stores, recreational places, et cetera. According to 2012-2013 Sydney HTS report, which is Household Travel Survey report, more than 40% of our trips destination are places other than school, home, or work,

or what you'd consider as non-frequent travel places. Now, emergency evacuation plan plays a crucial role in safe exit strategy of any building, but in such places with high likelihood of gathering of large crowds without a proper access to evacuation plan,

it becomes highly crucial to understand the impact of behavioral factors on the exit choice of the evacuees. Let's look at a couple of historical examples and see a few examples. On 26th February 1993, a urea nitrate hydrogen bomb was detonated in the basement of

north tower of World Trade Center. The structure sustained the damage and did not collapse into the south tower as per the malicious plans, but it ended up being one of the biggest or the largest scale evacuation exercise in the history with over 50,000 people evacuated

from the building. During the exercise, the logical thing for evacuees to do would have been to follow the fire evacuation plan down to the safe assembly area, but going through the accounts of the event, it was noted that more than 90% of north tower evacuees and 70%

of south tower evacuees reported walking through smoke. Now, when asked for reasons, the reasons stated among others were helping and warning others trying to fight the fire and I mean to the extent of just plain simple curiosity. In another event, a station nightclub

fire that broke out during a rock band performance on 20 February 2003 in Rhode Island claimed 100 innocent lives and injured 230 people. Incident records revealed that despite

the venue being equipped with four exits in total, more than two-thirds of the fatalities occurred near the main exit. As you can see on the top left picture on the screen where most evacuees tried to rush to when the fire broke out. This clearly raises a question,

what went wrong? Could more lives be saved in case we had a better understanding of how evacuees were going to react to the situation? The above two examples clearly demonstrate that there were factors that influenced evacuation behavior of people other than just the physical

capacity of the building, egress time, walking speed, etc. Clearly, human behavior or behavioral factors played an equally important role in defining the outcome of those incidences. These behavioral factors, however crucial and important, have been hard to incorporate into

planning softwares. Above, you can see few examples of most widely used evacuation simulation softwares. While many of these like WayOut, Pathfinder don't provide functionality to model behavioral factors, few others are capable of rule-based or conditional modeling of these

factors. Even when the functionality is there, often it is a cumbersome and difficult task to calibrate these factors due to lack of reliable quality data. This was taken up as one of the main challenges we would try and tackle through our research. Humans have primarily been modeled as homogenous entities without individual emotions or motivations. As described earlier,

data collection from real events is very difficult, while most of the stated preference experiments lack the sense of danger in participants. Large-scale field experiments have similar disadvantages along with the risks associated with large-scale crowds. Through this

study, we aimed to devise an experiment to analyze effects of human factors on exit choice using virtual reality techniques. Here's a brief overview of the methodology adopted for the study. Through literature review, we finalized the factors that we were going to test in the

study. The factors were clubbed into few groups and scenarios were designed. Once the scenario design was complete, suitable sites were located on Monash Clayton Campus to capture 360-degree immersive footage. Equipment used to capture and display the scenario footage can be seen

at the top right corner of the slide. Edited and finalized version of the footage was played to the recruits of the study and their responses were collected for model estimation. Here are a couple of sample videos trimmed from the final compiled video. As you can see in the example,

participants had full freedom to make 360-degree movements to scan the environment around them. Depending on the scenario, the footage was captured from a moving or a still frame reference. Each scenario and case had significant blank gap to record responses of the participants.

Let's quickly go over the design of all three scenarios. Scenario 1 tested three factors, exit familiarity, herding behavior, and exit distance. Participants would be shown an initial video in

which they would enter the virtual room from one of the two entrances as depicted by the blue dotted arrow. The participants would then stand at the position indicated by the blue circle. Based on the case, other participants may leave the room from one of the exits as depicted by

the gray circles before the participant in question was asked to make a choice. Scenario 2 tested effects of exit signs and crowding. Participants depicted by blue circle would walk through the corridor and stop at the T-junction where they would see an exit sign

with or without directional marking. Other participants depicted by gray circle moved in the direction depicted by gray dotted arrows. Scenario 3 tested the effects of

exit obstacles and physical stress or physical activity involved. Scenario 3 had two great separated exits. Participants would start at one of the two levels and make a choice. Some cases also had small obstacles blocking the exit that participants would have to go around to make an

exit. Here's a summary of the final results of the experiment. Discrete choice models estimated for all three scenarios using open source python-based software BioGene are shown in the three tables on the left. Beta of each factor in each scenario were final variable coefficients

estimated which is effectively the weight of the factor in the scenario. So for example beta f would be the weight of familiarity for that scenario. Negative sign of a coefficient would represent inverse relationship between the factor and choice of an exit. For example beta d which

is the coefficient of distance from an exit was modeled to be negative 0.496. This means that increase in distance of participant from an exit reduces the probability of that exit choice. ASC are alternate specific constants. So these are constants for alternate exit choices,

exit choice 1 and 2. As an input to the model ASC for first exit was always fixed to be 0. Interestingly scenario 1 and 2 show no inherent preference towards any exit while in scenario 3 which is the scenario with great separation ASC 2 which was for the first floor exit

came out to be negative and statistically significant indicating the fact that participants preferred to climb down the stairs over climbing up the stairs. Arc elasticity of a factor is change in probability of exit choice of a particular exit

when the factor is changed by a unit value. This helps in ascertaining the relative importance of factors against each other. So as you can see familiarity and exit sign and activity came out to be the most influential factors in scenario 1, 2 and 3 respectively.

It's one thing to come up with a model and generate results in an isolated environment and a different thing to line those results up against some reference. For example historical studies this could further validate the results of the experiment and provide additional confidence

in the methodology. So an experiment conducted in a warehouse setting by Benton and forensic reveals that exit familiarity was valued at least twice as much as distance from the exit by participants similar to the findings of our study. Bode and Kotling suggested in their study

that exit signage had a positive impact on exit choice again as confirmed by our study. The results are expected to vary as layout of the virtual environment changes but that is one of the strengths of this technique because numerous virtual settings can be tested for

influence of same factor and relativities between the factors can be calculated with greater precision to suit the needs of that scenario. In conclusion I would like to say that within the constraints of the study time resource and budgetary constraints it became

abundantly clear that the use of VR technology or virtual reality technology has numerous applications in understanding the influence of human factors in exit choice behavior. The model produced by the data collected from the study agreed with historic accounts of similar studies

and experiments. Methodology provided a fast safe and reliable means to test multiple factors in building presently constructed or maybe even in design phase as well. Thank you all for patiently listening to the presentation. I wish you all a great day ahead.