Agent-Based Simulation of Human Behavior In the Case of Dangerous Events Mikołaj Grotowski Jerzy Mikulik Faculty of Management Faculty of Management AGH University of Science and Technology AGH University of Science and Technology Kraków, Poland Kraków, Poland mikolaj.grotowski@agh.edu.pl jmikulik@zarz.agh.edu.pl Abstract—The present world is struggling to a greater extent, with the problem of ensuring security. Hence the II. STATE OF THE ART growing demand for simulations of dangerous events. Most of these events involve people's participation. In the case of large A. Approaches To the Problem of Evacuation groups, it is easier. However, the problem remains of simulating Simulation the actions of individuals, especially those who are under the In their work [1] Eric D. Kuligowski and Richard D. influence of strong emotions that usually accompany such Peacock, and later [2] from Bryan L. Hoskins, carried out a events. The authors have attempted to solve this problem based on solutions previously used only in the computer games detailed breakdown of the evacuation models. As you can see industry. The work discusses the basic assumptions, in these combinations, many models propose the use of a implemented mechanisms, and results of the first tests of the global approach to simulated people, which assumes that we use of these mechanisms in simulations of building evacuation study the overall flow of people. In such models, it is not processes. The work consists of the following parts: possible to introduce individual behavior of people, as they Introduction, review of existing solutions, the proposal of a do not exist as individual objects. It is also unintentional to solution, description of the implementation, and discussion of simulate with these models the situations resulting from the the results so far. intentional operation of the units present in the building since their impact is usually limited to individual units, and to a Keywords— Security, human behavior modeling, agent-based lesser extent to the flow of people as a whole. In the case of simulation. models treating individual evacuees individually, we can notice that the models that do not take into account the I. INTRODUCTION specific behavior of persons prevail or they consider them in a previously determined, deterministic way. However, more The need to ensure that safety is one of the basic human and more times there are more and more models attempting to needs. To ensure them, all knowledge available to man is simulate behavior in a more realistic way, through a used. No wonder, then, that the simulation of non-safe combination of scenarios and random factors. It should also phenomena has been used since the possibilities of computers be noted that despite the manufacturers' declarations, one can allowed them. However, most of the currently used not speak in such simulations about the use of artificial algorithms have the disadvantage that it does not allow to intelligence as a factor controlling evacuated units. What is investigate how human emotions can affect a simulated called AI by producers is only a more complex system of phenomenon. Our research aims to introduce the influence of assumptions and random factors. It does not contain a basic emotions to the variables under consideration in the learning factor for the definition of AI. At the same time, it simulation process. This is intentional because human seems that it is not advisable to introduce self-learning emotions have a very significant impact on the course of systems for such simulations because in reality, people who activities such as the evacuation of the building or the course are getting involved in the evacuation, for the most part, have of the attack. This is especially important when the dangerous no previous experience with this process. phenomenon is the result of intentional criminal activity, not the forces of nature. Another issue in discussing existing models is the way in which the route that evacuees are moving. Many models For research reasons, we have focused our research on the assume that it will be the optimal route, without taking into problem of simulating the evacuation process of a building. account the knowledge of the participants of the event. A The simulated cause of evacuation can be both a natural large group of models assumes that all people are moving in phenomenon (fire, construction disaster), as well as a the same direction, which is usually true in the case of events detrimental operation (bursting of an explosive charge, attack such as fire, but it can be a problem when other situations are with a firearm, attack with a melee weapon). simulated. Another problem is whether during the movement the influence of neighboring people is examined. This is important when trying to simulate such situations as panic during the evacuation, where the greatest threats are precisely the result of the interaction of individuals. Copyright © 2019 for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0) The last aspect to which we would like to draw attention perspective, conditional and probabilistic behavior, inter- is the way of presenting the evacuated area. At present, the person distance movement, and 3d space modeling. models considering evacuation on the plane prevail. The third dimension is either omitted or used only in visualization. This The basis for the model is the use of objects - agents with prevents a full analysis of the "field of vision" of the the ability to interact with the environment. simulated persons, which translates into a higher level of Each agent has a set of statistic defining its capabilities in abstraction in making decisions about behavior. It is also the reception of information and in its impact on the difficult in such an environment to carry out simulation of use environment. These statistics can change during the firearms, because, as demonstrated by experience, in such simulation. They can be divided into two categories, situations the key factor was often the possibility of hiding "psychic", responsible for the reception of stimuli and from the attacker, and these opportunities resulted mainly "physical" responsible for interacting with the environment. from three-dimensional nature of rooms. In addition, the agent may have assigned markers that indicate sensitivity or insensitivity to specific factors. At the B. Analysis of Requirements current stage, a model using six statistics is tested: Due to the fact that criminal activities are becoming more • Speed - determines the speed of the movement and more serious, it seems that the current approach to the problem of evacuation must change. Today, when planning • Strength - defines the possibility of an impact on the an evacuation, in the majority of cases only the fire hazard is environment taken into account. Potentially, it creates weak spots in the • Durability - defines resistance to the influence of building's security that can be used to attack. To counteract physical factors, reduces penalties to speed and this, during the simulation, not only fire scenarios or similar strength for such factors as the strength of other risks should be considered, but also potential criminal people or smoke activities. In order to be able to interpret the simulation results effectively, it is advantageous that they should be carried out • Knowledge - knowledge of the area, affects the chance on the same model. After tracing the available solutions, we to choose a better route without using the signage decided that we need a model that meets the following requirements: • Perceptiveness - a feature that increases the chances of choosing a route based on marking and reacting to • Enables simulation of criminal activities on par with changes in the environment natural phenomena • Mental resilience - a factor determining how the agent • Can simulate various behaviors of people, including is influenced by psychological factors non-rational ones, resulting from emotional states resulting from an event Each factor affecting the agent has the following parameters: a range of impact (distance or line of increase), • It is possible to simulate complicated interactions type of interaction (physical/psychological), and effect between people (modifiers to the characteristics of the agent). All impacts are modified by random distribution. Also included is a marker of • It is possible to differentiate the parameters of non-sensitivity or vulnerability. The insensitivity mark simulated persons decreases five times, and the vulnerability marker increases • It is possible to carry out a large number of the factor's impact on statistics five times. simulations at a reasonable time with various The path is determined as follows: first, it is tested by parameters and variants, which are also automatically means of a random factor modified by the perceptive eye, or created. the agent sees where there is a way out of a particular room. • The predicted route of movement is dynamically If so, the road is defined as the route with the smallest cost of created on the basis of information available for a interactions towards the evacuation marker. If this is not the specific person. case, a random test is performed with the feature of knowing if the agent still chooses a good exit. If not, it moves to the Similar models are used in the video game industry to site where the impact is the smallest. It is not necessary to create opponents for players - bots. Hence, we decided to use create a separate principle of following the crowd, because in the solutions used in creating games, as it is about simulating this case the lowest cost usually causes such behavior. the influence of emotions, interactions and the ability to make decisions under the influence of stress. Emotional effects are simulated by modifying characteristics - for example, panic is defined as an increase in strength and endurance along with a decrease in perception III. PROPOSED MODEL and knowledge. The influence of physical factors is simulated According to the classification contained in [1] and [2], in a similar way, for example, smokiness means a decrease in the created model can be defined as a behavioral model with strength, speed and perceptiveness. risk assessment capabilities, models that can simulate any type of building, continuous Grid/Structure, individual As you can see the basic principles of the model are quite simple, which allows for easy implementation and good computational efficiency. IV. THE PLANNED ASSESSMENT METHOD At the moment, it is planned to evaluate this model with two methods. First, the simulation results are compared to the passage of real events. Unfortunately, this must be real cases, not exercises, because this model focuses on phenomena that do not occur during tests. The second method is a comparison with other models, mainly with the BENTLEY SYSTEMS Legion model as the have most similar functionally. ACKNOWLEDGMENT The authors wish to thank AGH UST for providing financial means for the research and publication of the paper. REFERENCES [1] Erica D. Kuligowski, Richard D. Peacock “A Review of Building Evacuation Models,” Natl. Inst. Stand. Technol. Tech. Note 1471, U.S. GOVERNMENT PRINTING OFFICE WASHINGTON: July 2005 [2] Erica D. Kuligowski, Richard D. Peacock, Bryan L. Hoskins, “A Review of Building Evacuation Models, 2nd Edition” Natl. Inst. Stand. Technol. Tech. Note 1480, U.S. GOVERNMENT PRINTING OFFICE WASHINGTON: November 2010