Simulating Behaviours to face up an Emergency Evacuation

Simulating Behaviours to face up an Emergency Evacuation
Pablo Cristian Tissera, Alicia Castro                     A. Marcela Printista                           Emilio Luque
LIDIC-Departamento de Informбtica               LIDIC-Departamento de Informбtica             Departamento de Arquitectura y
Sistemas Operativos
Universidad Nacional de San Luis                 Universidad Nacional de San Luis           Universidad Autуnoma de Barcelona-
Argentina                        Conicet CCT-San Luis, Argentina                             Espaсa
Email:{ptissera,adcastro}@unsl.edu.ar                  Email: mprinti@unsl.edu.ar                  Email:emilio.luque@uab.es

Abstract Computer based models describing pedestrian               hybrid model where the dynamics of fire and smoke
behavior in an emergency evacuation play a vital role in the           propagation are modeled by means of CA and for simulating
development of active strategies that minimize the evacuation          people’s behavior we are using intelligent agent (IA) concept.
time when a closed area must be evacuated. The reference model
has a hybrid structure where the dynamics of fire and smoke                In order to cover a variety of scenarios suitable for
propagation are modeled by means of Cellular Automata and for          checking safety issues, in this model we will focus on
simulating people’s behavior we are using Intelligent Agents. The      pedestrian behavior. In the model, each individual or
model consists of two sub-models, called environmental and             pedestrian is an agent that will have its own thread of control
pedestrian ones. As part of the pedestrian model, this paper           and be able to run appropriate actions according to their own
concentrates in a methodology that is able to model some of the        state, the perceived environment and the messages provided by
frequently observed human’s behaviors in evacuation exercises.         the system (external stimuli) independently.
Each agent will perceive what is happening around, select the
options that exist in that context and then it makes a decision that       The proposed strategy is similar to the philosophical
will reflect its ability to cope with an emergency evacuation,         theories of Dennett and Bratman [9], [10], [11] where each
called in this work, behavior. We also developed simple exercises      agent will create a high-level model with actions to be taken to
where the model is applied to the simulation of an evacuation due      achieve a goal and plan how to implement them. The agent
to a potential hazard, such as fire, smoke or some kind of             perceives what is happening around, selects the options that
collapse.                                                              exist in that context and then the agent makes a decision that
will reflect its ability to cope with an emergency evacuation,
Keywords— Evacuation Simulation; Behaviors; Cellular                called in this work, behavior. Thus, in the model creation are
Automata; Intelligent Agents                                           involved three sub-processes, where different configurations of
them carry out different scenarios for the agent.
I.    INTRODUCTION
Our model considers that the making-decision process that
In the last years, several modeling approaches have been           an agent makes to evacuate will be driven by an engine of
proposed to deal with the emergency evacuations because the            behavior which concentrates different mindsets or reactions to
prediction of the people’s behavior is of great public interest.       face up to an emergency situation.
Microscopic approaches allow to investigate how the system
state evolves during the model runs. Within this category, there           Through interaction and coordinated evolution of CA and
is a diversity of models, including perhaps the best known are         IA models, it is possible to obtain a model capable of
those based on the Social Forces Model (SFM) [1], Cellular             simulating indoor environments with a finite number of exits
Automata (CA) [2], [3], [4], [5] and Multi-Agent System                that must be evacuated by individuals with different behaviors
(MAS) [6], [7]. Models based on the concept of social forces           due to an emergency evacuation. The proposed simulation
and cellular automata, can represent individuals as basic units        system allows specifying different scenes with a large number
of the system but they have the limitation of assuming that the        of people and environmental features, making easier the study
individuals depicted are homogeneous, i.e. the behavior of all         of the complex behaviors that arise when the people interact.
of them will be governed by the same set of rules [8].                 Our proposal could be used by architects, government
agencies, foundations, etc. in order to know the security threats
CA allow to generate “local” and “uniform” behaviors that          of a possible disaster, help with appropriate actions of
resemble the dynamics observed in real processes of fire and           prevention for in a quick and efficient way to evacuate a
smoke propagation. CA can be viewed as a simple model of a             building through the design of active policies that minimize the
spatially extended decentralized system made up of a number            evacuation time when circumstances require it.
of individual components (cells). The communication between
their cells is limited to local interaction. Each individual cell is      In section II we present the Hybrid Model for the
in a specific state which changes over time depending on the           Evacuation Simulation. Sub-section II-A resumes the sub-
states of its local neighbors.                                         model called environmental (EsM). Sub-section II-B explains
the pedestrian sub-model (PsM) and addresses the agent
However, these local features were not suitable for                 architecture adopted in this work. The section III explains the
representing certain aspects of people’s behavior that requires a      implementation of two behaviors commonly observed in
more specific and differentiated perspective. We developed a
emergency evacuation. In section IV we describe our work
with different instances of the problem at hand and report the                W: External wall cell       E: Empty cell       S: Cell with
performance analysis of each case. Finally, the section V                                                                     smoke
presents the conclusions and the section VI we put forward                    P: Cell with a person       PS: Cell with a person and smoke
possible future works.
O: Internal obstacle cell   SF: Cell with smoke and fire
II.    HYBRID MODEL
The use of cellular automata (CA) is attractive because it is                 The neighborhood considered in the model is Moore’s
a simple and extremely efficient model for computer                           Neighborhood, that includes the eight cells surrounding the
simulation. Unfortunately, these models impose certain                        central cell. With this choice we aim to provide to each
limitations that make it difficult the emergence of different                 individual in the system with all possible movement directions.
social behaviors, from the interaction of a heterogeneous group               Before the simulation starts, a variety of information related to
of individuals with their perceived environment.                              the outer walls, inner obstacles, individuals, combustible
The considered simulation model is able to take advantage                 locations, cell with fire and arrangement of the exits must be
of both cellular automata and intelligent agents [12]. It consists            specified. In our tools this task can be done by means of its
of two sub-models, called environmental (EsM) and pedestrian                  graphical interface.
(PsM). This model along with the computational methodology                        Model Evolution Rules: a cell in state W or O (outer wall or
allows us to build an artificial environment populated with                   obstacle) will not change its state throughout the simulation
autonomous agents, which are capable of interacting with each                 (Rules about the building). For fire and smoke spreading we
other. Fig. 1 shows the hybrid model.                                         use a probability-driven model [13]. A cell with smoke (S, SF
The EsM, based on CA, describes the spatial configuration                 or PS) at time t, will also have smoke in time t+1. If at time t
of the environment (geometry of space, exit doors, internal                   the central cell does not have smoke, but some of its adjacent
barriers, etc.) and models the processes of diffusion of smoke                cells have smoke, the central cell will also have smoke at time
and fire, while the PsM uses the concept of intelligent agents to             t+1 with a proportional a probability to the number of adjacent
describe the cognitive processes of individual agents and                     cells with smoke. The rules about fire propagation are
interactions among multiple agents in a specific environment.                 analogous to the rules about smoke propagation. If a cell has
Through interaction and coordinated evolution of these two                    smoke or fire, it will change its status from a min-value that
sub-models it is possible to obtain a model capable of                        indicates the recent emergence of the element (smoke or fire)
simulating indoor environments with a finite number of outputs                in the cell, to a max-value indicating that the cell has been fully
that must be evacuated by a group of people due to the threat of              saturated by the spread element (smoke or fire).
fire and the effect of the smoke. The evacuation exercises                       The simulation takes an updating time of 0.3 seconds by
adopt the CA model for advancing the simulation time.                         time-step. This value is the estimated time required by a
pedestrian for walking 0.4 m (size of a cell side).
For reasons of space we only include a summary of the CA
model, for details to refer to previous work.

B. Pedestrian Model

Intelligent agents (IA) have been used successfully in a wide range of applications. In artificial intelligence, an intelligent agent is an autonomous entity which observes and acts upon an environment and directs its activity towards

Fig. 1. Hybrid Model consisting of environmental and pedestrian sub-models. achieving goals. The PsM is the part of the hybrid model focuses on representing the human behaviors. Evacuation processes can be considered a special type of crowd dynamics,

A. CA Model for Environmental Simulation with the presence of hazards. Each agent has a set of static characteristics (such as sex, age and level of stress tolerance)

The CA are discrete dynamic systems with capacity to and dynamic (such as current stress level and speed). In develop complex behaviors from a simple set of rules.

addition to that, the crowd behaviors are complex phenomena,

Basically, these rules allow specifying the new state of a cell it is necessary to focus on various factors that determine the based on the state of the neighboring cells. In this way, it is human behavior and that may contribute to emergency possible to model complex dynamic systems from the behaviors.

specification of the local dynamics of its components. The

Cellular Space is a finite bi-dimensional array (grid) with In this paper, the agent behavior will be determined by their closed boundaries. Each cell of the cellular space represents 40 current perceptions and attitudes of reaction. This type of

Ч 40 cm2. Each cell can be in one of the states of the set Q= system provides solutions in dynamic and uncertain

{W, E, P,O, S, SF, PS}, where: environments, where the agent has only a partial view of the problem. The procedure can be followed in the Algorithm II.1.

At first, the agent must observe the environment and gather 1) If its plan is reasonable (step 9), i.e. the agent can the state of outside world with its inner world (step 1 of the achieve the objective with the current plan, then it will algorithm II.1). With this information the agent updates its execute the next step of the plan (step 6), otherwise the knowledge and starts to think about their possibilities of escape agent should select a new plan (step 10). For example,

(step 2). Consider for example two types of individuals: a on the way to an exit, an obstacle appears and then the person familiarized with the place (kp) and another that agent must find another route to the objective exit.

completely ignores the place to evacuate (ukp). The agent kp has a rich set of information to built their alternatives, whilst 2) If the agent decides that its alternative is not neither the world of ukp just contains what it is watching right now feasible nor desirable (step 5) then it will return to the and what it was able to learn until evacuation began. After step outer loop and deliberate about the convenience of

2, they will have different knowledge. select another objective (another exit) or even to change of strategy, resulting in this last case a change of state in

Next, the agent should contrast their alternatives before its engine of behavior (step 3). For example, an exit can making a decision. Both Dennett’s theory about intentional be blocked completely by the spread of the fire along systems, as well as Bratman’s theory of practical reasoning the evacuation evolution; therefore there will be no about humans, agrees that in any act of a rational agent at least possible path that leads to it.

involves a deliberative process and a process of practical reasoning leading to an action. Algorithm II.1: control loop of the agent()

This deliberative process will take into account both the while (NOT(success evacuation)) world and the agent options. These latter will be biased by its personality type and behavior. Will it choose the nearest exit? - Get next world state (1) will it choose the exit that it infers faster? The agent gets lost, - Update knowledge (2) will it follow the other agents? The different behaviors that an - Deliberates and Commits with an objective (3) agent can adopt represent their options. In our implementation, - Get plan to achieve the objective (4) each agent has an engine of behaviors allowing it to have different goals based on their different escape criterion. The do while (NOT (empty plan or success evacuation engine of behavior manages deliberative process (step 3). As or feasible objective)) (5) can be seen in Fig. 2, it is a non-deterministic finite automaton, where each node represents the implementation of a behaviors - Execute plan (6) while the transitions represent the event for which the do - Update knowledge (8) individual can change its behavior. Now, the deliberative If NOT (reasonable plan) (9) process concerns with combining new facts with existing Then - Get new plan (10) knowledge for solving different situations. At the end of the deliberation process the agent will get its objective, that End.

consists of the selection of a specific way to evacuate (”the III. BEHAVIORS agent decided to behave this way”) and it will commit with it.

For example, the agent ukp could have decided to evacuate by This section explains the process for two types of the nearest exit, then it must select the closer exit regardless of individual behaviors implemented till now. In the first the number of people heading to the same exit. behaviors, called the Nearest Exit, the agent will try to get out the closest exit to its current position. In this behavior the

Finally, the agent somehow must get the selected exit (its decision process will take into account the position of the plan) (step 4). The plan will consist in following a set of paths agent, the direction toward the nearest exit, the state of its that lead to the selected exit. In our implementation, the set of environment in relation to the progress of fire and smoke, but it paths is determined by Dijkstra’s algorithm from the cell ignores information from other alternative solutions, the occupied by the agent to the selected exit. behaviors of other agents and it will not take unexpected or altruistic decisions. The decision-making process selects the exit with the shortest PredDistj . This measure represents the estimated distance from current position to the exit j.

In the second behaviors, the agent will analyze different exits and choose one that it predicts the fastest exit to evacuate. In this behaviors, called the Best Predicted Exit, the decision process will consider the position of the agent, the state of its environment in relation to the progress of fire and smoke, the distance to alternative exits, the density

Fig. 2. The deliberative process implemented by an engine of behaviors.

of the crowd trying to evacuate for each exit (only if the agent can see the exit) and the stress level in relation to its

The decisions that the agent is taking are continually tolerance to it. Under this behavior, the agent does not adjusted by a process that takes into account both follow the actions of others, i.e. it remains in the category characteristics of the individuals and the dynamics of their of individual behavior but it only checks the orientation of perceptions. So, the agent does the following: the other agents to choose its way toward the best exit. As the evacuation progresses, the agent is predicting the cost agents expressed their desire to move to the nominated cell,

(in time) to evacuate by each of the exits that are available they should delay their movement until a conflict resolution in the environment. The inferred lower cost will indicate process is executed. To solve the collisions we changed the the best exit. For that, the decision-making process approach commonly used in other works. Instead of being the evaluates the following cost function: agents who decide, the selected cell is carrying out the agents competition process. The hosting decision is concentrated in

Costj=MinPredTime.Toj*PredDistj*Ij; each nominated cell.

if MinPredTime.Toj >= EvacPredTimej .

The conflict resolution process must solve two types of

In other case: situations:

Costj=EvacPredTimej*MinPredTime.Toj *PredDistj *Ij; • The conflict occurred when multiple agents chose the same free cell. In this case the process gives priority to the where: selection of agents with greater speed and fewer points of damage (parameter specified in the environment). If the

• Ij represents the estimated number of agents who conflict persists, the selection will be random.

intend to evacuate through the exit j ;

• The conflict occurred when multiple agents requested a cell

• EvacPredTimej represents the evacuation occupied by another agent. The process must check if the estimated time of exit j. This factor takes into cell will be free in the next time step. If it will be free, the account the intentions of other agents to escape by procedure of the same free cell is executed. Otherwise, the the door j; agent will not move of its current cell.

•    MinPredTime.Toj represents the minimum                      Finally, it is important to point out that after a prudential
estimated time needed by the agent to reach the         time, if an agent remains without advancing, it will try to
exit j considering a free path (unobstructed); and      change its neighborhood. In spite of failing to advance to the
desired direction, it is reasonable to want to keep moving until
•    PredDistj represents the estimated distance from        finding a way towards the exit.
current position to the exit j. This calculus is made
using the Dijkstra algorithm.
IV.     TEST-CASE SCENARIOS AND RESULTS
The experiments were carried out with EVAC Simulator
Note that for the case where MinPredTime.Toj>=                   [2], an integrated simulation system. EVAC is a system
EvacPredTimej it does not take into account the value of             developed in Java that allows the design and simulation of
EvacPredTimej , because if the spent time by the agent to reach      spatial environments in an explicit way. EVAC simulator
the door "j" is greater than the time taken to dislodge the door,    offers a friendly graphical interface which can be easily used
then when the agent arrives at the door, it would find it empty.     by non expert users. We performed a series of experiments in
order to test the behaviors under study focusing on showing the
The resulting procedure instructs the agent for which exit to    results of the interaction of the two sub-models. For all the
go (objective). In addition, the procedure involves a dynamic        cases, the environment is of 20 x 30 meters with 625
factor used to adjust the number of times the agent executes the     individuals and two Exits of 1.6 meters each one located in
action to evaluate the exits. This factor limits the effect of       opposite ends of the environment (called E1 and E2). We will
indecision of the agents that occurs when simultaneously             call βNE to those individuals whose configuration of behavior is
multiple agents make the same prediction of fastest exit to          Nearest Exit (NE) and βBPE to those whose configuration of
evacuate. As a result, as time goes on, agents can questioning       behavior is Best Predicted Exit (BPE).
the decision taken by themselves and try to execute the
deliberation process again (the agent comes and goes). This              With the purpose of obtaining acceptable statistical data,
factor depends on the environment size and it is dynamically         the results shown in Table I, II and III correspond to the
adjusted according the time elapsed since the beginning of the       average of 50 independent replications of each experiment. The
evacuation.                                                          corresponding confidence intervals were obtained, for the total
evacuation time (TET) (seconds), mean evacuation time
Path to the Selected Exit                                            (METxI) (seconds) and mean travelled distance (MDxI)
Each agent knows the objective exit. Based on the                 (meters) per individual to the exit. For the experiments called
perception of its environment, the agent must select at each         A and B, all of the individuals in the environment have βNE
time step, a position on the path to its exit.                       behavior. The cases of study C and D are analogous to the
previous ones but they have βBPE behavior. It is important to
If in the neighborhood there are free cells that come near       note that in the cases A and C, the environment is not affected
the exit, the agent will choose the one that most benefits grant     by the spread of fire and heat, but in cases B and D individuals
(leave it closer). If there is more than one, the choice of the      must adapt to a dynamic environment due to the spread of fire
position is random. If in the neighborhood there is no free cell,    and heat.
following the approach described above, the agent will select a
cell that is currently occupied by another agent. This can cause         As can be seen in Table 1, in the case A about 60% of
multiple agents try to occupy the same physical location             individuals decided to use E2. In the case B, a fire started in an
belonging to an exit way. To solve the problem, after the            area near E2. From the perception of this danger zone, agents
were going to that exit began a new deliberation process and             TABLE I: 625 INDIVIDUALS DISTRIBUTED EVENLY IN THE BUILDING.
some of them could change its objective.                             2 EXISTS. #EI INDICATES THE NUMBER OF PEDESTRIAN EXISTING DOOR EI
Case       TET (sec)       METxI (m)        MDxI (m)        #E1       #E2
Remember that in cases A and B, people choose the exit             A        98.91- 99.56     13.40-13.41     16.07-16.08     262.00    363.00
that they consider closest to them (for distance). The pedestrian      B       108.28-122.51     20.05-20.12     24.05-24.15     425.94    199.06
model considers zones of fire, smoke or heat as obstacles              C       101.63-106.40     14.43-14.44     17.32-17.33     301.86    323.14
(dynamic characteristic of the environment), so paths through          D       118.01-122.30     28.17-28.26     33.79-33.91     363.10    261.90
that zone will be considered invalid. When fire appears, agents
must re-plan their strategy to surround the fire or escape on the
other hand. Of the original 365 individuals who had the                  The results obtained for this series of experiments show us
intention to walk out the E2, 45% of them considered that the        that there is a decrease of approximately 10% of individuals
E1 was the nearest exit and therefore changed their objective,       who select the E2 as objective due to the appearance of fire and
while the rest of the agents surrounded the fire (by another         heat in the neighborhood of it (Table II cases E and F and exit
plan) but maintained their intention to go out the E2. Finally, in   # E2). Also it is possible to visualize that as there is increase the
this experiment, about 68% of individuals were evacuated by          amount of individuals βNE to evacuate by E1, the number of
the E1.                                                              individuals βBPE that decide to evacuate for the same exit
decreases (Table II). This behavior more risky observed in
This situation also causes an increase in the evacuation         individuals BPE is due to its perceptions they considered as a
times and the mean travelled distance per individual to the exit     factor to define its objective the congestion of the exits. Finally
as shown in Table 1 for cases A and B.                               as you would expect both the evacuation times as the mean
In a similar way we carried out a new series of experiments      travelled distance per individual (Table III) are increased due
(cases C and D), but this time the individuals can choose any        to the need of the individuals to surround the areas with fire
exit to evacuate according to its perceptions of the environment     and heat.
and a making-decision process to select the exit that the agent
predicts faster. In the case C, the emergent behavior of
individuals, results in equilibrium in the selection of exits.
Approximately 50% of individuals choose to evacuate by E2
while the remaining 50% choose evacuate by E1. In case D is
possible to see how the amount of individuals that decided to
evacuate for the E2 is decreased due to the threat of fire and the
heat in approximately 10% (around 61 individuals). The
differences that can be viewed between the evacuation times
and the mean travelled distance per individual of the
environments A vs. C and B vs. D, lies in that in the
environments C and D some individuals, after a process of             Fig. 3a. (right) Case A. Without fire - 100% TNE. E1: down left, E2: up right.
Fig. 3b. (left) Case B. With heat and fire - 100% TNE
deliberation, decided to change more than once their objective
(Exit through evacuate) along the evolution of the model, i.e.
the individuals re-evaluate its decision and obtain a new
TABLE II: 625 INDIVIDUALS DISTRIBUTED EVENLY IN THE BUILDING. 2
objective, which may be interpreted as a kind of “repentance”.       EXISTS. #Ei INDICATES THE NUMBER OF PEDESTRIAN EXISTING DOOR Ei. #Ei
Fig. 3a (right) shows the behavior of individuals to the case A.     βNE AND #Ei βBPE INDICATE THE NUMBER OF PEDESTRIAN βNE AND βBPE
In Fig. 3b (left), we can see how a lot of individuals who had       EXISTING DOOR EI.
decided to evacuate by E2 now goes to E1 due to the spread of             Case                 #E1 β NE            #E1 βBPE                #E1
fire and heat in the environment. Both figures were obtained               E                    153.56              175.78                329.34
through the display module of our tool when the evacuation                 F                    250.78              126.16                376.94
time is equal to 30 seconds.                                               G                    280.52              104.92                385.44
Case                 #E2 β NE            #E2 βBPE                #E2
The second set of experiments (cases E, F and G) has as                E                   139.44               156.22                295.66
aim show the existing interaction between both types of                    F                    42,22               205,84                248,06
individuals. In this set of experiments approximately the half of          G                    12.48               227.08                236.56
the individuals has βNE behavior, while the remaining
individuals are βBPE. The difference between these three case            TABLE III: 625 INDIVIDUALS DISTRIBUTED EVENLY IN THE BUILDING.
studies lies in the number of cells that can be achieved by the      2 EXISTS. EVACUATION TIMES AND MEAN TRAVELLED DISTANCE PER
spread of fire and heat. In the cases F and G the environment        INDIVIDUAL.
changes in a dynamic way due to the spread of fire and heat,                 Case            TET (sec)           METxI (m)            MDxI (m)
but the quantity of affected cells is minor for the case of study             E           96.73-106.02         14.12-14.13         16.94-16.95
F that for the case G. On the other hand, in the case study E the             F           121.72-125.75        20.35-20.40         24.41-24.48
environment is not affected by the spread of fire and heat.                   G           132.40-141.63        25.77-25.89         30.91-31.08
V. CONCLUSIONES                                At last, due to the incorporation of new features, the model
complexity increases. It knows that a simulation model with
In this paper we have presented a hybrid simulation model        computation time and memory requirement scaling linearly
to test behavior designs in an emergency evacuation. The             with the number of persons or the size of the environment is
model consists of two sub-models, called environmental (EsM)         desirable. In that direction, in the future we will extend the
and pedestrian (PsM). The paper considers that any decision is       model for supporting high performance simulation in order to
influenced by the knowledge, perceptions and attitude of             accelerate the simulation and take advantage of modern
people. These aspects lead, even in emergencies, to make and         computer architectures.
solve a deliberative process. The model pro-poses that the
making-decision process that an agent makes to evacuate, will                              ACKNOWLEDGEMENTS
be driven by an engine of behavior which concentrates
different mindsets or reactions to face up an emergency                 We appreciate the guidance received from the IA research
situation.                                                           group of National University of San Luis. This research has
been supported by the UNSL, Argentina, CONICET
This model along with the computational methodology              Argentina, the MICINN Spain and the MINECO (MICINN)
allows us to build an artificial environment populated with          Spain.
autonomous agents, which are capable of interacting with each
REFERENCES
other.
[1] D. Helbing and P. Molnбr, “Social force model for pedestrian
The paper develops a series of experiments. First we set up              dynamics,” Phys. Rev. E, vol. 51, pp. 4282–4286, May 1995.
a test environment for an evacuation of individuals whose
[2] P. C. Tissera, A. M. Printista, and M. Errecalde, “Evacuation
behavior is out for the predicted nearest door and then for the             simulations using cellular automata” Journal of Computer Science &
predicted faster door, both with and without spread of fire and             Technology, vol. 7, no. 1, pp. 14–20, 2007.
heat. In the second set of experiments, we analyze the
[3] V. J. Blue and J. L. Adler, “Emergent fundamental pedestrian flows
evacuation of pedestrian with mixed agent behaviors.                        from cellular automata microsimulation” Transporta-tion Research
Record, vol. 1644, pp. 29–36, 1998.
The experiments not only check the impact of individual
behavior in the time of evacuation, but also analyze the impact         [4] H. Klupfel, T. Meyer-Kцnig, J. Wahle, and M. Schreckenberg,
of the parameters that are included in the cost function                    “Microscopic simulation of evacuation processes on passenger ships”
in Proceedings of the Fourth International Conference on Cellular
evaluated by the decision making-process. Among others, we                  Automata for Research and Indus-try: Theoretical and Practical Issues
examined the impact of distance, population density and width               on Cellular Automata. London, UK: Springer-Verlag, 2000, pp. 63–
of the exits.                                                               71.

An important result to notice is the impact of the                  [5] C. Burstedde, K. Klauck, A. Schadschneider, and J. Zittartz,
“Simulation of pedestrian dynamics using a 2-dimensional cellular
emergence of a danger zone. When an agent perceives some                    automaton” Physics A: Statistical Mechanics and its Applications, vol.
dynamic factor, such as fire or smoke, it can enter into a                  295, no. 3-4, pp. 507–525, 2001.
process of deliberation of objectives (E1 or E2 in the current          [6] N. K. Pan, C. S. Han, K. Dauber, and K. H. Law, “A multi-agent
experiments) or re-plan the route in case of not being                      based framework for the simulation of human and social behaviors
reasonable the chosen plan. The deliberation process is                     during emergency evacuations” AI and Society. The Journal of
managed by an engine of behaviors that represents different                 Human-Centred Systems, vol. 22, no. 2, pp. 113–132, November
reactions to danger. Note that in all cases, the 625 individuals            2007.
knew and saw the exit, which optimized its decision making              [7] S. Sarmady, F. Haron, and A. Z. H. Talib, “Multi-agent simulation of
process. In case of ignoring this information, agents can make              circular pedestrian movements using cellular automata” in
Proceedings of the 2008 Second Asia International Conference on
sub-optimal decisions, increasing not only their MET but the                Modelling & Simulation (AMS). Washington, DC, USA: IEEE
total evacuation time (TET).                                                Computer Society, 2008, pp. 654–659.
VI. FUTURE WORKS                                [8] D. Helbing and A. Johansson, “Pedestrian, crowd and evacuation
dynamics” in Encyclopedia of Complexity and Systems Science, R.
One of the main advantages of the proposed model with                   A. Meyers, Ed. Springer, 2009, pp. 6476–6495.
respect to current CA models is that we can describe a set of           [9] D. C. Dennett, “Intentional systems” Journal of Philosophy 68, pp.
individuals with heterogeneous behaviors and provide a                      87–106, February 1971.
framework that will allow us to enhance and add new
[10] Dennett, The Intentional Stance. MIT Press, 1987.
behaviors to existing ones.
[11] M. Bratman, Intentions, Plans and Practical Reason. Harvard
Besides adding new behaviors, our next step is the study                 University Press, Cambridge, MA, 1987.
and consideration of those events that can cause behavioral
[12] P. C. Tissera, A. M. Printista, and E. Luque, “A hybrid simulation
changes in the engine. In the model, each agent has a set of                 model to test behaviour designs in an emergency evacuation” Procedia
psychological (level of stress and memory), physiological (age,              Computer Science, vol. 9, no. 0, pp. 266–275, 2012, proceedings of
sex, speed and level of health) and social (level of training and            the International Conference on Computational Science, ICCS 2012.
knowledge) characteristics that describe it, which can also be          [13] W. W. Hargrove, R. H. Gardner, M. G. Turner, W. H. Romme, and D.
classified as static (age, sex, etc.) or dynamic (level of health,           G. Despain, “Simulating fire patterns in heterogeneous landscapes”
speed, etc.). It will be necessary to analyze which events on                Ecological Modelling, vol. 135(2:3), pp. 243–263, 2000.
these characteristics, or news arising from the environment
sub-model, may cause the changes.