Comparison of Improved Floor Field Model and Other Models

Hyunwoo Nam, Suyeong Kwak and Chulmin Jun

Department of Geoinformatics, University of Seoul, Seoul, Korea, Republic of

Keywords: Floor Field Model, Pedestrian Dynamics, Cellular Automata, Indoor Evacuation, Microscopic Model.

Abstract: This study introduces an improved Floor Field Model (FFM) that models pedestrians using realistic physical

characteristics (size, shape, and posture). Through comparison with other well-known models, the areas of

improvement are elucidated. The FFM is a leading microscopic pedestrian model that uses cellular automation

(CA), but it does not accurately reflect the physical characteristics of pedestrians, such as their size, shape,

and posture. Therefore, it is difficult for the existing FFM to simulate certain phenomena, such as collisions

and friction between pedestrians. This study proposes an improved FFM that can simulate these phenomena,

and experiments were carried out to compare this model with other models, such as the existing FFM, Simulex,

and Pathfinder, to confirm the improvements. Through this experiment, it was confirmed that inter-pedestrian

phenomena, such as collisions, friction, and jamming, could be realistically simulated.

1 INTRODUCTION

The Floor Field Model (FFM) is a microscopic

pedestrian model that utilizes cellular automata (CA).

The FFM effectively models pedestrian movements

with simple arithmetic calculations. However, in the

FFM, the size and shape of the pedestrians in the

model differ significantly from the characteristics of

actual pedestrians (Burstedde et al., 2001; Kirchner et

al., 2002). To realistically simulate detailed physical

phenomena (collision, jamming, etc.), a shape similar

to the shape of an actual pedestrian should be used;

oval figures would fulfill this purpose. However, the

FFM assumes a circular pedestrian, and thus it is

difficult for it to reflect certain inter-pedestrian

phenomena, such as collisions and jamming.

Therefore, we have developed an improved

pedestrian model with additional features, such as the

pedestrian size, shape, posture, and turns. The

improved model follows the basic rules of the FFM,

and various factors, such as the lattice space and

posture determination probability, were modified and

added to include the abovementioned features.

Through these modifications, inter-pedestrian

phenomena, such as collisions and jamming, were

reflected in the pedestrian modeling process.

Moreover, comparative experiments were carried

out to confirm that the improved pedestrian model

resolves the limitations of the existing FFM. The

existing FFM and two other widely used models,

Simulex and Pathfinder, were selected for

comparison. The purpose of the models in use is to

assess the performance of the improved model. The

experiment was carried out in the first-floor space of

a building on the campus of the University of Seoul,

and various factors were investigated, including

changes in the evacuation time depending on the

evacuating population, assuming equal population

distribution among the available exits, and evacuation

conditions. Through this, it was confirmed that the

improved model showed effects that were not

previously seen in the existing FFM, such as collision

and jamming, and yielded evacuation experiment

results that were very similar to those of the models

currently in use.

2 RELATED WORKS

2.1 Floor Field Model

Burstedde first introduced the FFM in 2001

(Burstedde et al., 2001). The FFM was designed a

two-dimensional CA model and displayed various

factors that impact pedestrian movement using

various floor fields (Kirchner et al., 2002). The FFM

utilizes cellular spaces composed of lattices, and one

lattice typically encompasses a 40 cm × 40 cm square.

Detailed information on the FFM can be found in

Nam, H., Kwak, S. and Jun, C.

Comparison of Improved Floor Field Model and Other Models.

DOI: 10.5220/0005658700950101

In Proceedings of the 8th International Conference on Agents and Artiﬁcial Intelligence (ICAART 2016) - Volume 1, pages 95-101

ISBN: 978-989-758-172-4

previous related works (Burstedde et al., 2001;

Kirchner et al., 2002; Nishinari et al., 2004).

However, the FFM sets the pedestrian size and

shape according to the size of the lattice. Therefore,

the size and shape of the pedestrians in the FFM

results differ significantly from the size and shape of

actual pedestrians, meaning the model does not

reflect the influence of collisions and turns between

pedestrians. Although studies have been conducted to

improve the existing FFM, most of these studies

utilize the spatial structure, pedestrian size, and

movement methods as defined by the FFM (Henein,

2008; Kirchner et al., 2003; Kirchner et al., 2004;

Kirik et al., 2007; Kretz, 2006; Kwak et al., 2010;

Nishinari et al., 2005; Suma et al., 2012; Varas et al.,

2007; Yanagisawa and Nishinari, 2007; Yanagisawa

et al., 2009). The model suggested in this study is

different from the models in these previous studies, as

the overall characteristics of pedestrian size and the

placement methods of the FFM were approached

from a different point of view. Therefore, the

pedestrian posture can be defined, and the collision

phenomena resulting from changes in posture can be

modeled. The improved model is further explained in

Section 3.

2.2 Simulex

Simulex is an evacuation simulation model

introduced by Thompson (Thompson et al., 1997).

The spatial data can be structured using a CAD

interface, and an evacuation simulation can be

performed by placing exits and pedestrians within a

defined space. Moreover, Simulex was designed to

make it possible to conduct simulations in a complex

building setting with multiple floors.

Simulex defines data through the floors and stairs

used by pedestrians, links that link the two factors,

and the final exit of the building. Moreover, it uses a

distance map that defines the distance of each space

from the exit (Figure 1). As for the pedestrians,

various factors defining their motion are set,

including the placement location, the physical

characteristics of the pedestrian body type, and the

psychological characteristics dictating which exits

should be used and how quickly they should respond

to evacuation orders.

When the pedestrians are given orders to evacuate,

they carry out the evacuation process based on their

individual response times, and the evacuation routes

are obtained using the distance maps. Throughout the

moving process, the walking speed of each pedestrian

is determined by the preset pedestrian characteristics,

and in the case of bottlenecks, these speeds are

reduced to values near 0. Moreover, the pedestrians

can freely spin their bodies, and as such, collisions

occur between pedestrians.

Figure 1: UI and distance map of Simulex.

2.3 Pathfinder

Pathfinder is an agent-based evacuation simulator. It

contains various simulation functionalities, such as

the writing and editing of spatial data, an analytics

tool for three-dimensional results, in addition to the

basic pedestrian simulations. Pathfinder utilizes two

models, Society of Fire Protection Engineers (SFPE)

and steering, to determine the pedestrian movements

(Thunderhead Engineering, 2009).

The SFPE mode was developed based on the

concept proposed by (Nelson and MacLennan, 2002)

and represents the pedestrian movements as a flow.

Moreover, pedestrians are influenced by the exit

locations and pedestrian density and determine their

walking speeds in a kinetic manner. At the exit, the

width of the exit heavily impacts the walking speed.

The unique characteristic of the SFPE mode is that

there are no physical collisions between the

pedestrians. Therefore, the situation created by this

model is unrealistic, as numerous pedestrians are

modeled as being in the same locations. However,

while there are no physical collisions, i.e., when the

pedestrians are modeled as occupying the same space,

the density increases, which still has an impact on the

actions of the pedestrians (slower speeds).

The steering mode was developed as an

improvement to (Raynolds, 1999) and (Amor et al.,

2006). The steering mode focuses on providing a

natural depiction of pedestrian movements. The

construction of the model is not based on pedestrian

queues or the influence of the density but on the

phenomena that occur when the pedestrians move

naturally.

In both modes, the movement routes between each

mesh are calculated in an area that is divided by a

pedestrians in these routes, reaching critical mass, the

movement routes are recalculated. Through this

process, if a pedestrian bottleneck occurs, some

navigation mesh, and if there are too many

pedestrians choose alternative routes. The

ICAART 2016 - 8th International Conference on Agents and Artiﬁcial Intelligence

recalculation delay of the movement routes can be set

in the simulator parameters. In the SFPE mode, the

movements of the pedestrians are calculated in a

straight line, and in the steering mode, the B-Spline

algorithm is used to make their movements more

smooth and realistic (Figure 2).

Figure 2: Movement paths in SFPE (left) and steering (right)

modes (Thunderhead Engineering, 2009).

3 IMPROVED FLOOR FIELD

MODEL

This paper suggests a CA-based pedestrian model that

can reflect detailed physical phenomena (collision,

jamming, turns, etc.) and improves upon the

limitations of the existing FFM. Specifically, by

adding different sizes, stride lengths, postures, and

sight to the modeled pedestrians, additional factors

that could not be modeled in the FFM, such as

collisions and friction between pedestrians, can be

considered.

3.1 Size and Shape of Pedestrians

The size and shape of each pedestrian, which are

based on (Lim et al., 2006), were set at 50 cm × 30

cm: a rectangle instead of a square. As the FFM

utilizes a cell of 40 cm × 40 cm in size, the pedestrian

in this model was set to fit this size. This study has

decreased the size of the cell to 30 cm × 30 cm, and

the pedestrian no longer occupies one cell but parts of

four cells. Therefore, the pedestrian, sized 50 cm × 30

cm, may occupy all four cells by itself, and depending

on the situation, two pedestrians may share one cell.

Moreover, to place each pedestrian in four cells, the

pedestrians were placed on the inter-cell borders

instead of within the cell. In Figure 3(a), the

pedestrians are placed on the border points, and each

cell is occupied by either one or two pedestrians.

However, limiting conditions were defined to prevent

situations in which physical collisions occur and is

unable to place them properly, as shown in Figure

3(b).

Figure 3: An example of pedestrian shapes and placement:

(a) possible situations, (b) impossible situation.

3.2 Posture

In this paper, “posture” means the direction in which

each pedestrian is facing. The existing FFM does not

include posture information for the pedestrians.

However, the posture of a pedestrian is a very

important factor that influences the direction of

movement, sight, and decisions of the pedestrians.

Therefore, in this model, the pedestrians are given

various postures and face in various directions.

However, if the pedestrians can be modeled to face

any direction (360 degree range), the calculations

become very complex, and therefore the posture

directions were limited to eight directions.

Depending on the placements of nearby obstacles

and other pedestrians, each pedestrian can either turn

freely or not at all. If there are no limitations nearby,

they are able to change their postures freely, but if the

nearby areas are filled with pedestrians, turning

becomes impossible. That is, the postures of the

pedestrians influence each other, and these

interactions influence the overall pedestrian situation.

3.3 Posture Probability

The FFM uses transition probability to determine the

movement directions of the pedestrians. This study

used similar concepts to determine the postures of the

pedestrians; the method used in this study is termed

posture probability. The pedestrian must choose one

direction out of the eight possible directions as the

posture they will be facing. In a normal evacuation

situation, pedestrians would face the exit and move,

but if their sight is limited by factors such as power

outages, their postures could be set to face different

directions. Therefore, the model must be able to

reflect these different scenarios through parameter

control.

Posture probability is a method that determines

which of the eight direction near the pedestrian will

Comparison of Improved Floor Field Model and Other Models

be faced, given random variables. The pedestrians

choose their posture every timestep, and the chosen

posture influences their movement. This model only

allows movement at intervals of 90° to both the left

and right based on the current posture.

Figure 4: Simulation example of the improved model: (a)

beginning stage (t = 0), (b) middle stage, (c) final stage. (d)

Detailed pedestrian situation.

Apart from the previously mentioned factors,

various factors, such as the pedestrian stride sizes and

update rules, were also changed. Figure 4 shows

snapshots from a sample simulation of the improved

model. The pedestrians choose their posture and

movement direction at each timestep, and based on

these choices, they move towards a pre-determined

exit. Figure 4(d) shows that the shape of the

pedestrians is a rectangle, and their shape differs

based on their posture.

4 SIMULATIONS WITH OTHER

MODELS

This study sought to determine the characteristics of

a simulated scenario calculated through the improved

pedestrian model when compared with other models.

Therefore, a comparative analysis was conducted for

the five models described in this study: the improved

model, the existing FFM, Simulex, Pathfinder (SPFE,

Steering). The experimental conditions, methods, and

results are discussed in this section.

4.1 Experimental Conditions

The experiment took place in the first-floor space of

a university campus building. This space has six exits

and is composed of spaces with diverse purposes,

such as lecture rooms and libraries (Figure 5). This

space was selected so that diverse pedestrian

situations could be examined. The exits were

assigned numbers from 1 to 6, moving clockwise

from the top left corner. The dark areas in Figure 5

are close to an exit, and the light areas are far from an

exit. These colors have been assigned based on the

values of the Static Floor Field (SFF).

Figure 5: Experimental area.

The experiment was divided into two parts. The

first part was the comparison of the trends of changes

in the total evacuation time as the evacuating

population increased. The purpose of this part of the

experiment was to compare the influence of

phenomena such as bottlenecks, collisions, and

jamming, which increase with increasing evacuating

population in each model.

In the second part of the experiment, when the

evacuation population was set at a large enough value

to cause bottlenecks (1000), the number of

pedestrians using each exit and the time-series

evacuation situation by exit was compared. This is to

compare the evacuation situations of the different

models at each exit in a detailed manner with the same

evacuation population in each model.

4.2 Comparison of Total Evacuation

Time based on Size of Evacuating

Population

The total evacuation times of the five models were

compared as the evacuation population was varied

from 50 to 1200 people. The population distribution

among the rooms was randomized. Basic parameters

were used for Simulex and Pathfinder; for the FFM

and the improved model, the degree to which the

population wish to move quickly to the exit was set at





parameter (



2,1.5) (Kirchner et al., 2002). In

the improved model, the posture determination

parameter 



was set to be equal to 



. When 



2,

the pedestrians move quickly to the exit, and when





1.5, they move less quickly; therefore, the

evacuation time when 



2 is slightly lower.

The results of the experiment are summarized in

Figure 6. When the size of the evacuating population

increased in the FFM, the evacuation time increased

from approximately 40 to 70 s, showing an increase

ICAART 2016 - 8th International Conference on Agents and Artiﬁcial Intelligence

of approximately 30 s. This is because, the collision

and jamming phenomena do not occur regardless of

the number of pedestrians, and thus the evacuation

time changed very little. Therefore, the situation

identified as the limitations for FFM was confirmed

through the actual experiment results.

Figure 6: Comparison of evacuation times by model.

Considering the two scenarios of the improved

model, when 



2, the results are very similar to

the results of the Pathfinder (Steering) model.

Moreover, when 



1.5, the results are very similar

to the results of the Pathfinder (SFPE) model. The

two Pathfinder models reflect the phenomena of

collisions and jamming between pedestrians, and as

the bottlenecks become more intense, the evacuation

time increases. It can thus be concluded that the

interactions between pedestrians influence the results

of the improved model, similar to Pathfinder.

Therefore, when effects that were excluded in the

existing FFM were considered, the calculations

yielded evacuation times that are very similar to those

obtained by the currently widely used models.

Lastly, Simulex yielded longer evacuation times

than all other models. This is thought to be because

there is a stronger emphasis on the interaction

between pedestrians than in the other models, and the

walking speed was influenced by other factors.

4.3 Comparison of Evacuation

Situation by Exit

The experimental area in Figure 5 includes six exits,

and the width of each exit is approximately 2–3 m. In

this experiment, the evacuating population is set to be

large enough to cause bottlenecks (1000), and the

evacuation situations of the six exits are compared. In

both the FFM and the improved model, only the 





2 scenarios were tested. Figure 7 shows the

evacuating population by exit. In all models, exit 3

had the most people, followed by exit 2 and exits 1, 4,

and 6, which had approximately the same number of

people, and exit 5 had the fewest people.

When regarded on an exit-by-exit basis, in the two

Pathfinder models, approximately 120 and in the

other three models, approximately 90 people left

through exit 1. The other three models utilize the

distance map and SFF to move the pedestrians to the

exits (Thompson et al., 1997; Burstedde et al., 2001).

Each pedestrian selects an exit based on their distance

from the exit, and the spaces near each exit are

calculated in advance. The possibility of the

pedestrian located in each area moving to their

assigned exits becomes very high. Therefore, the

allocated population of these three models show

overall similarities in their movements. However, in

the Pathfinder simulation, if there are many

pedestrians along the route a pedestrian wishes to take,

an alternative route is calculated in real time

(Thunderhead Engineering, 2009). Therefore, there

were pedestrians that were initially assigned to exit 1

as well as other pedestrians who detoured from their

initial routes to different exits, resulting in a larger

outflow of pedestrians through this exit.

The Pathfinder models show fewer people leaving

through exit 2 compared with other models.

According to Figure 7, exit 2 had the second-largest

evacuating population and experienced severe

bottlenecks. Therefore, based on the alternative route

calculations, the bottlenecks occurring at exit 2 led to

pedestrians taking a detour to exit 1, and because the

other three models do not have the ability to calculate

alternative routes, the number of people using this

exit is similar in these models.

In all five models, approximately 400 pedestrians

were assigned to exit 3. Exits 4 and 5 are next to each

other in a hallway (shown in Figure 5). However,

because the hallway is closer to exit 4, the pedestrians

who passed along the hallway tended to decide to use

exit 4. However, in the Pathfinder simulations, when

there were bottlenecks at exit 4, the pedestrians

detoured to exit 5. Therefore, unlike the other models,

approximately 40 evacuating pedestrians used exit 5.

All five models treated exit 6 similarly.

Figure 8 compares the states of evacuation

through exit 1 in the five models. The time-series data

of the cumulative evacuation population are shown

for each model. The slope of the straight line is the

speed of evacuation; for larger slopes, the bottlenecks

have less influence, and for smaller slopes, the

bottlenecks have more influential. Figure 9 shows

snapshots of the evacuation simulation at similar

points in time for all four models except the SPFE

model.

Comparison of Improved Floor Field Model and Other Models

Figure 7: Number of evacuating pedestrians by exit.

Figure 8: Comparison of evacuation situations of exit 1 in

the five models.

Figure 9: Evacuation from exit 1: (a) Simulex, (b)

Pathfinder (Steering), (c) improved model, (d) FFM.

The simulation results in Figure 8 show that the

number of people accumulate most rapidly in the

FFM (broken line); after approximately 30 s, 95

pedestrians were evacuated. For the improved model

and the two Pathfinder models, the evacuation began

at approximately 15 s, after which the evacuation

speed was very similar. As shown in Figure 9(b) and

(c), there are differences in the situation of the

evacuation in the improved and steering models, but

as shown in Figure 8, they are very similar in terms

of evacuation speed.

However, in the Simulex model, the evacuation

process was much slower than in the other models.

Although the number of pedestrians in the Simulex

model was similar to those in the improved model, the

improved model completed evacuation at

approximately 50 s, whereas Simulex ended at

approximately 90 s. The reasons for this are the

structure of this building and the placement of

pedestrians. Figure 9(a) shows the Simulex

evacuation. Exit 1 is located in the bottom right hand

corner, and there is a large group of pedestrians in the

large room. To reach exit 1, the pedestrians must take

a detour. In this process, the collisions and jamming

were more influential than the improved model. As

shown in Figure 9(a), many bottlenecks were created

in the process of moving around the wall, so the speed

of evacuation was slow. However, in the other models,

despite the same structure and situation (Figure 9(b),

(c), (d)), bottlenecks occurred less than Simulex. For

the remaining five exits, the evacuation situation was

calculated differently based on the structure of the

building and the placement of pedestrians. The

existing FFM showed a very high evacuation speed,

the two Pathfinder models and the improved model

showed similar evacuation speeds, and Simulex

showed the slowest evacuation speeds.

5 CONCLUSION

This paper proposed an improved FFM model to

solve the limitations of the existing FFM and

identified the characteristics of the improved model

by comparing it with other models. Because the FFM

models pedestrian with a shape that differs from the

shape of actual pedestrians, certain phenomena, such

as collisions and jamming, cannot be simulated in this

model. Therefore, to improve this model, various

factors were added, including rectangular (rather than

square) pedestrians, pedestrian placement rules, and

postures, and these additions resulted in an improved

model. The improved model was compared with

models currently in use, such as Simulex and

Pathfinder (Steering, SFPE) to confirm that the

improved effects yielded accurate results.

A comparison experiment was carried out in two

parts using a specific space in a campus building.

First, changes in evacuation times resulting from

increasing evacuating population size were studied.

The FFM yielded a very low evacuation time even

when the evacuating population was large. However,

the improved model yielded results very similar to the

ICAART 2016 - 8th International Conference on Agents and Artiﬁcial Intelligence

100

Pathfinder results, and these results were

demonstrated to have overcome the limitations of the

previous model. Moreover, in the second part of the

experiment, in which the evacuation situations were

considered by exit, differences between the models

were observed with respect to the pedestrians

allocated to each exit, depending on the spatial

structure and the situation at the exits.

Overall, the improved model proposed in this

study seems to have overcome the limitations of the

FFM. In the future, it would be beneficial to conduct

further research to incorporate further physical

factors, such as inertia and pace, as well as

psychological factors common in evacuation

situations.

ACKNOWLEDGEMENTS

This research was supported by the Disaster Safety

Technology Development & Infrastructure

Construction Program funded by the Ministry of

Public Safety and Security(“NEMA-Infrastructure-

2014-116”)

REFERENCES

Amor, H. B., Murray, J. and Obst, O. (2006). Fast, Neat,

and Under Control: Arbitrating Between Steering

Behaviors. In S. Rabin(Ed.), AI Game Programming

Wisdom 3, pages 221-232.

Burstedde, C., Klauck, K., Schadschneider, A. and Zittartz,

J. (2001). Simulation of pedestrian dynamics using a

two-dimensional cellular automaton. In Physica

A:Statistical Mechanics and its Applications,

295(3):507–525.

Henein, C. (2008). Crowds Are Made of People: Human

Factors in Microscopic Crowd Models. In Ph.D. thesis,

Carleton University, Canada.

Kirchner, A., Klupfel, H., Nishinari, K., Schadschneider, A.

and Schreckenberg, M. (2004). Discretisation Effects

and The Influence of Walking Speed in Cellular

Automata Models for Pedestrian Dynamics. In J Stat

Mech 10:P10011.

Kirchner, A., Nishinari, K. and Schadschneider, A. (2003).

Friction effect and clogging in a cellular automatation

model for pedestrian dynamics. In Phys. Rev. E, 67,

056122.

Kirchner, A., and Schadschneider, A. (2002). Simulation of

evacuation processes using a bionics-inspired cellular

automaton model for pedestrian dynamics. In Physica

A:Statistical Mechanics and its Applications, 312:260-

276.

Kirik, E., Yurgel'yan, T. and Krouglov, D. (2007). An

intelligent floor field cellular automation model for

pedestrian dynamics. In Proceedings of The Summer

Computer Simulation Conference, pages 1031-1036.

Kretz, T. (2006). Pedestrian Traffic - Simulation and

Experiments. In PhD thesis, University Duisburg-

Essen.

Kwak, S., Nam, H. and Jun, C. (2010). An Indoor

Pedestrian Simulation Model Incorporating the

Visibility. In Journal of Korea Spatial Information

Society, 18(5):133-142.

Lim, W., Ryu T., Choi, H., Choi, H. and Chung, M. (2006).

A Comparision of Gait Characteristics between Korean

and Western Young People. In Journal of the

Ergonomics Society of Korea, 25(2):33-41.

Nelson, H. E. and MacLennan. H. A. (2002). Emergency

movement. In P. DiNenno(Ed.), The SFPE Handbook

of fire Protection Engineering, 3rd Edition, National

Fire Protection Association, Quincy, pages 3-367-3-

380.

Nishinari, K., Kirchner, A., Namazi, A. and

Schadschneider, A. (2004). Extended floor CA model

for evacuation dynamics. In IEICETrans. Inf. Syst, E87-

D:726-732.

Nishinari, K., Kirchner, A., Namazi, A, and

Schadschneider, A. (2005). Simulations of evacuation

by an extended floor field CA model. In Traffic and

Granular Flow ’03, pages 405-410.

Raynolds, C. W. (1999). Steering Behaviors For

Autonomous Characters. In Proceedings of the Game

Developers Conference 1999, Miller Freeman Game

Group, pages 763-782.

Suma, Y., Yanagisawa, D. and Nishinari, K. (2012).

Anticipation effect in pedestrian dynamics:Modeling

and experiments. In Physica A:Statistical Mechanics

and its Applications, 391:248-263.

Thunderhead Engineering. (2009). Pathfinder Technical

Reference. In Thunderhead Engineering Consultants,

Inc., Manhattan.

Thompson, P., Wu, J. and Marchant, E. (1997). Simulex

3.0: Modelling Evacuation in Multi-Storey Buildings.

In Fire Safety Science-Proceedings of the Fifth

International Symposium, pages 725–736.

Varas, A., Cornejo, M. D., Mainemer, D., Toledo, B.,

Rogan, J., Munoz, V. and Valdivia, J. A. (2007).

Cellular automaton model for evacuation process with

obstacles. In Physica A:Statistical Mechanics and its

Applications, 382:631-642.

Yanagisawa, D. and Nishinari, K. (2007). Mean-field

theory for pedestrian outflow through an exit. In Phys.

Rev. E, 76, 061117.

Yanagisawa, D., Kimura, A., Tomoeda, A., Ryosuke, N.,

Suma, Y., Ohtsuka, K. and Nishinari, K. (2009).

Introduction of frictional and turning function for

pedestrian outflow with an obstacle. In Phys. Rev. E,

80, 036110.

Comparison of Improved Floor Field Model and Other Models

101