A Multi-Layer Navigation Approach for Interactive Pedestrian Flow

Simulation in Digital Twins

Christoph Nellinger

, Jan Marius St

urmer

and Tobias Koch

German Aerospace Center, Institute for the Protection of Terrestrial Infrastructures, Sankt Augustin, Germany

Keywords:

Agent-Based Modeling, Pedestrian Flow Simulation, Digital Twin, Coupling Dynamics, Co-Simulation.

Abstract:

Pedestrian ﬂow simulation is crucial for accurately depicting daily activities and dynamics of infrastructures,

such as town halls, train stations, or airports. Current pedestrian ﬂow models often lack the capability to

interact with environmental changes in real-time or only focus on one-directional interactions via prescribed

events. To address this limitation, we propose a hybrid approach that combines graph-based methods for

large-scale navigation with the optimal steps model for small-scale navigation and locomotion of agents. This

combination enables dynamic updates according to environmental changes provided by other simulations. We

demonstrate the effectiveness of our proposed approach in an exemplary airport architecture where pedestrian

simulation is coupled with an electrical simulation, resulting in a successful bidirectional coupling. Speciﬁ-

cally, we consider a scenario where a saboteur agent meddles with an electrical circuit, causing a ripple effect

that impacts pedestrian behavior.

1 INTRODUCTION

In recent years, the concept of digital twins has be-

come increasingly signiﬁcant across various ﬁelds,

offering a new approach to understand and interact

with complex systems. Essentially, a digital twin

serves as a virtual replica of a physical entity or sys-

tem, providing a dynamic and immersive reﬂection of

its real-world counterpart (Grieves, 2015). This con-

nection between the physical and digital realms has

opened up new possibilities for exploration and inno-

vation in diverse sectors with applications in health-

care, urban planning, manufacturing, and infrastruc-

ture (Thelen et al., 2022).

As pedestrian dynamics are an important part of

the operation of different infrastructures, the demand

for computationally efﬁcient pedestrian ﬂow simula-

tions has grown in digital twins. Understanding hu-

man behavior and their interaction with technical sys-

tems is vital for optimizing infrastructure operation.

Moreover, forecasting the impact of incidents on in-

frastructure operation is essential to evaluate contin-

gency plans and to develop concepts for infrastructure

protection.

https://orcid.org/0009-0004-0528-2040

https://orcid.org/0009-0002-1490-6607

https://orcid.org/0000-0003-1279-0209

Current state-of-the-art tools enable us to simulate

and analyze complex phenomena with a high level

of accuracy and detail using prescribed events im-

pacting the simulation. However, there are no tools

available that allow direct real-time bidirectional cou-

pling with other simulations and/or sensors or actua-

tors, and few works have been published on that topic.

However, pedestrian behavior is inﬂuenced by various

environmental factors, including pathway accessibil-

ity, awareness of available routes, weather conditions,

and more. Many of these factors can be effectively

modeled and predicted using established simulation

techniques. Integrating these simulations within a

comprehensive digital twin of a socio-technical sys-

tem provides deeper insights into the dynamics of

pedestrian movement.

Imagine a scenario where a pedestrian simulation

interacts with an electrical simulation to model the

connection between human movement and building

infrastructure. As an employee enters a conference

room, their presence activates an occupancy sensor,

adjusting lighting, heating, ventilation, and security

settings for energy efﬁciency and comfort. However,

maintenance staff or malicious actors who interact

with power boxes can disrupt the system, impacting

not only lighting but also security mechanisms and

pedestrian ﬂow. This includes electrical door fail-

ures that can create bottlenecks and alter evacuation

Nellinger, C., Stürmer, J. M. and Koch, T.

A Multi-Layer Navigation Approach for Interactive Pedestrian Flow Simulation in Digital Twins.

DOI: 10.5220/0013555700003970

In Proceedings of the 15th International Conference on Simulation and Modeling Methodologies, Technologies and Applications (SIMULTECH 2025), pages 295-304

ISBN: 978-989-758-759-7; ISSN: 2184-2841

295

routes.

The tight integration between the pedestrian sim-

ulation and the electrical simulation opens up new

ways of forecasting the system’s behavior under un-

foreseen conditions. By dynamically updating the

navigation model based on environmental changes,

the digital twin ensures optimal pedestrian ﬂow and

accessibility, contributing to the overall resilience and

operational performance of the building.

To achieve acceptable computation times, while

being able to update the navigational model through-

out the simulation, we propose a two-layer naviga-

tion model with a microscopic navigation based on a

variation of ﬂoor ﬁeld cellular automata and a graph-

based macroscopic navigation similar to the works of

Kneidl et al. (2013). By cutting the ﬂoor ﬁelds based

on the rooms of the architecture at hand, we reduce

the amount of necessary computations. By dynami-

cally updating the navigation graph at run time, we

allow for reacting to environmental changes in real-

time.

This paper is structured as follows: Section 2 pro-

vides a brief overview of pedestrian simulation and

highlights studies that explore its coupling with other

simulators. Section 3 outlines the models employed

for each navigation layer and offers a detailed expla-

nation of the agent’s navigation process within the

simulation. Additionally, it describes the interaction

framework and data model used to couple simula-

tions. Section 4 presents the simulation example,

detailing the pedestrian simulation setup, including

geometry and parameter values, and introducing the

electrical simulator as the coupled counterpart. The

results of the corresponding simulation are presented

in Section 5, followed by a discussion of current lim-

itations and potential future improvements in Sec-

tion 6.

2 RELATED WORK

Previous studies incorporating agent-based pedestrian

simulation into digital twins have primarily focused

on trafﬁc interactions (Wang et al., 2023), crowd be-

havior (White et al., 2021), and emergency evacuation

scenarios (Han et al., 2020; Umemoto et al., 2024).

These implementations often rely on fundamental

models such as cellular automata or social force mod-

els. Cellular automata represent pedestrian movement

on a discrete grid, where agents transition between

cells based on predeﬁned local rules, making them

computationally efﬁcient but sometimes limited in re-

alism. In contrast, social force models treat pedestri-

ans as particles inﬂuenced by attractive and repulsive

forces, capturing continuous movement dynamics and

interactions with obstacles or other agents. The opti-

mal steps model (K

oster et al., 2011) bridges the gap

between social force models and cellular automata by

allowing pedestrians to move on a continuous plane

while using a ﬂoor-ﬁeld-based cellular automata ap-

proach to ﬁnd the optimal stepping position.

Other pedestrian movement models aim to en-

hance realism by incorporating microscopic behav-

ior and strategic decision-making. These approaches

not only reﬁne small-scale interactions, such as steer-

ing around congestion, but also integrate high-level

path planning (Seitz and K

oster, 2012; Kneidl et al.,

2013; Asano et al., 2010). Some of these models

use graph-based navigation with A*-based algorithms

(Hart et al., 1968). Such methods are widely applied

in areas like computer games, where they enable the

adaptive movement of non-player characters (Cui and

Shi, 2011).

While bidirectional coupling between pedestrian

simulation and other systems remains an underex-

plored area, a few studies have begun to investigate

this approach. One example couples the Simulation

of Urban MObility (SUMO) with the Unity3D game

engine to study the interaction between pedestri-

ans and connected vehicles (Artal-Villa and Olaverri-

Monreal, 2019). This coupling allows for real-time

interaction and provides a more dynamic understand-

ing of pedestrian behavior in relation to trafﬁc sys-

tems. Another study builds on this by using the

CARLA-SUMO co-simulation framework, integrat-

ing it with the CAVE Automated Virtual Environ-

ment (CAVE) to allow users to interact with the sim-

ulation (Wang et al., 2023). A third study couples

SUMO (Simulation of Urban Mobility) and Vadere

with OMNeT++ to co-simulate pedestrian move-

ment alongside telecommunications network behav-

ior (Schuhb

ack et al., 2019). This integration enables

an analysis of how pedestrian behavior might impact

or be impacted by communication infrastructure, al-

beit with a limitation: the use of discrete-event simu-

lation in OMNeT++ requires that the pedestrian sim-

ulation be restarted with new starting conditions for

every event-based change in the telecommunications

network, which restricts real-time coupling. Despite

these promising efforts, research in this area remains

sparse, with these three studies among the very few

addressing such bidirectional coupling.

3 METHODS

For small-scale navigation, we use the optimal steps

model (Seitz and K

oster, 2012), which is based on

SIMULTECH 2025 - 15th International Conference on Simulation and Modeling Methodologies, Technologies and Applications

296

Figure 1: The black outline depicts the geometry of the

building, as extracted from the ﬂoor plan. The brown ar-

eas are target areas, added subsequently. Differently col-

ored vertices and edges describe the different steps of the

construction of the navigation graph.

the fact that pedestrians naturally move in a discrete

way, step-by-step. On the larger scale, our naviga-

tion graph cuts the underlying geometry of the build-

ing into different sections that are separated by doors.

These separate rooms serve as the natural boundaries

for the calculation of the ﬂoor ﬁelds used on the mi-

croscopic layer. This approach signiﬁcantly reduces

the computational time required for computing ﬂoor

ﬁelds of pedestrian destinations. For many buildings

these structural graphs are tree-shaped, but for more

complex buildings such as train stations, airports or

town halls, these graphs can include circular features

with the possibility to cross different rooms to get to

the same destination.

To elucidate the individual steps of our approach

in detail, we utilize a simpliﬁed example geometry, il-

lustrated in Fig. 1. The geometry represents a small

building composed of four rooms with dimensions of

approximately 4.5 meters by 4 meters, and two hall-

ways of similar size (4.5 meters by 8 meters) con-

nected by doors with widths of around 1 meter. The

walls are represented by black lines, while door areas

are shaded in gray. Three of the rooms have desig-

nated target areas that function as pedestrian destina-

tions (note the three brown rectangles within the up-

per rooms in Fig. 1). Inside the bottom left hallway,

we have an entrance area, and inside the bottom right

hallway, there is an exit area, depicted as rectangles

shaded in brown.

3.1 The Optimal Steps Model

The optimal steps model combines the advantages

of different approaches to model pedestrian move-

ment, namely cellular automata and social force mod-

els. Building upon the cellular automata approach of

oster et al., 2011), the optimal steps model uses re-

pulsive potentials of other pedestrians and obstacles

and travel times to their target destinations to create

left door

bottom door

target area

Figure 2: Navigation ﬁelds for the different points of in-

terest, i.e. doors and target areas, within the middle-right

upper room of our test geometry from Fig. 1.

a ﬂoor ﬁeld and later on chooses the optimal step for

each agent. These two phases are referred to as navi-

gation and locomotion.

3.1.1 Navigation

The ﬂoor ﬁelds are created as scalar ﬁelds represent-

ing the travel times of a wavefront traversing from a

destination through the room of the building at a cer-

tain speed that can be adjusted based on the distance

to walls and other obstacles, to which pedestrians nat-

urally try to keep a distance. Given there are no ob-

stacles on the way, the value of the ﬂoor ﬁeld will lin-

early match the Euclidean distance to the destination.

The propagation of the wave can be described by the

Eikonal equation:

V (⃗x) · |

⃗

∇N (⃗x)| = 1, N(⃗x) = 0 for ⃗x ∈ Z (1)

where V (⃗x)) is the velocity ﬁeld and N (⃗x) is the travel

time to position⃗x of the wave starting at target area Z.

V (⃗x) =

(

0, ⃗x ∈ E

min(1,δ

(⃗x) ·

), ⃗x ̸∈ E

(2)

E denotes the area covered by walls and obstacles and

(⃗x) is the distance between position⃗x and the clos-

est obstacle. The wave cannot pass through walls and

obstacles and the speed of the wave linearly increases

from 0 to 1 within a certain distance d

from these.

If the distance is higher than d

, the wave travels at a

constant speed of 1.

We use the fast marching method (Sethian, 1996)

as implemented in scikit-fmm

to solve the Eikonal

equation and compute these ﬂoor ﬁelds efﬁciently.

scikit-fmm: the fast marching method for Python,

https://github.com/scikit-fmm/scikit-fmm

A Multi-Layer Navigation Approach for Interactive Pedestrian Flow Simulation in Digital Twins

297

Figure 3: Discretization of the circles reﬂected by points.

The three arrows represent six possible positions for the

next step.

The set of three ﬂoor ﬁelds, two for the doors, one for

the target area, within the middle-right upper room is

shown in Fig. 2.

3.1.2 Locomotion

To ﬁnd the optimal next step, the repulsive potential

of agents in close proximity are considered in addition

to the value of the ﬂoor ﬁeld.

(⃗x) = P

(⃗x) +

∑

i=1,i̸=l

p,i

(⃗x) +

∑

j=1

o, j

(⃗x) (3)

(⃗x) is the attractive potential of the target t evalu-

ated at position ⃗x. P

p,i

(⃗x) is the repulsive potential

of pedestrian i and P

o, j

(⃗x) the repulsive potential of

obstacle j, affecting a pedestrian at position ⃗x.

In contrast to a cellular automaton, this model

uses local optimization on one or more circles around

each pedestrian, taking into account a discrete number

of positions (as shown in Fig. 3), while maintaining

movement on a continuous plane by shifting the ori-

entation of the circle between steps. −P is evaluated

as a utility function on positions along the discretized

circle and the original starting position to determine

the optimal next step. In our implementation we use

two circles representing a large step and a small step

(half-distance) and evaluate a total of 32 different po-

sitions per step.

3.2 The Navigation Graph

The construction process of the navigation graph is di-

vided into three steps. First, we generate a ﬂoor graph

consisting of the rooms and doors of the building as

vertices. An edge between a door-vertex and a room-

vertex exists exactly when the door is part of one of

the boundary walls of the corresponding room. This is

the case when a room has a door to the system bound-

ary or the door connects two rooms with each other.

This graph can be obtained from the building’s ge-

ometry, for example, by using computer vision tech-

niques on existing ﬂoor plans. This way we obtain the

red graph in Fig. 1.

Second, we add vertices and edges that represent

areas in which pedestrians take part in certain pro-

cesses, such as queuing up for a ticket purchase or sit-

ting down waiting for a train to arrive. Each of these

areas is represented by a vertex as they are possible

destinations for pedestrians. We then add an edge be-

tween these vertices and the room-vertex representing

the room they are located in. These are the blue dotted

edges in Fig. 1.

Finally, we remove all the vertices representing a

room and add edges between each pair of each of their

neighbors since those are arbitrarily set and not imme-

diately relevant for the navigation of the pedestrians.

We end up with a graph whose vertices are either pos-

sible pedestrians’ destinations or doors they have to

pass through to navigate between destinations.

In our example, this is the graph consisting of

the orange and blue vertices connected by the green

dashed edges. As edge weights of our graph, we use

the travel times used in the optimal steps model.

Macroscopic

Navigation

Layer

Agent gets new target

Yes

Reached

current

target?

Step to next

position

Are there

more intermediate

targets?

Remove current

target from list of

intermediate

targets

Find optimal

next position

(Optimal Steps

Model)

List of intermediate

targets (doors) is

loaded (shortest path in

navigation graph)

Start navigation towards

current target (first intermediate target)

Yes

Participate in

process

Yes

Target area

process

completed?

Microscopic

Navigation

Layer

Figure 4: This ﬂowchart diagram illustrates the complete

navigation process of an agent, from target assignment to

the calculation and execution of individual movement steps.

The two layers of navigation are highlighted in orange

boxes.

SIMULTECH 2025 - 15th International Conference on Simulation and Modeling Methodologies, Technologies and Applications

298

3.3 Pedestrian Routing

The model uses a destination-based routing, i.e.,

when an agent is created within the simulation, they

are assigned a target area. Agents proceed to navigate

towards this target until they reach it. Once the agent

reaches their destination, they might spend a certain

amount of time there, taking part in a process associ-

ated with the type of target area. Agents move freely

inside the target area for the duration of the task. De-

pending on the target area and the associated task, a

speed reduction factor is applied.

Examples of this are spending time buying a ticket

at a ticket counter, performing a certain task at a work

station, walking around in a small shop inside a pub-

lic building or sitting down in a waiting area. Upon

completing their task, at a target area the target will

assign another target area to the agent.

In our simulation, we assign new targets with a

certain probability, which might be derived from an-

alyzing pedestrian data from the modeled building or

selected by using expert knowledge. Once assigned

a new target, the navigation algorithm calculates a

shortest path towards the next target based on the nav-

igation graph and assigns intermediate targets, which

are the doors that have to be passed through. For the

shortest path calculation, the A* search algorithm is

used.

Fig. 4 illustrates the basic navigation approach.

The route choice of a pedestrian is determined at the

time of target assignment. Thus, changes in the nav-

igation graph during the travel time to this target will

not lead to an immediate rerouting. If an agent tries to

pass through a door that has been locked, the area will

trigger a recalculation of the path towards the target.

This recalculation takes into account the new state of

the navigation graph and works for an arbitrary po-

sition on the plane as the starting position, in order

to be able to trigger the rerouting from any position.

This is important if we want to portray events like

an emergency protocol in a public building where an

announcement through speakers to leave the building

immediately is made, leading to a rerouting of pedes-

trians from their current position.

The basic idea of how to use the navigation graph

to reroute from an arbitrary position is to add the po-

sition as a temporal vertex to the navigation graph

with edges connecting it to all doors of the current

room. The edge weights are the values of the ﬂoor

ﬁelds of those doors evaluated at the position of the

agent. Shortest path calculation is then performed on

this temporal graph to determine the new list of inter-

mediate targets. If there is no way to reach the cur-

rently assigned target anymore, a new target will be

assigned. Target choice naturally depends on the use

case for this scenario.

Another possibility of consecutive target assign-

ment for our simulation is giving the agent a full list

of consecutive targets, either at generation or at cer-

tain targets in the building. This is used if there is

a process consisting of different sub-processes in dif-

ferent areas, such as the security scan at an airport, or

if the order of processes an agent will take part in is

already known before the simulation.

3.4 The Data Model and Interaction

Framework

The interaction among simulators is enabled through

a shared data source, structured around a comprehen-

sive and extensible data model as proposed by Franke

et al. (2023). This model is designed to represent indi-

vidual components, such as walls, obstacles, pedestri-

ans, pedestrian target areas and others. Each compo-

nent is characterized by a set of attributes relevant to

its function within the simulations. Some attributes,

such as a Universally Unique Identiﬁer (UUID) or

the position and rotation of the component, are avail-

able in all components. In addition, some compo-

nents deﬁne special attributes, such as the velocity

and destination of pedestrians, or the length and cur-

rent ﬂow of power cables. Communication between

simulators is implemented using an NGSI-LD (Next

Generation Service Interface-Linked Data) informa-

tion model, managed through a FIWARE Orion

con-

text broker. This approach ensures standardized and

efﬁcient data exchange.

Each simulation running simultaneously sub-

scribes to the components actively involved in its pro-

cesses. In the case of the pedestrian simulation, the

subscribed components include individual pedestri-

ans, pedestrian target areas, doors, and the power fuse.

The subscription synchronizes the state of these com-

ponents between the internal data model of the pedes-

trian simulation and the shared data source.

Fig. 5 illustrates the software architecture used to

couple the pedestrian simulation with a generic sim-

ulation of a system that interacts bidirectionally with

pedestrian ﬂows.

4 SIMULATION EXAMPLE

In order to demonstrate the capabilities of our model,

we examine a scenario in which a saboteur disrupts

FIWARE Orion: https://github.com/telefonicaid/

ﬁware-orion

A Multi-Layer Navigation Approach for Interactive Pedestrian Flow Simulation in Digital Twins

299

Figure 5: This diagram illustrates the coupling process between the electrical simulation software and the pedestrian stream

simulation. The FIWARE Orion Context Broker synchronizes the associated data models.

Figure 6: This labeled ﬂoor plan illustrates the airport archi-

tecture used in our simulation example, marking all relevant

stations in the agent’s navigation process. The red line rep-

resents the electrical circuit which connects the fuse in the

maintenance room with the electrical door.

airport operations by removing a fuse. This scenario,

previously used as an example in the introduction,

serves as a test case for our approach. To simulate

this event, we couple our pedestrian simulation with

an electrical simulation using the interaction frame-

work described in Section 3.4. The pedestrian simu-

lation captures the movement dynamics of individu-

als within the airport, while the electrical simulation

models the resulting system failure. The electrical

simulation is introduced in detail in a dedicated sub-

section at the end of this section.

Fig. 6 depicts the simpliﬁed airport departure ter-

minal architecture that is used as input to the simula-

tion. The architecture consists of two connected halls.

Hall 1 includes entrances, check-in counters, several

shops, and a maintenance room. Hall 2 contains the

security check area and exits leading to the gates. The

two halls are linked by two doors, one on the left side

and the other on the right side. The electrical mo-

tors at the doors leading from Hall 1 to Hall 2 are

connected to a fuse box within the target area of the

maintenance room. However, the left door and the

right door are connected to separate fuses.

Thirty minutes of airport operation are simulated

in real-time with one simulation step executed ev-

ery 0.2 seconds. Pedestrians are generated at the en-

trances based on a normal distribution that reﬂects the

expected departure times of their ﬂights (parameters

are provided in Table 1). Each airline has four associ-

ated check-in counters, and passengers select one of

them as their target with equal probability. Agents

are modeled as circles of radius 25cm (Weidmann,

1993). Upon entering, each pedestrian proceeds to

a check-in counter associated with their ﬂight, where

they queue for processing. The check-in process is

represented by a waiting time at the check-in counter.

After completing check-in, they walk to the security

check area and queue for one of the available security

check lines. The security check process consists of

three steps: (1) depositing hand luggage for scanning,

(2) undergoing a personal scan, and (3) retrieving

their hand luggage. Processing times are assumed to

be normally distributed (Schultz, 2010). The param-

eters for the distributions associated with the target

areas are given in Table 2. Once processed, pedestri-

ans move towards the designated exit associated with

their ﬂight’s gate and are removed from the simula-

tion.

At a randomly determined time during the simu-

lation, drawn from a normal distribution with a mean

of 900 seconds and a standard deviation of 300 sec-

onds, an agent of type Saboteur is introduced. This

agent proceeds to the maintenance room in Hall 1 and

activates the target area connected to the fuse box.

The activation of the area represents the removal of

SIMULTECH 2025 - 15th International Conference on Simulation and Modeling Methodologies, Technologies and Applications

300

Table 1: Agent creation parameters.

Flight Agents Mean Std. Dev. Color

Lufthansa 20 300 s 300 s cyan

Eurowings 20 800 s 300 s purple

Air Lingus 20 1200 s 300 s green

Ryanair 20 1500 s 300 s yellow

Table 2: Waiting times at process steps.

Target Area Mean Time Std. Dev.

Check-In Counters 40 s 5 s

Luggage Deposit 25 s 6 s

Security Scan 20 s 5 s

Luggage Withdrawal 30 s 8 s

the fuse. This event triggers the electrical simulation,

which recalculates the working status of all compo-

nents within the affected power circuit. Here, the

switching of the fuse deactivates the motor opening

the left door between Hall 1 and Hall 2 (marked in

red in Fig. 6). Consequently, the left door becomes

unavailable, and remains closed.

The unavailability of the left door prompts the

pedestrian simulation to dynamically recalculate the

shortest paths for all affected agents. Pedestrians orig-

inally planning to use the left door are rerouted to al-

ternative paths, ensuring continuity in the simulation

despite the disruption. The simulation is run 20 times

to compensate for the inﬂuence of random variables.

4.1 Electrical Simulation Model

The electrical simulation model is designed to model

the logical behavior of a power circuit, capturing

whether components are turned on or off. The model

represents the circuit as a tree-like graph, where nodes

(components) are connected by edges (electrical con-

nections). Three key statuses deﬁne the state of each

node in the simulation:

• Power Status: Indicates whether a node receives

power, which is determined by its parent node’s

operational status and power supply.

• Switch Status: Determines whether an external

switch is turned on or off, controlling the ﬂow of

power to the node.

• Working Status: Reﬂects whether a node is ac-

tively functioning, combining the power status

and switch status using logical conjunction.

The simplistic simulation model dynamically re-

calculates the statuses of all subsequent components

in response to changes in switch or main power sup-

ply states. This ensures that the system accurately re-

ﬂects the operational conditions of the power circuit.

off

fuse switch

door motor

(a) before sabotage

(b) after sabotage

Figure 7: Example of a hierarchically structured electrical

circuit as created in the simulation showing the correspond-

ing states before (a) and after (b) the sabotage.

The electric system used in our demonstration

setup and the corresponding states before (a) and af-

ter (b) the sabotage are shown in Fig. 7. The sabo-

teur entering the target area triggers a fuse on the right

tree side, leading to a state change of the right system

branch. Thus, the fuse switch cuts the power supply

of the door motor, leaving the door out of service.

5 RESULTS

The output of the simulation provides a detailed rep-

resentation of the agents’ states throughout the en-

tire simulation time span, including time (frame), x-

coordinate, y-coordinate, viewing direction (angle),

absolute velocity V, velocity in x-direction Vx, and

velocity in y-direction Vy. A visual representation of

this spatio-temporal data is shown in Fig. 8.

The simulation progresses through two distinct

phases: before the manipulation of the fuse (Phase 1,

Fig. 8a), and after the fuse has been removed (Phase

2, Fig. 8b). During Phase 1, pedestrian movement

remains undisturbed, with the left door fully opera-

tional. The Saboteur enters the simulation after 754

seconds and walks directly towards the maintenance

room. In Phase 2, after the saboteur agent removed

the fuse, the electrical simulation updates the power

circuit states, marking the left door as non-functional.

This update is promptly communicated back to the

pedestrian simulation, leading to a re-evaluation of

the navigation graph. As a result, all pedestrians adapt

their routes to avoid the affected area, as evidenced by

the trajectory data presented in Fig. 8b. The number

of agents traversing each door is listed in Table 3, fur-

ther quantifying the impact of the event on pedestrian

ﬂow. While during the ﬁrst phase no agents checking

in at the left check-in counters pass through the right

door, passenger paths are altered during the second

A Multi-Layer Navigation Approach for Interactive Pedestrian Flow Simulation in Digital Twins

301

Table 3: Number of agents traversing each of the doors by

time.

Time Interval Door1 Door2

0s - 983 s 20 6

983 s -1800 s 0 46

Table 4: Duration of upstream and downstream communi-

cation.

Interval Min Max Average

upstream 16 ms 24 ms 21 ms

downstream 24 ms 27 ms 25 ms

phase.

To assess the efﬁciency of the bidirectional com-

munication between the pedestrian and electrical sim-

ulations, we measured the time delay between the

agent removing the fuse and the pedestrian simulation

receiving the updated status of the electrical door.

The total communication delay consists of two

components: Upstream, i.e., the time needed to

pass the information of the fuse removal from the

pedestrian simulation to the electrical simulation, and

downstream, i.e., the time between the reception of

the information in the electrical simulation and the

corresponding update of component attributes of the

electrical door in the pedestrian simulation. The cor-

responding times were tracked using the logs of the

context broker. Measurements across the 20 runs are

shown in Table 4.

Additionally, the duration of each simulation

phase was recorded: Phase 1 lasted an average of

1124 seconds, reﬂecting the normal distribution of the

saboteur’s creation time (mean = 900 s) plus the time

the saboteur needed to reach the power box (average

227 s), while Phase 2 averaged 676 seconds accord-

ingly.

These results demonstrate the responsiveness of

the integrated simulation framework and conﬁrm the

successful integration of the pedestrian ﬂow simula-

tion with the electrical simulation, allowing real-time

interaction between the two domains.

6 DISCUSSION

In this work, a novel multi-layer navigation model has

been proposed. We leverage the advantages of agent-

based modeling to portray the complex behavior of

individuals, while reducing the necessary computa-

tions to achieve a suitable framework for live coupling

with other simulators within digital twins of socio-

technical systems. The results of our simulation sug-

gest a promising representation of pedestrian behav-

ior and their interaction with technical infrastructure

features within a building environment.

We did not conduct any experimental studies to

validate our agents’ behavior. However, scientiﬁc val-

idation of the underlying model has been performed

by von Sivers and K

oster (2015). By expanding these

validated models to include dynamic environmental

updates, we aim to provide a more comprehensive

simulation for studying pedestrian dynamics in com-

plex buildings.

Moreover, our model allows direct real-time cou-

pling between different simulation models, as it is

not based on a predeﬁned event tree like other state-

of-the-art implementations of pedestrian ﬂow simula-

tions, such as Vadere (Kleinmeier et al., 2019).

Additionally, the navigation model employed

within the pedestrian simulation proves advanta-

geous for such interactions, enabling efﬁcient and dy-

namic path recalculations in response to environmen-

tal changes. As shown in the output trajectories, our

model successfully integrates temporal variations in

the environment, such as the locking of a door result-

ing in the temporary inaccessibility of a pathway.

In our current implementation, agents are rerouted

by interacting with the door. However, multiple

pedestrians walking toward the same locked door,

each trying to open it before choosing a new path to-

ward their target, is not realistic behavior. Instead,

the implementation can be improved by allowing one

pedestrian to attempt to open the door and triggering

the rerouting of all pedestrians steering toward this

door who are either in a certain proximity or within

the same room. Different methods of modeling the

available information about the accessibility of tar-

get areas and passages might be applied depending

on the use case. How pedestrians share information

about which passages are open and other navigation-

related details varies greatly depending on the speciﬁc

use case and is a topic that requires further research.

Another issue with our current implementation is

that agents sometimes overstep into a locked door be-

fore realizing it is inaccessible, which leads to prob-

lems. This issue can be attributed to the fact that the

navigation ﬁeld of a door spans both adjacent rooms.

To mitigate this problem in the future, we propose re-

placing single nodes for doors in the navigation graph

with two nodes: one representing each room that the

door connects. By doing so, we will introduce two

separate navigation ﬁelds per door. As a consequence,

inaccessibility can be represented by removing the

edge between these two nodes rather than the nodes

SIMULTECH 2025 - 15th International Conference on Simulation and Modeling Methodologies, Technologies and Applications

302

(a) t = 0 s - 755 s (b) t = 755 s - 983 s

Figure 8: Agent trajectories for different time windows: (a) Until the saboteur removes the power fuse, (b) After the electrical

left door is not operational. The red trajectory shows the path of the saboteur.

themselves, as is currently done. This change should

improve the overall reliability and robustness of our

system while also reducing the likelihood of agents

overstepping into locked doors.

The optimal-steps model does not capture the

fact that people will already adjust their pathing to

steer around congestions of other pedestrians in sight

when they are still far away, unless they estimate that

congestion will dissolve before they reach that area.

Thus, our model is also subject to this limitation. This

could be resolved by incorporating line-of-sight con-

siderations between congestion and the agent, using

its orientation in the navigation graph, as shown in

previous work (Kneidl et al., 2013).

Furthermore, the shown example assumes prior

knowledge of the building geometry for all agents.

In reality, however, not all pedestrians possess this

knowledge and may rely on their sight and informa-

tion signs for their routing decisions at a strategic

level. Proper use of in-between targets and target as-

signment within these in-between targets can help ad-

dress this issue.

Currently, we have not accounted for multiple

ﬂoor levels within a building. However, stairs and

other connection points between ﬂoor levels can be

integrated to enhance our method. Including the real-

time accessibility of elevators connecting different

ﬂoors into the pedestrian ﬂow model might be an in-

teresting application for the presented approach.

7 CONCLUSION

Pedestrian ﬂow simulation plays a crucial role in re-

alistically depicting the daily activities and dynamics

of infrastructure, such as airports and train stations.

However, current models are unable to incorporate the

inﬂuence of external factors by coupling the simula-

tion with other simulators or sensor data.

In this work, we propose a hybrid navigation ap-

proach that enhances the optimal steps model with an

interactive navigation graph while preserving its ad-

vantages. The software architecture we use does not

impose limitations on the number of coupled simu-

lations and enables seamless information sharing be-

tween them. Consequently, other technical systems,

such as water infrastructure, can also be integrated.

We have successfully applied our method using a

minimal example setup. The integration of pedes-

trian ﬂow simulation with electrical simulation repre-

sents a step forward in creating comprehensive digital

twins for buildings, as it expands the scope of scenar-

ios that can be simulated and optimized. Improving

the navigation model and its implementation to accu-

rately handle problems like over-stepping, as well as

expanding it in order to simulate buildings with mul-

tiple ﬂoors will be part of future work.

REFERENCES

Artal-Villa, L. and Olaverri-Monreal, C. (2019). Vehicle-

pedestrian interaction in sumo and unity3d. In New

A Multi-Layer Navigation Approach for Interactive Pedestrian Flow Simulation in Digital Twins

303

Knowledge in Information Systems and Technologies:

Volume 2, pages 198–207. Springer.

Asano, M., Iryo, T., and Kuwahara, M. (2010). Microscopic

pedestrian simulation model combined with a tacti-

cal model for route choice behaviour. Transportation

Research Part C: Emerging Technologies, 18(6):842–

855.

Cui, X. and Shi, H. (2011). A*-based pathﬁnding in modern

computer games. International Journal of Computer

Science and Network Security, 11(1):125–130.

Franke, K., St

urmer, J. M., and Koch, T. (2023). Automated

simulation and virtual reality coupling for interactive

digital twins. In 2023 Winter Simulation Conference

(WSC), pages 2615–2626. IEEE.

Grieves, M. (2015). Digital twin: Manufacturing excellence

through virtual factory replication. White Paper.

Han, T., Zhao, J., and Li, W. (2020). Smart-guided pedes-

trian emergency evacuation in slender-shape infras-

tructure with digital twin simulations. Sustainability,

12(22):9701.

Hart, P. E., Nilsson, N. J., and Raphael, B. (1968). A for-

mal basis for the heuristic determination of minimum

cost paths. IEEE Transactions on Systems Science and

Cybernetics, 4(2):100–107.

Kleinmeier, B., Z

onnchen, B., G

odel, M., and K

oster, G.

(2019). Vadere: An open-source simulation frame-

work to promote interdisciplinary understanding.

Kneidl, A., Hartmann, D., and Borrmann, A. (2013). A hy-

brid multi-scale approach for simulation of pedestrian

dynamics. Transportation Research Part C: Emerging

Technologies, 37:223–237.

oster, G., Hartmann, D., and Klein, W. (2011). Micro-

scopic Pedestrian Simulations: From Passenger Ex-

change Times to Regional Evacuation, pages 571–

576. Springer Berlin Heidelberg.

Schuhb

ack, S., Daßler, N., Wischhof, L., and K

oster, G.

(2019). Towards a bidirectional coupling of pedes-

trian dynamics and mobile communication simula-

tion. EPiC Series in Computing, 66(13):60–67.

Schultz, M. (2010). Entwicklung eines individuenbasierten

Modells zur Abbildung des Bewegungsverhaltens von

Passagieren im Flughafenterminal. PhD thesis, Tech-

nische Universitaet Dresden.

Seitz, M. J. and K

oster, G. (2012). Natural discretization of

pedestrian movement in continuous space. Physical

Review E, 86(4):046108.

Sethian, J. A. (1996). A fast marching level set method for

monotonically advancing fronts. Proceedings of the

National Academy of Sciences, 93(4):1591–1595.

Thelen, A., Zhang, X., Fink, O., Lu, Y., Ghosh, S., Youn,

B. D., Todd, M. D., Mahadevan, S., Hu, C., and Hu,

Z. (2022). A comprehensive review of digital twin

- part 1: modeling and twinning enabling technolo-

gies. Structural and Multidisciplinary Optimization,

65(12).

Umemoto, D., Kikuchi, M., Terui, A., Abe, K., Shimizu,

R., Hirashige, K., Ito, N., and Noda, I. (2024). Urban

scale pedestrian simulation in kobe city center. Artiﬁ-

cial Life and Robotics, 29(2):211–217.

von Sivers, I. and K

oster, G. (2015). Realistic stride length

adaptation in the optimal steps model. In Trafﬁc and

Granular Flow ’13, pages 171–178, Cham. Springer

International Publishing.

Wang, Z., Zheng, O., Li, L., Abdel-Aty, M., Cruz-Neira,

C., and Islam, Z. (2023). Towards next generation of

pedestrian and connected vehicle in-the-loop research:

A digital twin co-simulation framework. IEEE Trans-

actions on Intelligent Vehicles, 8(4):2674–2683.

Weidmann, U. (1993). Transporttechnik der Fußg

anger,

volume 2 of Schriftreihe des IVT Nr. 90. Institut

ur Verkehrsplanung, Transporttechnik, Straßen- und

Eisenbahnbau Z

urich, Erg

anzte Auﬂage edition.

White, G., Zink, A., Codec

a, L., and Clarke, S. (2021). A

digital twin smart city for citizen feedback. Cities,

110:103064.

SIMULTECH 2025 - 15th International Conference on Simulation and Modeling Methodologies, Technologies and Applications

304