Previous Article in Journal
Contrastive Learning with Image Deformation and Refined NT-Xent Loss for Urban Morphology Discovery
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
IJGI
Volume 14
Issue 5
10.3390/ijgi14050197
ijgi-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Evacuation Behavioural Instructions with 3D Motions: Insights from Three Use Cases

by
Ruihang Xie
1,
Sisi Zlatanova
1,*,
Jinwoo (Brian) Lee
1 and
André Borrmann
2
1
Faculty of Arts, Design, and Architecture, School of Built Environment, The University of New South Wales, Sydney, NSW 2033, Australia
2
Chair of Computing in Civil and Building Engineering, School of Engineering and Design, Technical University of Munich, 80333 Munich, Germany
*
Author to whom correspondence should be addressed.
ISPRS Int. J. Geo-Inf. 2025, 14(5), 197; https://doi.org/10.3390/ijgi14050197
Submission received: 19 March 2025 / Revised: 1 May 2025 / Accepted: 8 May 2025 / Published: 8 May 2025
(This article belongs to the Special Issue Indoor Mobile Mapping and Location-Based Knowledge Services)
Browse Figures
Versions Notes

Abstract

:
During emergency evacuations, pedestrians may use three-dimensional (3D) motions, such as low crawling and climbing up/down, to navigate above or below indoor objects (e.g., tables, chairs, and stair flights). Understanding how these motions influence evacuation processes can facilitate the development of behavioural instructions. This study examines the influence of 3D motions through a simulation-based method. This method combines a voxel-based 3D indoor model with an agent-based model. Three use case studies are elaborated upon, considering varying building types, agent numbers, urgency levels, and demographic differences. These case studies serve as exploratory demonstrations rather than validated simulations grounded in real-world evacuation experiments. Our findings are as follows: (1) Three-dimensional motions may create alternative and local 3D paths, enabling agents to bypass congestion, particularly in narrow corridors and confined spaces. (2) While 3D motions may help alleviate local congestion, they may intensify bottlenecks near exits, especially in highly crowded and high-urgency scenarios. (3) As urgency and agent numbers increase, differences in evacuation efficiency between scenarios with and without 3D motions are likely to diminish. We suggest further investigation into evacuation behavioural instructions, including the following: (1) conditional use of 3D motions in different buildings and (2) instructions tailored to different demographic groups. These use cases illustrate new directions for evacuation managers to consider the incorporation of 3D motions.

1. Introduction

Pedestrians can dynamically change their motions during evacuation. In low-urgency situations, they primarily walk upright on floor slabs, stairs, and ramps. Nevertheless, they may also use 3D motions within constrained vertical spaces or under higher urgency levels. For example, video recordings illustrate that pedestrians crawled or bent their bodies to walk below the smoke and jumped up to a chair to move faster in high-urgency situations [1].
A recent study [2] defined 3D pedestrian motions in evacuations as consisting of 3D movements and height-related 3D actions. Three-dimensional movements can be classified into five groups: walking upright on movable or dynamic objects (e.g., tables and chairs), bent-over walking, knee and hand crawling, and low crawling. Three-dimensional actions involve stepping up/down, jumping up/down, and climbing up/down. Understanding how these motions influence evacuation processes can help evacuation managers develop behavioural instructions across diverse scenarios with diverse furniture, equipment, and complex room layouts featuring non-standard stairs and ramps.
Existing research on evacuation behavioural instructions focuses on two key aspects: decision-making and physical movement of individuals [3,4]. First, instructions can be developed to enhance decision-making processes. For example, Ma et al. (2016) [5] examined the potential benefits of effective leadership in crowd evacuations. Their findings indicate that only a few leaders can successfully guide a large crowd of evacuees towards the exit in an enclosed space. Catal et al. (2019) [6] also explored the use of augmented reality technology in the design of an evacuation training game, which was implemented to train employees in navigating to the nearest exit during fire or earthquake evacuations. Second, behavioural instructions can focus on how individuals physically navigate an evacuation scenario. For example, a study [7] investigated pedestrian evacuation strategies under limited visibility, examining different navigational approaches, including walking along walls, following the average movement direction, and tracking the crowd’s average position. Their findings suggest that walking along the wall is more effective in low-density environments, whereas following the average movement direction is more effective in high-density scenarios.
Few studies have been conducted on behavioural instructions related to 3D motions. One study [8] examined the impact of passing over and pushing obstacles on evacuation performance. Their findings suggest that encouraging appropriate obstacle-crossing behaviours near the front of classrooms or along walls in traditional learning environments can enhance evacuation efficiency. Conversely, discouraging obstacle-pushing behaviours and ensuring adequate space near exits in active learning classrooms can further improve evacuation outcomes. However, their study did not differentiate between various 3D motions, such as jumping, low crawling, or walking on elevated objects such as tables. Our recent research has suggested that 3D motions can enhance evacuation efficiency, particularly at lower urgency levels, by providing alternative 3D paths to bypass congestion. However, the benefits of 3D motions diminish as urgency and agent numbers increase. It has also been recommended that evacuation training and exercises should incorporate 3D motions. Despite these insights, existing studies do not consider diverse building scenarios or demographic factors like age and gender diversity. The limited behavioural instructions on 3D motions could limit evacuation performance, underscoring the need for further research in this area.
Thus, this study aims to propose behavioural instructions incorporating 3D motions through three use cases. A simulation-based method is used to explore these use cases. In comparison with other methods, such as empirical evacuation experiments with human participants [9,10,11] or studies using animal behaviour as a proxy for human movement in emergencies [12], evacuation simulation offers a more practical approach to assessing safety conditions. Conducting real-world experiments can be costly, time-consuming, and ethically challenging, especially when accounting for emergency conditions and diverse demographic groups. In contrast, simulation allows evacuation managers to generate diverse ‘what-if’ scenarios through key parameter adjustments, such as crowd density, exit availability, and hazard locations, which enable various analyses of evacuation processes across various conditions and thus support the development of behavioural instructions [13].
A previously developed simulation-based method is used in this study, which combines a voxel-based 3D indoor model with an agent-based model. The voxel-based model enables simulating 3D motions in complex indoor environments. Meanwhile, the agent-based model applies predefined rules to determine whether, when, where, and how agents perform 3D motions. Through three detailed use cases considering varying building types, agent numbers, perceived urgency levels (PULs), and demographic differences, we investigate how 3D motions may influence evacuation processes and propose corresponding behavioural instructions. These case studies serve as exploratory demonstrations rather than validated simulations grounded in real-world evacuation experiments. The study contributes to informing evacuation managers about both immediate response strategies and long-term preparedness and educational measures for pedestrians.
The following section, Section 2, reviews previous work on evacuation training and exercises, evacuation simulation models for 3D motions, and 3D indoor models for evacuation simulations. Section 3 details the method comprising a voxel-based 3D indoor model, an agent-based model, and key assumptions. Section 4 presents case studies, including building selection and the evacuation scenario setup. Section 5 presents results from three key metrics. Finally, Section 6 and Section 7 discuss recommendations for behavioural instructions and propose directions for future work.

2. Related Work

By examining the literature on these topics, we identify concepts, models, approaches, and research gaps in existing evacuation training and exercises, evacuation simulation models, and 3D indoor models. This review establishes the foundation for our investigation.

2.1. Evacuation Training and Exercises

Risk reduction solutions comprise four key phases: mitigation, preparedness, response, and recovery. Mitigation aims to prevent or reduce the impact of emergencies through planning and design. Preparedness enhances the readiness of managers and individuals through training and exercises. Response involves actions to manage and minimise the event’s consequences, while recovery focuses on restoring normal conditions after the event [14,15].
Among these phases, evacuation training and exercises serve as a critical component of preparedness and are mandated by many building standards worldwide [16]. Trainings and exercises are conducted to prepare individuals for evacuation and evaluate the performance of both response strategies and evacuees as part of risk reduction efforts [17]. The literature has highlighted their importance in improving evacuation performance. For example, evacuation exercises conducted twice in the same town in Italy resulted in a 20% reduction in evacuation time, based on the comparison of evacuation times observed between the two trials [18]. Similarly, Feng et al. [19] reported notable improvements in evacuation knowledge and self-efficacy among participants following drills conducted in a virtual environment.
Moreover, evacuation training and exercises help overcome known behavioural issues during evacuation, such as the affiliation to familiar routes inside a building [20] or failure to evacuate due to reduced risk perception [21]. Therefore, behavioural instructions from evacuation managers are essential during training and exercises to instruct pedestrians in both decision-making and physical movement [3,4]. To further strengthen evacuation preparedness, behavioural instructions incorporating 3D motions should be considered and investigated, particularly in complex indoor environments.

2.2. Evacuation Simulation Models for 3D Motions

Several studies have incorporated 3D motions into evacuation simulation models, predominantly utilising social force models, cellular automata (CA) models, and agent-based frameworks. First, adaptations of social force models have been employed to simulate partial 3D motions, such as low crawling in fire and smoke scenarios [22] or stepping over small obstacles like chairs and boxes [23]. CA models have also been used to capture various locomotion patterns, including walking upright, bent-over walking, and knee and hand crawling [24]. Another study [8] extended CA modelling to examine human–obstacle interactions, specifically passing over or pushing obstacles such as desks and chairs. Additionally, agent-based models have been applied to simulate 3D motions, with one study [25] developing a prototype system incorporating low crawling and jumping up/down, while another [26] focused on defining behavioural rules for knee and hand crawling. A review article [1] further highlighted the adaptability of agent-based models, noting their capability to integrate diverse 3D motions into individual agent behaviours with relative ease.
However, recent review papers [1,27] have highlighted significant limitations in the previous models for 3D motions. First, these models were largely restricted to a limited set of 3D motions (e.g., low crawling, knee and hand crawling) due to inadequate representation of the height dimension in 3D space. They did not fully identify navigable spaces above or below indoor objects (e.g., desks and staircases) that could facilitate vertical movement. As a result, evacuation path calculations failed to account for the additional distances introduced by movement in the vertical dimension. Furthermore, pedestrian interactions with indoor objects through different 3D motions were not comprehensively modelled, and the underlying decision-making processes governing such motions remained largely unexplored.
To address the above-mentioned gaps in simulating 3D motions, we present an agent-based model. The proposed model enables evacuation managers to explore a variety of scenarios and pedestrian characteristics, thereby supporting the development of behavioural instructions that incorporate 3D motions into evacuation training and exercises. This paper presents case studies demonstrating how such behavioural instructions can be formulated.

2.3. Three-Dimensional Indoor Models for Evacuation Simulations

A 3D indoor model enriched with detailed semantic, attribute, geometric, and topological information is essential for supporting evacuation simulation. Such models should represent navigable spaces (e.g., rooms and corridors), physical components (e.g., floor slabs, walls and furniture), and openings such as doors and windows [28,29].
The Industry Foundation Classes (IFC), CityGML, and IndoorGML are among the most widely used indoor modelling standards. IFC models, in particular, serve as a valuable data source for evacuation simulation, as they provide information on floor slabs, spaces, and doors, which can be extracted and transformed into navigation models [30,31]. Several researchers [28,32,33,34] have developed 3D indoor models based on IFC models, while one study [35] introduced a semantic framework with obstacle-specific attributes (e.g., bottom area, height and texture) to determine whether objects can be bypassed, removed, or passed over. However, these standard models and related studies have yet to fully capture navigable spaces above or below indoor objects (e.g., desks and stairs) and their integration with other navigable spaces in 3D space.
The concept of a navigable surface has also been used to define areas where agents can move. Existing studies have employed various geometric and structural methods to identify these surfaces, representing them either as navigable meshes or discrete cells. Navigable meshes, widely utilised in studies [36,37,38,39,40], are typically derived from the 2D/2.5D geometry of floor slabs, ramps, stairs, and escalators. Represented as interconnected triangles, they allow pedestrians to navigate seamlessly within a continuous 2D/2.5D space by moving in three directions from any given triangle. However, navigable meshes are not well suited for simulating 3D motions. Since they exclude areas occupied by physical components, such as furniture or overhead structures, these regions remain empty and non-navigable for agents [27].
Discrete cell-based approaches partition indoor spaces into a set of non-overlapping units, typically represented as 2D cells or voxels. Two-dimensional cells have been widely adopted in CA and certain agent-based models, as demonstrated in previous studies [41]. These cells vary in shape and attributes, including circular nodes [19] and square grids incorporating elevation or slope [33]. However, 2D cells are insufficient for simulating 3D motions, such as jumping up or down, as they lack height representation. Voxels extend 2D cells into three dimensions, enabling a more detailed representation of indoor environments. They have primarily been used to model staircases [42,43,44,45,46]. Gorte et al. [32] further utilised voxels to delineate navigable spaces on floor slabs and stairs. Existing research [2,27] highlighted the value of 3D voxel models in fully identifying navigable areas above and below indoor objects and establishing their spatial connectivity.
Recent studies [2,47] introduced automated methods for generating a voxel-based 3D indoor model to support simulating 3D motions. These methods classify indoor spaces into freely navigable spaces (P-spaces), navigable spaces under conditions (C-spaces), and non-navigable spaces (N-spaces). C-spaces represent areas above or below indoor objects where 3D motions are possible under specific conditions. From these classifications, navigable surfaces are identified within P-spaces and C-spaces, while vertical links are established to connect these spaces and their corresponding surfaces across different heights. The resulting 3D indoor model is well-suited for simulating 3D motions.

3. Method

This study employs a simulation-based method to investigate how 3D motions influence evacuation processes in diverse scenarios and to provide relevant behavioural instructions. This method builds on previous studies [2,47], integrating a voxel-based 3D indoor model, which captures navigable surfaces, vertical links, and indoor objects, with an agent-based model that governs 3D motions through predefined behavioural rules. Together, this method enables the simulation of various types of 3D motions. To isolate the effect of 3D motions, we make five key assumptions. The following subsections describe the method in detail.

3.1. Simulation Models

As mentioned above, the method to simulate 3D motions consists of two primary components:
  • A voxel-based 3D indoor model that consists of indoor objects, navigable surfaces, and vertical links.
  • An agent-based model that specifies pre-defined rules governing different stages to perform 3D motions.
A voxel-based 3D indoor model has been created based on distinct types of 3D motions [2,47]. As shown in Figure 1a, navigable surfaces allow agents to evacuate by moving above or below indoor objects, such as floor slabs, tables, chairs, and stair flights. The model also incorporates vertical links, which connect spaces and surfaces at different heights, enabling agents to move up or down, like jumping up to or down from a table. Additionally, specific voxels within the surfaces and links, known as transfer and anchor voxels, serve as starting and ending points for 3D actions. Voxel connectivity follows a 10-adjacency model: six face-connected neighbours in 3D space and four edge-connected neighbours on a 2D surface (see Figure 1b). Each navigable voxel has two states: occupied (by agents) or unoccupied. This capability fills a critical gap left by traditional 2D/2.5D mesh-based and cell-based models, which often struggle to handle 3D spatial relationships, and ensures more realistic simulation of pedestrian motions in 3D space.
Additionally, an agent-based model is being investigated by us to determine whether, when, where, and how agents perform 3D motions. The model consists of four groups of rules that control agent motions: initialisation rules, decision-making rules, indoor-agent interaction rules, and agent motion rules. The initialisation rules define the model’s foundational parameters, including agent footprints, motion types, and speed characteristics. Pedestrian shapes in different 3D motions are abstracted into square footprints, and PULs are introduced as core parameters to control agents’ motivation and desired speeds. The decision-making rules specify whether and when each agent would be triggered to perform 3D motions. These decisions are influenced by local density, visibility, individual motion speed, and the presence of specific indoor objects. The indoor-agent interaction rules extend the classical social force model [48] to simulate micro-interactions in 3D space. These include repulsive and attractive forces exerted between agents and indoor objects across both horizontal and vertical dimensions. The inclusion of vertical forces is particularly critical for modelling 3D actions such as stepping up/down and jumping up/down. The agent motion rules determine the specific 3D motions performed by an agent and the manner in which they are executed. These rules govern transitions between different navigable surfaces through vertical links and incorporate vector field guidance to simulate 3D motions and complex 3D paths.
Our simulation method incorporates a comprehensive set of behaviours tailored to simulate 3D motions. These include not only 3D movements (e.g., crawling and bent-over walking) and 3D actions (e.g., climbing up/down and stepping up/down) but also decision-making processes that determine whether and when these motions are triggered. Furthermore, we explicitly account for microscopic interactions between pedestrians and indoor objects. These interactions are modelled based on the psychological tendency of pedestrians either to maintain distance from other pedestrians and objects or to move towards goals. In addition, behaviours related to exit choice and collision avoidance based on the shortest paths and vector fields have been integrated into the model, enabling agents to make global evacuation path decisions and adjust their local paths in response to their surroundings. Regarding exit capacity, we do not impose specific numerical constraints on doors or exits. Instead, the original geometries—such as the width and height of openings—are preserved, thereby naturally reflecting the physical limitations of exits during congestion.
Evacuation simulation models have been used to anticipate where, when, and why adverse events emerge, allowing for evaluating evacuation status in indoor environments and serving as a foundation for architectural design and emergency management choices, proposals, and planning [49]. Such models can further support the development of behavioural instructions, particularly those incorporating 3D motions, to improve preparedness and response in complex indoor environments [1].
Experimental or realistic evacuation datasets on 3D motions in emergencies are highly scarce [1]. There are ethical restrictions on conducting evacuation experiments involving real emergencies. Moreover, a single dataset is often the only reference for studying an individual scenario. Thus, we investigate use cases to explore how the developed models and approaches can investigate the potential influence of 3D motions and inform evacuation behavioural instructions. These case studies are intended as exploratory demonstrations rather than validated simulations.

3.2. Assumptions Made for the Simulations

To ensure the focus remains on the effect of 3D motions, five key assumptions are made for the simulation-based investigation. These assumptions are defined to reduce complexity while maintaining the validity of the results within the defined scope of the study, as outlined below:
  • All simulated agents are assumed to possess prior knowledge of the building layout, including the locations of exits and corridors. This eliminates the need for agents to explore indoor environments, reducing variability due to learning or hesitation [49].
  • The simulations exclude the impact of specific disaster scenarios (e.g., fires, earthquakes, or toxic gases). This prevents external hazards from influencing agent behaviours, such as reduced visibility due to smoke or altered motion paths from structural damage [43].
  • Agents are assumed to evacuate immediately at the start of the simulations without delays due to decision-making, hesitation, or external influences [50]. This ensures that evacuation time is determined solely by agent motions, particularly 3D motions.
  • Agents make independent evacuation decisions without instructions from evacuation managers or external systems. This prevents external interventions from influencing path selection, ensuring a focus on autonomous navigation [50].
  • Agents are assumed to use the shortest available path to the nearest exit along predefined navigable routes [51]. This ensures that improvements in evacuation efficiency result from local 3D motion strategies rather than global path or exit adaptions.

4. Three Use Cases

We use three case studies to explore how 3D motions may influence evacuation processes across various conditions, including building types, agent numbers, PULs, and demographic differences. The findings inform behavioural instructions and offer insights into how 3D motions can be integrated into evacuation training and exercises to enhance efficiency and safety.

4.1. Building Scenario Selection

Three criteria are used to select representative building scenarios of varying complexity and scale:
  • Building function: selecting buildings with diverse functions enables an exploration of how 3D motions may respond to distinct spatial configurations.
  • Building scale: incorporating buildings of different scales enables an assessment of how 3D motions may affect evacuation in small and large buildings. This variation provides insights into behavioural instructions for different building scales.
  • Crowd sizes: buildings should involve a high density of pedestrians concentrated within a single room or space rather than being dispersed randomly across different rooms. This setup allows the investigation to focus on the effects of 3D motions while minimising the influence of other factors such as exit choices.
Based on these criteria, we selected three IFC models. IFC models were used as they offer a broad range of geometric and semantic information on building elements [52], which support the creation of voxel-based 3D indoor models. Figure 2 shows the selected buildings and their voxel-based models: a restaurant, a teaching building’s hall, and a canteen. The restaurant and canteen were retrieved from the public information model repository (http://openifcmodel.cs.auckland.ac.nz/), accessed on 21 December 2024, while the hall was generated from architectural drawings to accurately reflect its spatial characteristics. It is important to note that any IFC model containing detailed 3D building information can be an input for creating a corresponding voxel-based 3D indoor model.
Furniture (e.g., desks, tables, and chairs) was added to the models to enable the simulation of 3D motions. The furniture was arranged according to the buildings’ functions and basic pedestrian flow patterns. With the developed method, navigable surfaces and vertical links were identified in the three IFC models. A voxel size of 15 cm was adopted based on a prior evaluation study [53], which balances computational efficiency and spatial accuracy.

4.2. Evacuation Scenario Setup

To develop behavioural instructions that incorporate 3D motions, three use cases were defined and classified into ‘with 3D motions’ and ‘no 3D motions’ scenarios, considering different agent numbers, PULs, and demographic distributions.
In the ‘with 3D motions’ scenario, agents could perform different 3D motions. The ‘no 3D motions’ scenario functioned as the control scenario, where agents were restricted to walking on floor slabs, while all other conditions remained identical to the ‘no 3D motions’ scenario. The two scenarios provide a baseline for a direct comparison of model output metrics. Across both scenarios, agent numbers, demographic distributions, and PULs were used to define nine sub-scenarios in which 3D motions may influence evacuation processes, as described below:
(1)
Use case 1: Restaurant evacuation
A small indoor restaurant with a single exit and seating arrangements with tables, chairs, and partitions poses distinct evacuation challenges. The layout can obstruct evacuation paths, requiring individuals to navigate around or across obstacles to reach the exit. Three different agent numbers are considered for the restaurant:
  • Small group (16 agents): representing a low-occupancy scenario.
  • Medium group (30 agents): simulating a more typical occupancy level during dining hours.
  • Large group (50 agents): representing a high-density scenario typical of weekends or holidays, characterised by increased congestion and motion restrictions.
This case provides insights into whether and how behavioural instructions should promote or discourage the use of 3D motions in confined spaces. The small group consists of 16 agents, ensuring that each gender and age group is represented according to the following defined demographic distribution. This selection prevents any category from being excluded.
(2)
Use case 2: Hall evacuation
A medium-sized indoor hall, such as a lecture theatre, event space, or conference hall, poses distinct evacuation challenges due to the presence of rows of seating, staircases, and elevated platforms. Three agent group sizes are considered for the hall:
  • Small group (50 agents): simulating a moderate occupancy level during an event.
  • Medium group (90 agents): simulating a well-attended gathering where crowd motions become more complex.
  • Large group (130 agents): modelling a densely populated event with potential bottlenecks at exits.
This case examines whether 3D motions improve evacuation efficiency in medium-sized spaces and whether exit designs should include features that support controlled 3D motions to enhance evacuation efficiency. This provides insights into behavioural instructions for exits and crowd control measures in public buildings.
(3)
Use case 3: Canteen evacuation
A large indoor canteen with numerous tables and service counters poses significant evacuation challenges due to high occupant density. Three agent group sizes are considered for the canteen:
  • Small group (150 agents): representing a low-occupancy scenario typical of off-peak hours.
  • Medium group (200 agents): simulating a typical dining period with a high but manageable occupancy.
  • Large group (250 agents): representing a high-density scenario where congestion significantly affects evacuation dynamics.
This case assesses whether 3D motions provide a practical advantage in large spaces and high-density evacuations and whether exit strategies should include behavioural instructions regulating 3D motions. Findings can inform behavioural instructions for large indoor spaces, helping reduce bottlenecks and delays at exits.
Three PULs, i.e., 0.3, 0.5, and 0.7, were incorporated in each sub-scenario to simulate varying urgency levels. PUL values range within (0, 1]. Higher values correspond to greater urgency, whereas lower values indicate non-urgent conditions. Low urgency (PUL = 0.3) represents calm evacuation, where agents move deliberately. Medium urgency (PUL = 0.5) represents a moderate urgency level, where agents exhibit heightened but controlled urgency. High urgency (PUL = 0.7) represents evacuation with significant pressure, where agents evacuate at a faster pace. These simulations provide critical insights into how 3D motion-based behavioural instructions can be tailored to different urgency levels.
Additionally, demographic differences were incorporated into each sub-scenario to represent a general population mix, ensuring the simulations reflect diverse agent characteristics. A gender ratio of 50% male and 50% female agents was maintained to achieve gender-related balance. Age groups were proportionally distributed, considering the functions of the selected buildings, based on the following assumptions, with 20% adolescents (12–18 years), 70% adults (19–59 years), and 10% older adults (60–70 years).
The starting locations of all agents were randomly assigned in each scenario. Parameter values in the agent-based model were determined based on prior studies [28,54,55,56,57,58,59,60,61], relevant assumptions, and an extensive trial-and-error process through observations of 3D motion patterns, with visualisations used to refine the representation of pedestrian behaviours. These parameters control the decision-making process, mobility constraints and capabilities, and agents’ interactions during simulations. A comprehensive list of parameter values, along with their sources and justifications, is provided in Appendix A. A total of 35 parameters were selected. For example, 12 parameters pertain to movement speeds, while 3 parameters define the likelihood of performing 3D motions. Note that the value of these parameters could be user-defined, specified by research or based on a pre-determined distribution.
Each simulation scenario was repeated 15 times to mitigate random variations and enhance the statistical robustness of the findings. Following each simulation run, the following key performance metrics were recorded: (1) number of agents performing 3D motions, (2) number of agents moving above, and (3) total evacuation time (TET): the duration from the onset of evacuation until the last agent exits the building.

5. Results

Unity and Microsoft Visual Studio 2022 were used to develop a simulation prototype. The simulations were conducted on a system equipped with an Intel Core i7-12700K@5.00 GHz CPU, NVIDIA GeForce RTX 3080 GPU, and 32 GB of RAM. The simulation results are presented below.

5.1. Number of Agents Performing 3D Motions

Across all scenarios, an increasing occurrence of 3D motions is observed as PULs rise. This trend is particularly pronounced in buildings with larger agent numbers.
Figure 3a shows the number of agents performing 3D motions in the restaurant. At a PUL of 0.3, the large group has the highest participation (13 agents), whereas the medium and small groups show lower engagement in 3D motions. At a PUL of 0.5, the large group experiences a significant increase (21 agents), while the medium and small groups decline by one agent each. At a PUL of 0.7, the large group stabilises (20 agents), while the medium group shows a slight increase, and the small group ceases to perform 3D motions. Furthermore, Figure 3b illustrates the demographic breakdown of agents performing 3D motions. Adult males account for nearly all 3D motions across all PULs, while adolescents and older adults rarely perform 3D motions. Even at higher PULs, this pattern persists, with adult females (5 agents) participating only in the large group.
Figure 4a shows the number of agents performing 3D motions in the hall. This scenario exhibits a higher occurrence of 3D motions than the restaurant. At a PUL of 0.3, the large group records the highest engagement in 3D motions (11 agents). As urgency rises to 0.5, the number of agents performing 3D motions increases across all groups, with the large group reaching 17 agents. At a PUL of 0.7, 3D motion engagement stabilises, showing slight declines in the medium and large groups. In addition, Figure 4b illustrates the demographic breakdown of agents performing 3D motions in the hall. Unlike in the restaurant scenario, adolescents exhibit minimal engagement in the hall, while older adults demonstrate some participation, particularly at higher urgency levels. At PULs of 0.5 and 0.7, older adults begin to adopt 3D motions in the medium and large groups, maintaining a small yet consistent presence. However, adult males continue to be the primary adopters of 3D motions.
Figure 5a illustrates the number of agents performing 3D motions in the canteen. The canteen exhibits the greatest occurrence of 3D motions among the three buildings. At a PUL of 0.3, the large group already records 35 agents performing 3D motions, while the medium and small groups record 28 and 24 agents, respectively. At a PUL of 0.5, all groups experience significant increases, with 84 agents in the large group. At a PUL of 0.7, 3D motion occurrence continues to increase, with 91 agents in the large group, 65 in the medium group, and 45 in the small group. These findings suggest that agents increasingly rely on 3D motions in highly crowded environments in response to rising urgency. Moreover, Figure 5b illustrates the demographic breakdown of agents performing 3D motions in the canteen. Compared to the restaurant and hall, the canteen exhibits greater diversity in 3D motion occurrence across age and gender groups. While adult males still dominate, older adults and adolescents begin participating at higher PULs. At a PUL of 0.5, older adults in the large group (19 agents) and adolescent males (5 agents) adopt 3D motions. At a PUL of 0.7, adolescents (both male and female) and older adults reach their highest participation rates.

5.2. Number of Agents Moving Above

A significant portion of agents performing 3D motions do so by moving above movable objects rather than below them. As the number of agents moving below is relatively lower, the analysis focuses on 3D motions above movable objects.
As shown in Figure 6a, the restaurant scenario exhibits relatively low engagement in 3D motions. At a PUL of 0.3, 12 agents in the large group perform 3D motions to move above movable objects, while the medium and small groups show fewer cases (5 and 2 agents, respectively). As the PUL increases to 0.5, the large group reaches 20 agents, but engagement remains minimal in the other groups. At a PUL of 0.7, 19 agents in the large group continue moving above, while the medium group sees a slight increase (2 agents), and the small group ceases engaging in such motions. The specific 3D motions used to move above in this scenario include climbing up/down, jumping up/down, and walking up above movable objects, whereas moving below is limited to low crawling.
Figure 6b suggests that the hall exhibits a more widespread use of 3D motions to move above movable objects than the restaurant. At a PUL of 0.3, 10 agents utilise moving above in the large group, with the medium and small groups showing slightly lower engagement (10 and 7 agents, respectively). As the PUL increases to 0.5, the large group reaches 13 agents, while the medium and small groups also increase. At a PUL of 0.7, engagement stabilises, with 11 agents in the large group and 11 and 8 agents in the medium and small groups, respectively. The specific 3D motions observed in the hall include jumping up/down and bent-over walking on movable objects, while low crawling remains the sole below-object movement.
The canteen exhibits the highest engagement with moving above (see Figure 6c). At a PUL of 0.3, 32 agents in the large group perform 3D motions to move above movable objects, while the medium and small groups show 24 and 23 agents, respectively. As the PUL increases to 0.5, the large group reaches 73 agents, with the medium and small groups increasing to 54 and 36, respectively. At a PUL of 0.7, engagement continues to rise, with 78 agents in the large group, 58 in the medium group, and 40 in the small group. The 3D motions observed in the canteen primarily involve jumping up/down and walking up on movable objects, with low crawling as the only observed below-object motion.

5.3. Total Evacuation Time

We examined the effect of 3D motions on TET across different building scenarios, agent numbers, and PULs. Notably, the only difference between the ‘with 3D motions’ and ‘no 3D motions’ scenarios is the activation of 3D motions. The evacuation simulation model and parameter values remain identical across both scenarios, except for the minimum PUL threshold that determines 3D motion activation. The results suggest that the influence of 3D motions varies across the evacuation scenarios. While 3D motions generally lead to faster evacuation in some cases, their impact is not uniform across all conditions.
In the restaurant, slight variations in TETs are observed (see Figure 7a). For the small group, TETs remain nearly identical across all PULs, with 3D motions offering no significant advantage (e.g., at 0.3 PUL, times are 17.44 s with 3D motions and 17.52 s without). The medium group exhibits marginal improvements when 3D motions are enabled, particularly at higher PULs. Similarly, 3D motions lead to significantly shorter evacuation times in the large group at a PUL of 0.3 (33.38 s vs. 36.06 s), while 3D motions slightly benefit faster evacuations at a PUL of 0.7 (13.52 s vs. 14.28 s). These results indicate that in this relatively small and confined restaurant, 3D motions may not significantly improve evacuation efficiency when the agent count is low and urgency is high.
In the hall, 3D motions generally improve evacuation efficiency, particularly at higher PULs (see Figure 7b). At a PUL of 0.3, the small group experiences faster evacuation with 3D motions (44.7 s vs. 46.64 s), and the medium and large groups show a similar trend. At 0.5 PUL, 3D motions continue to provide a time advantage, while the large group shows the least noticeable difference (26.85 s vs. 26.42 s). However, at 0.7 PUL, the benefit diminishes slightly, with the medium and large groups showing marginal differences (17.2 s vs. 18.42 s and 17.75 s vs. 18.08 s). These results suggest that in larger environments, 3D motions help mitigate congestion, particularly under lower PULs, but their effectiveness decreases at PULs.
Figure 7c suggests the canteen shows the most complex relationship between 3D motions and evacuation efficiency. At 0.3 PUL, 3D motions generally lead to faster evacuations in the small and large groups, reducing TETs by up to 5% (e.g., 64.63 s vs. 68.26 s for the small group and 87.33 s vs. 92 s for the large group). At 0.5 PUL, the impact of 3D motions varies, with agent groups experiencing minimal differences. Notably, at 0.7 PUL, the large group exhibits a slight increase in TET with 3D motions (33.19 s vs. 29.92 s), indicating that in highly crowded environments, 3D motions may introduce additional complexity that does not always lead to faster evacuations. Given the findings, a key question arises: why do 3D motions lead to slower evacuations in large agent groups at high urgency levels?
To answer this question, we focus on the canteen with 250 agents at 0.7 PUL to examine the influence of 3D motions on evacuation processes. Figure 8 presents speed variation maps for both scenarios under 3D motion and no-3D motion conditions. For the speed variation maps, the maximum speed achieved by any agent within each scenario was identified as a reference value, and all agent speeds were expressed as percentages of this scenario-wide maximum. By comparing the maps, we can observe the areas labelled with black circles. Three-dimensional motions introduce alternative 3D paths, enabling agents to move above or below tables to bypass congested areas. This flexibility helps alleviate bottlenecks in narrow corridors, leading to a more distributed motion pattern and reducing local congestion. However, while these 3D paths improve flow efficiency in some regions, they also redirect agents towards exits more rapidly, leading to heightened congestion and slower speed around exit points compared to the no-3D motion condition. This suggests that while 3D motions can enhance spatial navigation and mitigate localised congestion, their overall effectiveness is influenced by exit capacity and crowd distribution. Figure 9a,b depict screenshots of a simulation of the canteen scenario where agents climbed up to a table, proceeded with bent-over walking, and climbed down. In contrast, one agent performed low crawling below the table. Figure 9c,d present agent trajectories with speed variations around the table in top and 3D views.

6. Discussions

We applied three use cases to explore how the developed models can be used to investigate the influence of 3D motions on evacuation processes and to inform evacuation behavioural instructions. These case studies serve as exploratory demonstrations rather than validated simulations of 3D motions with real-world evacuation experiments. The findings are as follows: (1) It is possible that 3D motions may provide alternative local 3D paths, enabling agents to navigate above or below movable objects and bypass congestion, particularly in narrow corridors and confined spaces with furniture. However, whether 3D motions are triggered may vary based on the indoor environment, agent numbers, and urgency levels. (2) The effect of 3D motions on TET may be diverse. It is hypothetical that while 3D motions may alleviate congestion in certain areas, they may also cause agents to move more quickly towards exits, which can increase congestion near exits, particularly in large groups at high urgency levels. (3) As urgency levels and agent numbers increase, differences in evacuation efficiency between 3D and non-3D motion conditions are likely to become less pronounced. This is likely due to increased congestion near exits, which may limit the benefits of 3D motions in high-density scenarios.
While these findings offer initial insights, they are derived from case studies and thus should be interpreted as conceptual explorations rather than definitive conclusions. Based on these exploratory findings, we suggest that future research further explores behavioural instructions, including the following:
  • Conditional use of 3D motions: Pedestrians may be advised to prioritise 3D motions only when necessary to bypass localised congestion (e.g., narrow corridors or areas with furniture) rather than indiscriminately using them, as this may worsen exit congestion. Specifically, the following should be considered:
    • In confined indoor environments with few movable objects (e.g., restaurants and pubs), 3D motions may be used sparingly, as their benefits are possibly limited except in scenarios with higher urgency and a larger number of pedestrians. Instead, directing pedestrians towards unobstructed exits may be more appropriate.
    • In moderately complex indoor environments with a certain level of pedestrian number (e.g., halls and stores), it is possible that 3D motions may help navigate temporary congestion points, particularly near furniture.
    • In large, high-density environments (e.g., canteens and exhibition centres), 3D motions are likely to alleviate local congestion, but exits may be carefully managed to prevent severe congestion at exits.
  • Tailoring behavioural instructions to demographics as follows:
    • It is possible that adults may benefit from targeted training to improve their motion capability and minimise unnecessary congestion near exits.
    • Further investigation is warranted into whether older adults need to be provided with clear evacuation paths that reduce the use of 3D motions for possibly safer and more accessible evacuation routes.
    • Whether adolescents require training on when and how to effectively use such motions may need exploration to enhance evacuation efficiency without compromising safety.
    • Further investigation is needed into evacuation behaviours of parents with children and of groups such as couples or friends to prioritise safer routes and reduce the reliance on 3D motions.
By investigating these behavioural instructions, it is possible that evacuation managers can leverage the benefits of 3D motions while mitigating potential drawbacks. We acknowledge that some values assigned to parameters in the simulation models within the case studies were based on assumptions and tests due to the limited availability of empirical data. These parameter values can be refined through various empirical methods, including real-time data collection using sensors, smartphones, or cameras during real-world emergency evacuations, evacuation drills simulating emergency conditions, and controlled laboratory experiments designed to measure the characteristics of 3D motions. Data collection can also consider different demographic groups, such as adults, adolescents, and children. Repeated trials under varying spatial configurations (e.g., narrow corridors, stairs and obstacles) and environmental conditions (e.g., low visibility and smoke presence) can further enhance the accuracy and representativeness of the collected parameter values.
However, even if empirical data were available from a single emergency event, its applicability to future evacuation scenarios would be inherently constrained due to variations in spatial configurations, furniture placement, and diverse individual behaviours. Nevertheless, our method provides a flexible framework that allows evacuation managers to adjust parameter values and simulate various conditions, enabling them to test different scenarios and develop adaptive behavioural instructions.

7. Conclusions

This study formulates evacuation behavioural instructions incorporating 3D motions through three use case studies. It outlines the key assumptions to ensure a controlled focus on the effect of 3D motions. It then presents the criteria employed to select representative building scenarios, including building function, building scale, and crowd sizes. Subsequently, three use cases are introduced—restaurant, hall, and canteen—each with different agent numbers, PULs, and demographic groups. For each use case, evacuation scenarios are designed with ‘3D motions’ and ‘no 3D motions’ conditions to enable comparative analysis. Detailed configurations of building setups, agent distributions, and simulation parameters are provided, and key metrics such as the number of agents performing 3D motions, the number of agents moving above, and TET are recorded and analysed. The results demonstrate that 3D motions may improve evacuation efficiency by offering alternative local 3D paths, particularly in scenarios with narrow corridors, junctions, and confined spaces with furniture. However, exit congestion may mitigate the benefit of 3D motions. Based on these findings, the study suggests evacuation behavioural instructions that incorporate 3D motions. Future research building on these findings and addressing the study’s limitations could explore the following directions:
  • Conduct real-world experiments and collect empirical data regarding the parameters used in the agent-based model to enable in-depth validation.
  • Integrate with dynamic exit/path choice models and perform simulations in more complex and diverse indoor environments, such as multi-level buildings.
  • Investigate how age, gender, mobility constraints, and psychological stress levels affect the likelihood of individuals using 3D motions during evacuations.

Author Contributions

Conceptualisation, Ruihang Xie, Sisi Zlatanova, and Jinwoo (Brian) Lee; Data Curation, Ruihang Xie; Formal Analysis, Ruihang Xie and Sisi Zlatanova; Investigation, Ruihang Xie and Sisi Zlatanova; Methodology, Ruihang Xie, Sisi Zlatanova, Jinwoo (Brian) Lee, and André Borrmann; Resources, Ruihang Xie and Sisi Zlatanova; Software, Ruihang Xie; Supervision, Sisi Zlatanova, Jinwoo (Brian) Lee, and André Borrmann; Validation, Ruihang Xie and Sisi Zlatanova; Visualisation, Ruihang Xie; Writing—Original Draft, Ruihang Xie; Writing—Review and Editing, Ruihang Xie, Sisi Zlatanova, and André Borrmann. All authors have read and agreed to the published version of the manuscript.

Funding

This research received no external funding.

Data Availability Statement

The data presented in this study is available on request from the corresponding author.

Acknowledgments

This research is supported by the University International Postgraduate Award and HDR Essential Costs of Research Funding Support from the University of New South Wales, Sydney, Australia.

Conflicts of Interest

The authors declare no conflicts of interest.

Abbreviations

The following abbreviations are used in this manuscript:
3DThree-dimensional
PULsPerceived urgency levels
CACellular automata
IFCIndustry Foundation Classes
TETTotal evacuation time

Appendix A

Table A1. Parameters and their assigned values for the simulations.
Table A1. Parameters and their assigned values for the simulations.
ParametersQuantityValues for Adolescents
Male/Female		Values for Adults
Male/Female		Values for Older Adults
Male/Female
v m a x i , u Maximum speed of walking upright4 m/s/3 m/s5 m/s/4 m/s3 m/s/2 m/s
v m i n i ,   u Minimum speed of walking upright1 m/s/0.5 m/s1 m/s/0.8 m/s0.5 m/s/0.5 m/s
v m a x i ,   b Maximum speed of bent-over walking3 m/s/2 m/s4 m/s/3 m/s1.5 m/s/1 m/s
v m a x i , k Maximum speed of knee and hand crawling1.5 m/s/1 m/s2 m/s/1.5 m/s1 m/s/0.5 m/s
v m a x i , l Maximum speed of low crawling1 m/s/0.5 m/s1.5 m/s/1 m/s0.5 m/s/0.3 m/s
v s i , v j i , v c i .Speeds of stepping up/down, jumping up/down and climbing up/down4 m/s/3 m/s5 m/s/4 m/s2 m/s/1 m/s
α i , β i , γ i , δ i Effect weights of PULs on the desired speeds of walking upright, bent-over walking, knee and hand crawling, low crawling* 0.9/0.9* 1/1* 0.8/0.8
[ D m i n , D m a x ] Local density range of other agents to affect if an agent performs 3D motions* [2, 3.5]/[2, 3.5] ped/m2* [1.5, 4]/[1.5, 4] ped/m2* [2.5, 3.5]/[2.5, 3.5] ped/m2
R *Radius from an agent’s footprint centre to detect other agents* 0.8 m/0.6 m* 1 m/0.8 m* 0.6 m/0.4 m
D a i Density threshold of other ahead agents to affect if an agent performs 3D motions* 3.5/4 ped/m2* 3/3.5 ped/m2*4/4.5 ped/m2
d v i , a v i , h v i   Distance, angle and eye height of an agent’s visibility* 2.5 m, 120°, 155 cm/2.5 m, 120°, 145 cm*3 m, 120°, 170 cm/3 m, 120°, 160 cm*2 m, 100°, 160 cm/2 m, 100°, 155 cm
R i , m Radius between an agent and an M-object to influence if perform 3D motions* 1 m/0.7 m* 1.5 m/1.2 m* 1 m/0.7 m
w m i n i Threshold of PULs to control if an agent is eligible to perform 3D motions* 0.4/0.5* 0.3/0.4* 0.5/0.6
p o i Probability of performing 3D motions, influenced by other agents who surround an agent * 0.5/0.4* 0.6/0.5* 0.4/0.3
p m i .Probability of performing 3D motions, influenced by other ahead agents through an agent’s visibility* 0.5/0.4* 0.6/0.5* 0.4/0.3
p d i Probability of performing 3D motions, influenced by an agent’s minimum desired speed* 0.5/0.4* 0.6/0.5* 0.4/0.3
p i Probability of moving up * 0.7/0.6* 0.8/0.7* 0.6/0.5
R u i Radius within which an agent detects another agent moving up.* 0.4 m/0.5 m* 0.4 m/0.3 m* 0.5 m/0.6 m
τ i Acceleration time of an agent0.5 s0.5 s0.5 s
r i Radius of an agent0.2 m/0.2 m0.25 m/0.25 m0.2 m/0.2 m
m i Mass of an agent55 kg/45 kg70 kg/55 kg65 kg/50 kg
A i Parameter for repulsive force of an agent200020002000
B i Parameter for repulsive force of an agent0.080.080.08
K Parameter for squeeze force of an agent1.2 × 1051.2 × 1051.2 × 105
k Parameter for squeeze force of an agent2.4 × 1052.4 × 1052.4 × 105
K i h Parameter for horizontal attraction force of an agent* 1.2 × 105* 1.2 × 105* 1.2 × 105
k i v Parameter for vertical interaction force of an agent* 0.05* 0.05* 0.05
Note: * denotes that the value range is assigned based on assumptions and an extensive try-and-error process by observing the patterns of 3D motions while visualising the changes.

References

  1. Xie, R.; Zlatanova, S.; Lee, J.B. 3D indoor-pedestrian interaction in emergencies: A review of actual evacuations and simulation models. In Proceedings of the International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Sydney, Australia, 19–21 October 2022; pp. 183–190. [Google Scholar] [CrossRef]
  2. Xie, R.; Zlatanova, S.; Lee, J.B.; Aleksandrov, M. A motion-based conceptual space model to support 3D evacuation simulation in indoor environments. Int. J. Geo Inf. 2023, 12, 494. [Google Scholar] [CrossRef]
  3. Haghani, M.; Yazdani, M. How simple behavioural modifications can influence evacuation efficiency of crowds: Part 1. Decision making of individuals. Transp. Res. Part C Emerg. Technol. 2024, 166, 104763. [Google Scholar] [CrossRef]
  4. Haghani, M. Optimising crowd evacuations: Mathematical, architectural and behavioural approaches. Saf. Sci. 2020, 128, 104745. [Google Scholar] [CrossRef]
  5. Ma, Y.; Yuen, R.K.K.; Lee, E.W.M. Effective leadership for crowd evacuation. Phys. A Stat. Mech. Its Appl. 2016, 450, 333–341. [Google Scholar] [CrossRef]
  6. Catal, C.; Akbulut, A.; Tunali, B.; Ulug, E.; Ozturk, E. Evaluation of augmented reality technology for the design of an evacuation training game. Virtual Real. 2020, 24, 359–368. [Google Scholar] [CrossRef]
  7. Wang, P.; Cao, S. Simulation of pedestrian evacuation strategies under limited visibility. Phys. Lett. A 2019, 383, 825–832. [Google Scholar] [CrossRef]
  8. Wang, K.; Fu, Z.; Li, Y.; Qian, S. Influence of human-obstacle interaction on evacuation from classrooms. Autom. Constr. 2020, 116, 103234. [Google Scholar] [CrossRef]
  9. Peacock, R.D.; Hoskins, B.L.; Kuligowski, E.D. Overall and local movement speeds during fire drill evacuations in buildings up to 31 stories. Saf. Sci. 2012, 50, 1655–1664. [Google Scholar] [CrossRef]
  10. Jiang, R.; Wang, Y.; Xie, R.; Zhou, T.; Wang, D. Effect of stairway handrails on pedestrian fatigue and speed during ascending evacuation. ASCE-ASME J. Risk Uncertain. Eng. Syst. Part A Civ. Eng. 2022, 8, 6022001. [Google Scholar] [CrossRef]
  11. Xie, R.; Pan, Y.; Zhou, T.; Ye, W. Smart safety design for fire stairways in underground space based on the ascending evacuation speed and BMI. Saf. Sci. 2020, 125, 104619. [Google Scholar] [CrossRef]
  12. Shiwakoti, N.; Sarvi, M.; Rose, G.; Burd, M. Animal dynamics based approach for modeling pedestrian crowd egress under panic conditions. Transp. Res. Part B Methodol. 2011, 45, 1433–1449. [Google Scholar] [CrossRef]
  13. Şahin, C.; Rokne, J.; Alhajj, R. Human behavior modeling for simulating evacuation of buildings during emergencies. Phys. A Stat. Mech. Its Appl. 2019, 528, 121432. [Google Scholar] [CrossRef]
  14. Zlatanova, S.; Holweg, D. 3D Geo-information in emergency response: A framework. In Proceedings of the Fourth International Symposium on Mobile Mapping Technology, Kunming, China, 29–31 March 2004; Available online: https://www.gdmc.nl/publications/2004/3D_Geo-information_Emergency_Response.pdf (accessed on 1 May 2025).
  15. Russo, F.; Rindone, C. Methods for risk reduction: Training and exercises to pursue the planned evacuation. Sustainability 2024, 16, 1474. [Google Scholar] [CrossRef]
  16. Menzemer, L.W.; Ronchi, E.; Karsten, M.M.V.; Gwynne, S.; Frederiksen, J. A scoping review and bibliometric analysis of methods for fire evacuation training in buildings. Fire Saf. J. 2023, 136, 103742. [Google Scholar] [CrossRef]
  17. Astarita, V.; Guido, G.; Haghshenas, S.S.; Haghshenas, S.S. Risk reduction in transportation systems: The role of digital twins according to a bibliometric-based literature review. Sustainability 2024, 16, 3212. [Google Scholar] [CrossRef]
  18. Russo, F.; Rindone, C.; Trecozzi, M.R. The role of training in evacuation. In Proceedings of the 8th International Conference on Simulation in Risk Analysis and Hazard Mitigation, Island of Brac, Croatia, 19–21 September 2012; pp. 491–502. [Google Scholar] [CrossRef]
  19. Feng, Z.; González, V.A.; Amor, R.; Spearpoint, M.; Thomas, J.; Sacks, R.; Lovreglio, R.; Cabrera-Guerrero, G. An immersive virtual reality serious game to enhance earthquake behavioral responses and post-earthquake evacuation preparedness in buildings. Adv. Eng. Inform. 2020, 45, 101118. [Google Scholar] [CrossRef]
  20. Sime, J.D. Movement towards the familiar: Person and place affiliation in a fire entrapment setting. Environ. Behav. 1985, 17, 697–724. [Google Scholar] [CrossRef]
  21. Kinateder, M.T.; Kuligowski, E.D.; Reneke, P.A.; Peacock, R.D. Risk perception in fire evacuation behavior revisited: Definitions, related concepts, and empirical evidence. Fire Sci. Rev. 2015, 4, 1. [Google Scholar] [CrossRef]
  22. Guo, N.; Ding, J.; Ding, Z.; Zhu, K.; Wu, C. Crawling evacuation from a room: Experiment and modeling. J. Stat. Mech. Theory Exp. 2021, 3, 033415. [Google Scholar] [CrossRef]
  23. Liu, Q.; Lu, L.; Zhang, Y.; Hu, M. Modeling the dynamics of pedestrian evacuation in a complex environment. Phys. A Stat. Mech. Its Appl. 2022, 585, 126426. [Google Scholar] [CrossRef]
  24. Zheng, Y.; Jia, B.; Li, X.; Jiang, R. Evacuation dynamics considering pedestrians’ movement behavior change with fire and smoke spreading. Saf. Sci. 2017, 92, 180–189. [Google Scholar] [CrossRef]
  25. Tugarinov, M.M.A.; Shulga, I.D.; Yurchenko, E.A.; Ermakov, A.D. 3D-simulation of emergency evacuation. J. Phys. Conf. Ser. 2020, 1680, 12052. [Google Scholar] [CrossRef]
  26. Tang, F.; Ren, A. GIS-based 3D evacuation simulation for indoor fire. Build. Environ. 2012, 49, 193–202. [Google Scholar] [CrossRef]
  27. Xie, R.; Zlatanova, S.; Lee, J. 3D indoor environments in pedestrian evacuation simulations. Autom. Constr. 2022, 144, 104593. [Google Scholar] [CrossRef]
  28. Aleksandrov, M.; Heslop, D.J.; Zlatanova, S. 3D indoor environment abstraction for crowd simulations in complex buildings. Buildings 2021, 11, 445. [Google Scholar] [CrossRef]
  29. Boguslawski, P.; Zlatanova, S.; Gotlib, D.; Wyszomirski, M.; Gnat, M.; Grzempowski, P. 3D building interior modelling for navigation in emergency response applications. Int. J. Appl. Earth Obs. Geoinf. 2022, 114, 103066. [Google Scholar] [CrossRef]
  30. Thunderhead Engineering, Pathfinder Technical Reference Manual 2020. Available online: https://support.thunderheadeng.com/docs/pathfinder/2020-1/ (accessed on 9 May 2022).
  31. Oasys, Massmotion Help Guide 2019. Available online: https://www.oasys-software.com/wp-content/uploads/2019/06/MassMotion-10.0-Help-Guide.pdf (accessed on 9 May 2022).
  32. Gorte, B.; Aleksandrov, M.; Zlatanova, S. Towards egress modelling in voxel building models. In Proceedings of the ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Kuala Lumpur, Malaysia, 1–3 October 2019; pp. 43–47. [Google Scholar] [CrossRef]
  33. Xiong, Q.; Zhu, Q.; Du, Z.; Zhu, X.; Zhang, Y.; Niu, L.; Li, Y.; Zhou, Y. A dynamic indoor field model for emergency evacuation simulation. ISPRS Int. J. Geo Inf. 2017, 6, 104. [Google Scholar] [CrossRef]
  34. Diakite, A.A.; No, L.D.I.A.-V.; Biljecki, F.; Isikdag, U.; Simmons, S.; Li, K.; Zlatanova, S. IFC2IndoorGML: An open-source tool for generating IndoorGML from IFC. In Proceedings of the International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Nice, France, 6–11 June 2022; pp. 295–301. [Google Scholar] [CrossRef]
  35. Deng, H.; Tian, M.; Ou, Z.; Deng, Y. A semantic framework for on-site evacuation routing based on awareness of obstacle accessibility. Autom. Constr. 2022, 136, 104154. [Google Scholar] [CrossRef]
  36. Kim, H.; Han, S. Crowd evacuation simulation using active route choice model based on human characteristics. Simul. Model. Pract. Theory 2018, 87, 369–378. [Google Scholar] [CrossRef]
  37. Zhang, L.; Wang, Y.; Shi, H.; Zhang, L. Modeling and analyzing 3D complex building interiors for effective evacuation simulations. Fire Saf. J. 2012, 53, 1–12. [Google Scholar] [CrossRef]
  38. Zhao, Y.; Lu, T.; Li, M.; Wu, P. The microscopic characteristics of escape behaviours from a three-dimensional lecture theatre under conditions of good and zero visibility. Saf. Sci. 2019, 118, 641–653. [Google Scholar] [CrossRef]
  39. Kneidl, A.; Hartmann, D.; Borrmann, A. A hybrid multi-scale approach for simulation of pedestrian dynamics. Transp. Res. Part C Emerg. Technol. 2013, 37, 223–237. [Google Scholar] [CrossRef]
  40. Kneidl, A.; Borrmann, A.; Hartmann, D. Generation and use of sparse navigation graphs for microscopic pedestrian simulation models. Adv. Eng. Inform. 2012, 26, 669–680. [Google Scholar] [CrossRef]
  41. Li, Y.; Chen, M.; Dou, Z.; Zheng, X.; Cheng, Y.; Mebarki, A. A review of cellular automata models for crowd evacuation. Phys. A Stat. Mech. Its Appl. 2019, 526, 120752. [Google Scholar] [CrossRef]
  42. Jun, H.; Huijun, S.; Juan, W.; Xiaodan, C.; Lei, Y.; Musong, G. Experiment and modeling of paired effect on evacuation from a three-dimensional space. Phys. Lett. A 2014, 378, 3419–3425. [Google Scholar] [CrossRef]
  43. Wei, J.; Zhang, H.; Guo, Y.; Gu, M. Experiment of bi-direction pedestrian flow with three-dimensional cellular automata. Phys. Lett. A 2015, 379, 1081–1086. [Google Scholar] [CrossRef]
  44. You, L.; Zhang, C.; Hu, J.; Zhang, Z. A three-dimensional cellular automata evacuation model with dynamic variation of the exit width. J. Appl. Phys. 2014, 115, 224905. [Google Scholar] [CrossRef]
  45. Fichtner, F.W.; Diakité, A.A.; Zlatanova, S.; Voûte, R. Semantic enrichment of octree structured point clouds for multi-story 3D pathfinding. Trans. GIS 2018, 22, 233–248. [Google Scholar] [CrossRef]
  46. Staats, B.R.; Diakité, A.A.; Voûte, R.L.; Zlatanova, S. Detection of doors in a voxel model, derived from a point cloud and its scanner trajectory, to improve the segmentation of the walkable space. Int. J. Urban Sci. 2019, 23, 369–390. [Google Scholar] [CrossRef]
  47. Xie, R.; Zlatanova, S.; Aleksandrov, M.; Lee, J.B. A voxel-based 3D indoor model to support 3D pedestrian evacuation simulations. J. Build. Eng. 2024, 98, 111183. [Google Scholar] [CrossRef]
  48. Helbing, D.; Farkas, I.; Vicsek, T. Simulating dynamical features of escape panic. Nature 2000, 407, 487–490. [Google Scholar] [CrossRef]
  49. Chen, J.; Shi, T.; Li, N. Pedestrian evacuation simulation in indoor emergency situations: Approaches, models and tools. Saf. Sci. 2021, 142, 105378. [Google Scholar] [CrossRef]
  50. Lin, J.; Zhu, R.; Li, N.; Becerik-Gerber, B. How occupants respond to building emergencies: A systematic review of behavioral characteristics and behavioral theories. Saf. Sci. 2020, 122, 104540. [Google Scholar] [CrossRef]
  51. Ronchi, E. Developing and validating evacuation models for fire safety engineering. Fire Saf. J. 2021, 120, 103020. [Google Scholar] [CrossRef]
  52. Liu, L.; Li, B.; Zlatanova, S.; van Oosterom, P. Indoor navigation supported by the industry foundation classes (IFC): A survey. Autom. Constr. 2021, 121, 103436. [Google Scholar] [CrossRef]
  53. Xie, R.; Zlatanova, S.; Lee, J.B. Voxel size performance in 3D indoor model creation for 3D evacuation simulations. In Proceedings of the International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Fremantle, Perth, WA, Australia, 22–25 October 2024; pp. 501–508. [Google Scholar] [CrossRef]
  54. Cao, L.; Davis, G.A.; Gallagher, S.; Schall, M.C.; Sesek, R.F. Characterizing posture and associated physiological demand during evacuation. Saf. Sci. 2018, 104, 1–9. [Google Scholar] [CrossRef]
  55. Kady, R.A.; Davis, J. The impact of exit route designs on evacuation time for crawling occupants. J. Fire Sci. 2009, 27, 481–493. [Google Scholar] [CrossRef]
  56. Wang, Q.; Song, W.; Zhang, J.; Ye, R.; Ma, J. Experimental study on knee and hand crawling evacuation for different age group students. Int. J. Disaster Risk Reduct. 2020, 48, 101613. [Google Scholar] [CrossRef]
  57. Dong, S.; Huang, P.; Wang, W. An optimization method for evacuation guidance under limited visual field. Phys. A Stat. Mech. Its Appl. 2022, 607, 128221. [Google Scholar] [CrossRef]
  58. Korhonen, M.T.; Mero, A.A.; Alen, M.; Sipilä, S.; Häkkinen, K.; Liikavainio, T.; Viitasalo, J.T.; Haverinen, M.T.; Suominen, H. Biomechanical and skeletal muscle determinants of maximum running speed with aging. Med. Sci. Sports Exerc. 2009, 41, 844–856. [Google Scholar] [CrossRef]
  59. Meng, Q.; Zhou, M.; Liu, J.; Dong, H. Pedestrian evacuation with herding behavior in the view-limited condition. IEEE Trans. Comput. Soc. Syst. 2019, 6, 567–575. [Google Scholar] [CrossRef]
  60. Papaiakovou, G.; Giannakos, A.; Michailidis, C.; Patikas, D.; Bassa, E.; Kalopisis, V.; Anthrakidis, N.; Kotzamanidis, C. The effect of chronological age and gender on the development of sprint performance during childhood and puberty. J. Strength Cond. Res. 2009, 23, 2568–2573. [Google Scholar] [CrossRef] [PubMed]
  61. Huang, L.; Luo, K.; Xu, J.; Xu, X.; Dong, Z. Experimental study on the elderly’s walking speed considering smoke density and visual acuity in buildings. Saf. Sci. 2021, 134, 105075. [Google Scholar] [CrossRef]
Ijgi 14 00197 g001
Figure 1. (a) Schematic diagram of navigable surfaces and vertical links around a table. (b) Six face-connected and four horizontal edge-connected neighbours of a voxel.
Figure 1. (a) Schematic diagram of navigable surfaces and vertical links around a table. (b) Six face-connected and four horizontal edge-connected neighbours of a voxel.
Ijgi 14 00197 g001
Ijgi 14 00197 g002
Figure 2. Original IFC models of three buildings and their voxel-based models for the investigation. (a) Restaurant. (b) Hall. (c) Canteen. Navigable surfaces for walking upright on floor slabs and vertical links for stepping up/down (green), navigable surfaces for walking upright on movable objects and vertical links for jumping up/down (blue), navigable surfaces for bent-over walking and vertical links for climbing up/down (cyan), navigable surfaces for knee and hand crawling (magenta), and navigable surfaces for low crawling (red).
Figure 2. Original IFC models of three buildings and their voxel-based models for the investigation. (a) Restaurant. (b) Hall. (c) Canteen. Navigable surfaces for walking upright on floor slabs and vertical links for stepping up/down (green), navigable surfaces for walking upright on movable objects and vertical links for jumping up/down (blue), navigable surfaces for bent-over walking and vertical links for climbing up/down (cyan), navigable surfaces for knee and hand crawling (magenta), and navigable surfaces for low crawling (red).
Ijgi 14 00197 g002
Ijgi 14 00197 g003
Figure 3. (a) Number of agents performing 3D motions in the restaurant. (b) Breakdown of agents performing 3D motions by demographic groups in the restaurant.
Figure 3. (a) Number of agents performing 3D motions in the restaurant. (b) Breakdown of agents performing 3D motions by demographic groups in the restaurant.
Ijgi 14 00197 g003
Ijgi 14 00197 g004
Figure 4. (a) Number of agents performing 3D motions in the hall. (b) Breakdown of agents performing 3D motions by demographic groups in the hall.
Figure 4. (a) Number of agents performing 3D motions in the hall. (b) Breakdown of agents performing 3D motions by demographic groups in the hall.
Ijgi 14 00197 g004
Ijgi 14 00197 g005
Figure 5. (a) Number of agents performing 3D motions in the canteen. (b) Breakdown of agents performing 3D motions by demographic groups in the canteen.
Figure 5. (a) Number of agents performing 3D motions in the canteen. (b) Breakdown of agents performing 3D motions by demographic groups in the canteen.
Ijgi 14 00197 g005
Ijgi 14 00197 g006
Figure 6. Number of agents moving above. (a) Restaurant. (b) Hall. (c) Canteen.
Figure 6. Number of agents moving above. (a) Restaurant. (b) Hall. (c) Canteen.
Ijgi 14 00197 g006
Ijgi 14 00197 g007
Figure 7. (a) TET in the restaurant across different agent numbers and PUL levels. (b) TET in the hall across different agent numbers and PUL levels. (c) TET in the canteen across different agent numbers and PUL levels.
Figure 7. (a) TET in the restaurant across different agent numbers and PUL levels. (b) TET in the hall across different agent numbers and PUL levels. (c) TET in the canteen across different agent numbers and PUL levels.
Ijgi 14 00197 g007
Ijgi 14 00197 g008
Figure 8. Speed variation maps for the canteen with 250 agents at 0.7 PUL under 3D motion and no-3D motion conditions. (a) With 3D motions. (b) No 3D motions.
Figure 8. Speed variation maps for the canteen with 250 agents at 0.7 PUL under 3D motion and no-3D motion conditions. (a) With 3D motions. (b) No 3D motions.
Ijgi 14 00197 g008
Ijgi 14 00197 g009
Figure 9. Screenshots of a simulation demonstrating agents climbing up/down, walking up, and low crawling around a table and 3D trajectories. (a) Visualisation in the original model. (b) Visualisation with navigable surfaces and vertical links represented by voxels. Navigable surfaces for walking upright on floor slabs and vertical links for stepping up/down (green), navigable surfaces for walking upright on movable objects and vertical links for jumping up/down (blue), navigable surfaces for bent-over walking and vertical links for climbing up/down (cyan), navigable surfaces for knee and hand crawling (magenta), and navigable surfaces for low crawling (red). (c) The 3D agent trajectories in the top view. (d) The 3D agent trajectories in the 3D view.
Figure 9. Screenshots of a simulation demonstrating agents climbing up/down, walking up, and low crawling around a table and 3D trajectories. (a) Visualisation in the original model. (b) Visualisation with navigable surfaces and vertical links represented by voxels. Navigable surfaces for walking upright on floor slabs and vertical links for stepping up/down (green), navigable surfaces for walking upright on movable objects and vertical links for jumping up/down (blue), navigable surfaces for bent-over walking and vertical links for climbing up/down (cyan), navigable surfaces for knee and hand crawling (magenta), and navigable surfaces for low crawling (red). (c) The 3D agent trajectories in the top view. (d) The 3D agent trajectories in the 3D view.
Ijgi 14 00197 g009
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Xie, R.; Zlatanova, S.; Lee, J.; Borrmann, A. Evacuation Behavioural Instructions with 3D Motions: Insights from Three Use Cases. ISPRS Int. J. Geo-Inf. 2025, 14, 197. https://doi.org/10.3390/ijgi14050197

AMA Style

Xie R, Zlatanova S, Lee J, Borrmann A. Evacuation Behavioural Instructions with 3D Motions: Insights from Three Use Cases. ISPRS International Journal of Geo-Information. 2025; 14(5):197. https://doi.org/10.3390/ijgi14050197

Chicago/Turabian Style

Xie, Ruihang, Sisi Zlatanova, Jinwoo (Brian) Lee, and André Borrmann. 2025. "Evacuation Behavioural Instructions with 3D Motions: Insights from Three Use Cases" ISPRS International Journal of Geo-Information 14, no. 5: 197. https://doi.org/10.3390/ijgi14050197

APA Style

Xie, R., Zlatanova, S., Lee, J., & Borrmann, A. (2025). Evacuation Behavioural Instructions with 3D Motions: Insights from Three Use Cases. ISPRS International Journal of Geo-Information, 14(5), 197. https://doi.org/10.3390/ijgi14050197

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop