Evaluating the Efficacy of Agent-Based Modeling in Analyzing Pedestrian Dynamics within the Built Environment: A Comprehensive Systematic Literature Review

Abstract

The dynamics of pedestrian behavior within the built environment represent a multifaceted and evolving field of study, profoundly influenced by shifts in industrial and commercial paradigms. This systematic literature review (SLR) is motivated by the imperative to comprehensively investigate and assess the built environment through the lens of pedestrian modeling, employing advanced modeling tools. While previous scholarship has explored the interplay between the built environment and pedestrian dynamics (PD), there remains a conspicuous gap in research addressing the utilization of agent-based modeling (ABM) tools for a nuanced evaluation of PD within these contexts. The SLR highlights the essential and practical benefits of using ABM to study PD in built environments and combine related theories and practical projects. Beyond theoretical discussions, it emphasizes the real-world contributions of ABM in understanding and visualizing how people behave in urban spaces. It aims to provide deep insights for both researchers and urban planners. By thoroughly examining recent research, it not only explores the practical uses of ABM but also reveals its broad implications for various aspects of pedestrian behavior in built environments. As a result, this SLR becomes a key resource for understanding the crucial role of ABM in studying the complexities of our surroundings. The findings and discussion here highlight ABM’s vital role in bridging the gap between theory and practice, improving our understanding of pedestrian behavior in urban settings. Furthermore, this study outlines promising avenues for future research, thereby fostering continued exploration and innovation in the dynamic realm of pedestrian behavior within built environments.

1. Introduction

Understanding PD [1,2] in walking-predominant transportation cities is crucial for effective urban planning, infrastructure optimization, and urban design [3]. The movements and interactions of pedestrians significantly influence the functionality and safety of urban spaces. Traditional modeling approaches, such as macroscopic and microscopic models, have provided valuable insights into pedestrian behavior. However, these methods often fall short in capturing the complexity and variability of real-world pedestrian interactions.

ABM emerges as a powerful computational tool to address these limitations as a key computer-based simulation [3]. ABM simulates the actions and interactions of autonomous agents, offering a natural and adaptable approach to model PD. By mimicking individual behaviors and their responses to environmental cues, ABM provides a nuanced and detailed understanding of pedestrian movements in urban settings as a developed interface [4].

PD encompasses the study of how individuals navigate through urban spaces, responding to environmental cues, social interactions, and personal goals. These dynamics are profoundly influenced by shifts in industrial and commercial paradigms, leading to changes in urban structure, transportation systems, and land use [5]. As cities continue to grow and evolve, understanding pedestrian behavior becomes pivotal in promoting sustainable urban planning paradigms, optimizing infrastructural elements, and resolving intricate problem scenarios.

ABM is becoming well known and used in many fields because it mimics the behavior of a complex system where agents interact with each other and their surroundings using straightforward local rules [6]. The effectiveness of this technique in forecasting pedestrian movement in urban areas, the spread of infectious diseases, forecasting traffic movement, and the behavior of economic systems has increased attention paid to the effectiveness of the technology. Four key points that were summarized by Ali Bazghandi [3] focus on the superiority of ABM compared to other modeling techniques: (i) it captures emergent events; (ii) it gives a natural description of a system; (iii) it is adaptable; and (iv) it is time- and cost-efficient.

In this context, ABM has emerged as a potent computational paradigm for simulating PD within the built environment [7]. ABM stands out as a modeling approach capable of capturing emergent phenomena, offering naturalistic depictions of intricate systems and exhibiting temporal and resource efficiency [7]. The flexibility of ABM to handle diverse tasks, its bottom-up approach, and its ability to use random elements make it a great tool for improving models of pedestrian behavior. However, using ABM to study pedestrian dynamics has many challenges. One major issue is the need for well-defined models, as ABM’s success depends heavily on how well the model is designed. Also, performance can be a problem when dealing with a large number of agents.

This area of research has employed various modeling methods to understand and simulate how people move in urban settings. These methods can generally be divided into three categories: macroscopic, microscopic, and mesoscopic models [8].

Macroscopic Models: These models offer a broad view of pedestrian flows, treating crowds like a continuous fluid. They are useful for understanding overall movement patterns and flow characteristics in detail, but neglect individual movements [9]. The Lighthill–Whitham–Richards (LWR) model falls in this category.

Microscopic Models: These models focus on individual pedestrian behaviors and interactions. They model each pedestrian as an individual entity, allowing for a detailed analysis of interactions and decision-making processes. The social force model and CA are examples that have been widely used in PD research [10].

Mesoscopic Models: These models bridge the gap between macroscopic and microscopic approaches by incorporating elements of both. They consider group-level behaviors while still accounting for some individual interactions. They are useful when computational efficiency is crucial and detailed individual behaviors are less critical [11].

Comparison with ABM: ABM stands out from traditional modeling approaches by simulating the actions and interactions of autonomous agents. This method provides a more nuanced and adaptable approach to modeling pedestrian dynamics, capturing emergent behaviors resulting from individual interactions. ABM’s flexibility makes it especially suitable for complex urban environments where pedestrian behaviors are influenced by various factors [12].

The stages of development of ABM in PD are as follows.

• Early Development: The initial ABM focused on modeling simple behaviors and interactions of agents. These models demonstrated ABM’s potential to capture complex dynamics that were challenging to model using traditional methods [13].
• Advancements in Computational Power: More sophisticated ABM models were created when computational power improved. These models included higher levels of detail and more complex interaction rules, leading to more accurate simulations of PD in urban settings [14].
• Integration with Urban Planning: Recently, ABM has been integrated with urban planning tools. This integration provides urban design with more efficient and safer urban spaces [15].

While traditional models have provided valuable insights into pedestrian dynamics, ABM excels by capturing individual interactions’ complexity and variability. This review highlights ABM’s practical applications and limitations, offering a comprehensive understanding of its role in urban planning and design. The evolution from cellular automata CA to ABM marks a significant milestone, enabling the introduction of diverse agents and the exploration of nonlinear dynamics, allowing researchers to model pedestrians as intelligent, autonomous agents interacting with their environment and each other [16].

In the macroscopic approach, every individual and their interactions are modeled. In this approach, the averages can be used without losing information, as all possible sources of heterogeneity can be incorporated into the studies. Contrarily, the microscopic approach does not explicitly mention any micro-level property or system. The underlying assumption in these models is that aggregation of results can be used without losing any information. These models are applied successfully in the prediction of certain traffic situations, while the mesoscopic approach tries to make the most of the computational power by incorporating as many individuals as possible without deteriorating the quality of results. Although each individual cannot be tracked down, the models allow individual properties [3,4].

In the literature, John von Neumann and Stanislaw Ulam began exploring microsimulation using CA at the Los Alamos National Laboratory in New Mexico [3] (Cohen, 2018) [17]. They reproduced patterns and behaviors with significant accuracy using simple rules. The next step in the simulations was the development of ABM, which allowed the introduction of multi-heterogeneous agents. Nonlinear dynamics were approached from a different angle by studying the local interactions between agents. Each individual in the pedestrian stream is modeled as an intelligent and autonomous agent. Pedestrians (agents) move and interact with other agents independently through a set of decision rules.

Recently, several studies [17,18,19] have supported ABM as an effective method for simulating pedestrian systems via computer. ABM, a modeling technique gaining popularity across various fields such as economics, biology, ecology, and social science, involves portraying the entities of study as agents within a specific environment. These agents engage with one another and with their surrounding environment according to a series of established decision-making rules.

Despite the growing use of ABM in various fields, there is a notable research gap in comprehensively understanding its practical applications and limitations specifically in pedestrian dynamics. This review seeks to bridge this gap by synthesizing existing studies to provide a clearer picture of how ABM can be effectively utilized in urban planning contexts. The novelty of this study lies in its focused examination of ABM within the built environment, which has not been extensively covered in previous literature.

This systematic literature review aims to be a valuable resource by highlighting the benefits of ABM in studying pedestrian dynamics within built environments. By combining theoretical foundations with practical insights from significant projects, the review bridges the gap between theory and practice. Pedestrian behavior in urban settings is a complex field influenced by industrial and commercial changes. ABM can provide crucial insights into this domain. This review evaluates ABM’s effectiveness, addressing its practical applications and limitations. By synthesizing theoretical and practical aspects, it underscores ABM’s contributions to urban planning and design, enhancing our understanding of pedestrian behavior and guiding future research.

Agent:

According to agency theory, an agent is a piece of software that acts independently and interacts with its environment to achieve its goals. Agents are ideal for understanding human actions due to their proactive character and autonomy (physical situations or mental states) (Martinez-Gil et al., 2017) [18]. According to Ronald et al. [19], an agent is described as “a software component that is embedded in its environment, autonomous, and not under external control, adaptive to changes in its surroundings and proactive in consistently pursuing goals, communication capacities of obtaining goals, flexibility, resilience, and social interaction with other actors.”

Built environment:

A built environment holds a vital role in urban areas or public spaces. The area can be considered an entity that can be understood based on its dynamics or the complexity of understanding pedestrian movement in place [19]. Furthermore, the built environment can be interpreted as being built for a particular event or an activity (sports, pilgrimage, or any event).

Built environment categorization:

The built environment significantly affects pedestrians’ behaviors in various environments. Study [19] proposed categorizing the qualities of settings, walking behaviors, map representation, and projected pedestrian traffic. This “physical” method of categorization simplifies the modeling process. For instance, psychologists believe that instead of the number of individuals present, our feelings are affected by how comfortable we feel about being in a crowded area. Based on this concept, the built environment was further classified into small, large, mixed-mode, open, and hybrid spaces.

Small-scale enclosed spaces: These are made up of little rooms linked with hallways and doors. For instance, buildings frequently feature a lot of confined environments (e.g., offices and meeting rooms).

Large-scale enclosed spaces: These are often more substantial structures with broad layouts. For instance, a sports stadium has a space with seats, aisles, and exits.

Mixed-mode spaces: These comprise spaces connecting pedestrians to building entrances and other streets that may be shared with vehicles or public transportation. The pedestrian must navigate past static items, such as benches, trash cans, and garden areas. A line, which may be an organized line of individuals waiting to enter a popular store or an unorganized line of people waiting to cross the street or wait for a bus, is yet another component of this ecosystem.

Open spaces: These are made up of open spaces that may have specific paths set aside for them. Behaviors include drifting, repeated pauses, and potentially longer stays for picnics or sightseeing because the aim is most likely to relax.

Hybrid spaces: These often contain pedestrian-friendly or low-traffic locations with several attractions, including sports stadiums or universities. They incorporate actions from enclosed, mixed-mode, and open-space environments. Examples of behaviors in an open-space environment include wandering and afternoons on the lawn (e.g., moving around lecture theaters).

Pedestrian Dynamics:

Pedestrians are the most complex agents of the model. Since pedestrians cannot acquire a complete information set at a given moment, the technique of satisficing optimization is used, whereby pedestrians will try to get to their destination by minimizing the cost of their movement. Several pedestrian behaviors can occur in this environment and influence the model type. Purposeful and familiar, purposeful and unfamiliar, purposeless, required behavior (panic situation, evacuation behavior), forced waiting, and temporal constraints are some of the common pedestrian behavior types.

For pedestrians, these are the randomly generated parameters adopted [17]:

• The distance of vision: How far the pedestrian can see.
• The angle of vision: Determines the angle of vision.
• Noise: Determines the random angle to turn when facing an obstacle.
• Efficiency: Defines a threshold of acceptance between the shortest path and a more indirect alternative.
• Patience: Defines the threshold for waiting.
• Risk-taker: Defines how much utility difference s/he will accept.

With the advancement of tools and technology, PD in built environments can also be classified into different pedestrian modeling categories [20,21,22,23], illustrated in Figure 1 (Martinez-Gil et al., 2017) [18]. Figure 2 depicts the schematic diagram of pedestrian flows with intersecting flows, bottlenecks, and lane formation of pedestrians.

Agent-Based Modeling (ABM):

Based on the aforementioned information, ABM is one of the approaches to represent this kind of model for visualization and implementation. The behaviors of cars, pedestrians, traffic, the environment, and how they interact will be explicitly described from a theoretical perspective. In practice, the behavioral characteristics will be modeled, and the resulting dynamics will be explored by running several simulations. Executing the model several times under identical initial conditions, a specific emergent, an emerging behavior of the system is expected. Cohen (2018) [17] mentioned that the models constructed would be provided with an overview, design concept, and detail (ODD) protocol to enable comparison with other ABMs. Based on the computing capacity, the algorithm employs the idea of ODD to facilitate unambiguous communication between the environment and agents, allowing the latter to perceive and respond appropriately.

Finally, particular attention will be devoted to communicating and visualizing the model as efficiently as possible. The design principles of (i) simplify, (ii) emphasize, and (iii) explain will be used as a guideline to orientate the efforts of this work. Then, the models use the ABMs to resolve complicated issues, where CA is a suitable example of ABM.

Table 1. Advantages and disadvantages of ABMs.

Pros	Cons
Flexibility and Adaptability	Computational Complexity
Heterogeneity: Models diverse agents (Heath et al., 2009) [17]	High Resource Demand: Requires significant computational resources (Manzo, 2014) [24]
Ability to iteratively learn and add intelligence without disturbing its current operation. (Derksen et al., 2012) [21]	Long Simulation Times: Detailed models may take a long time to run (Niemann et al., 2021) [22]
Complex Behavior: Captures intricate interactions (Macal and North, 2010) [23]
Dynamic Environments: Adapts over time (Manzo, 2014) [24]
Detailed Insights	Model Validation and Verification
Micro-Level Analysis: Insights into individual behaviors (Richetin et al., 2009) [25]	Complex Validation: Difficult to validate complex models (Collins et al., 2024) [26]
Emergent Properties: Understands system-level outcomes from agent interactions (Bonabeau, 2002) [12]	Data Requirements: Needs high-quality, detailed data (Manzo, 2014) [24]
Scalability	Model Development
Scalable Models: Adjusts to various sizes and complexities (Parviero, 2022) [27]	Time-Consuming: Developing models takes significant time and expertise (Taylor et al., 2016) [28]
Increasing availability of micro-data to support agent-based models, and advances in computer performance (Macal and North, 2006) [23]	Programming Skills: Requires good programming knowledge and familiarity with ABM tools (Fabris, 2023) [29]
Interdisciplinary Applications	Efficiency-oriented solutions
Wide Applicability: Used in multiple fields like economics, social sciences, biology, and AI (Axtell and Farmer, 2022) [30]	Both large- and small-scale simulations are vulnerable to memory flaws which could invalidate experimental results (Antelmi et al., 2023) [31].
Experimentation	Interpretation of Results
Conducts scenario analyses (Assefa et al., 2021) [21]	Complex Results: Results can be difficult to interpret (Sun et al., 2015) [32]
	Overfitting: Risk of overfitting to specific data (Srikrishnan and Keller, 2021) [33]
Uncertainty and Sensitivity
Parameter Sensitivity: Results can be sensitive to parameter choices and initial conditions (Borgonovo et al., 2022) [34]	Stochastic Variability: Requires multiple runs and statistical analysis to manage variability (Hunter and Kelleher, 2020) [35]

describes the advantages and disadvantages of this modeling technique.

Figure 3 depicts the importance of an environment for an agent, crowd flows, and pedestrian navigation within the environment. Developing a model between the agent and the environment is essential for simulation-based PD.

Figure 4 displays three layers of the model design concept, namely, emphasizing model design, adaptation, and interaction.

The structure of this SLR consists of three phases as Figure 5 planning, conducting, and reporting reviews. Section 2 explains the research technique. The findings of several studies, study topics, common procedures, data formats, and performance approaches are presented in Section 3. In Section 4, current solutions are discussed with their contributions, management and academic consequences, limits, future research directions, and the scientific value of this study.

2. Research Methodology

The philosophy for developing this SLR was adopted from a previous study (Kitchenham et al., 2002) [37]. There are three phases to the research process. The initial planning phase addresses defining research topics, designing, and verifying review methods. The discovery and selection of relevant research, data extraction, and the information synthesis process are addressed in the second step. Writing and validating the SLR falls in the third phase. The flow of the three stages is depicted in Figure 6.

2.1. Plan Review

The development of review protocols and the key research questions are described together with the relevant search technique in the initial stage of the research process.

Research Questions

The following research questions were established in this SLR, where potentially all of them can be addressed with appropriate solutions.

RQ 1: To what extent can ABM accurately simulate pedestrian dynamics in urban environments, considering the impact of different modeling parameters and software platforms?

RQ 2: What are the key challenges and limitations of ABM in modeling pedestrian behavior in complex built environments, and how can these challenges be addressed to improve modeling accuracy?

RQ 3: How can ABM simulations of pedestrian dynamics provide valuable insights for enhancing urban planning, design, and policy decisions, particularly in optimizing transportation hubs and public spaces for pedestrian safety and accessibility?

2.2. Review Protocols

This review follows a structured protocol to ensure a comprehensive and unbiased synthesis of the existing literature on ABM in pedestrian dynamics. The review protocol outlines the objectives, scope, and procedures for conducting the literature review.

2.3. Search Strategy

The search strategy involves systematically searching multiple academic databases, including but not limited to PubMed, Scopus, and Web of Science. Keywords such as “agent-based modeling,” “pedestrian dynamics,” “urban planning,” and “simulation models” were used to identify relevant studies. Boolean operators and filters were applied to refine the search results and ensure the inclusion of pertinent articles.

2.3.1. Searching Keywords

To ensure that the evaluation closely covered the built environment and technologies used for the microscopic environment using ABM, we focused our search on the most pertinent terms. Hence, we searched for phrases before performing the following steps:

• Extracting the significant distinct terms based on our research questions.
• We used different terms as keywords, such as PD, ABM.
• Updating our search terms with keywords from relevant papers.

We used the main alternatives and added “OR operator” and “AND operator” to obtain the maximum number of relevant articles, as depicted in

Table 2. Inclusion and exclusion criteria.

ID	Keywords
1	(“Pedestrian Dynamics” OR “PD”) AND (“Built Environment” OR “Built”) AND (“Agent Based Modeling” OR “ABM”)
2	(“Pedestrian Dynamics” OR “PD”) AND (“Built Environment” OR “Built”) AND (“Agent Based simulations” OR “Simulations”)
3	(“Macro” OR “PD”) AND (“Built Environment” OR “Built”) AND (“Agent Based Modeling” OR “ABM”) AND (“Simulations”)
4	(“Micro” OR “PD”) AND (“Built Environment” OR “Built”) AND (“Agent Based Modeling” OR “ABM”)
5	(“Micro” OR “Macro”) AND (“Built Environment” OR “Built”) AND (“Agent Based Modeling” OR “ABM”) AND (“Simulations”)

.

2.3.2. Literature Resources

The foundational review studies sourced their pertinent literature from databases including Web of Science, Scopus, ACM Digital Library, Springer, Science Direct, and IEEE Explorer. Such repositories, encompassing ISI and Scopus-indexed articles as well as selected papers from significant conferences, offer the broadest array of high-quality research on the topic. The search phrase was created using sophisticated search options provided by these databases. Our search included articles published from 2011 through 2023.

By focusing on articles published from 2011 onwards, the authors aimed to include the most recent and relevant research in the field of PD and ABM. This ensures that the SLR encompasses the latest developments, methodologies, and findings, which is essential for providing up-to-date insights.

Research in fields like computational modeling can evolve rapidly. By focusing on a narrower timeframe, the authors can conduct a more thorough and detailed review of the selected articles.

2.4. Conduct Review

According to the research questions, keywords, and protocols as a reference in this stage, this phase mainly deals with article inclusion and exclusion, as mentioned in

Table 3. Inclusion and exclusion criteria according to the RQ, keywords and protocol.

Inclusion Criteria

Studies that employed ABM for pedestrian dynamics. Articles published in peer-reviewed journals. Research focusing on urban planning and design contexts.

Exclusion Criteria

Studies not related to pedestrian dynamics. Articles not published in English. Non-peer-reviewed publications, such as conference proceedings and book chapters.

.

2.4.1. Study Selection

Figure 7 depicts the entire study selection procedure. A total of 1500 items were found through online search. Following filtration utilizing title, keyword, inclusion, and exclusion criteria, 300 articles remained. There were 96 articles on various concepts, such as visualization of civil engineering designs, maps, and model implementation using BIM, and 124 articles from other fields like economics and health were removed. Fifty articles are removed from the list after reading the entire article based on the RQ and information synthesis.

The methodology for choosing pertinent articles through keyword searches and article selection is detailed in Figure 7. Articles that were duplicates or did not cover all the research questions were excluded. The criteria for assessing the quality of studies are presented in

Table 4. Quality checklist.

No.	Questions
1	Was there a strong focus on PD, such as micro/macro?
2	Was the study able to describe how important ABM simulations tool is for designing the model?
3	Was any efficient tool or algorithm used to develop the PD model for the built environment?
4	Did the study concentrate on the basic approaches of ABM for the built environment?
5	Did any study cover all the aspects of PD instances?

, which primarily facilitated the identification of relevant, detailed, and exhaustive studies.

2.4.2. Data Extraction

Table 5. Data extraction.

Study
Study Research Problem Contributions
RQ1: PD
RQ2: PD instances used for built environment
RQ3: Application and implications: ABM tools used in PD

shows the data extraction process in systematically collecting relevant information from each included study. Key data points extracted included study objectives, methodology, key findings, limitations, and implications for urban planning. A standardized data extraction form was used to ensure consistency and accuracy in data collection.

Table 6. Summary of the selection process.

Search Database	Keywords Used	Initial Results	Screened Articles	Full-Text Reviewed	Included Studies
PubMed	“pedestrian dynamics” AND “Agent-Based Modeling”	300	60	30	4
Scopus	“urban planning” AND “simulation models”	500	150	40	9
Web of Science	“built environment” AND “ABM”	400	100	20	6
IEEE Xplore	“pedestrian behavior” AND “simulation”	200	30	10	3
ACM Digital Library	“Agent-Based Modeling” AND “public spaces”	200	30	10	2
Total	Combined across all databases	1600	370	110	26

gives a summary of the selection process.

2.5. Analysis

The analysis involved synthesizing the extracted data to identify common themes, trends, and gaps in the literature. A narrative synthesis approach was used to provide a comprehensive overview of the state of research on ABM in pedestrian dynamics. The findings were organized to highlight the practical applications and limitations of ABM, with a focus on its relevance to urban planning and design.

2.5.1. Information Synthesis

At this stage, the gathered data were amalgamated to address the questions posed by our study, for which we utilized a narrative synthesis method. Results are discussed below.

2.5.2. Report Review

Data extracted from the primary studies were used to answer the three research questions. The guidelines [37] were closely followed in the reporting of results.

3. Results

Approximately 30 studies were included in this review. All studies were relevant to dynamic pedestrian types, the need for analysis thereof in the built environment, and to address the categories of PD using ABM to shortlist the number of studies focusing on the built environments used (

Table 7. RQ Studies.

Research Questions	Studies
RQ1: PD	24
RQ2: PD instance used for built environment
RQ3: Application and implication of ABM tools used in PD

).

Table 8. Main categories of the research papers.

Pedestrian Dynamic Literature
Public Space Optimization	Urban Design and Planning	Emergency Response and Evacuation Planning	Transportation Hub Design
Andrew Crooks et al. (2015) [38]. Nova Asriana (2021) [39]. Gabriele Filomena and Judith A. Verstegen (2021) [40].	López Baeza et al. (2021) [41]. Gabriele F. et al. (2020) [40]. Kostas Cheliotis (2020) [14]	Zi-Xuan Zhou et al. (2021) [42]. Ren-Yong Guo et al. (2012) [43] Ruggiero Lovreglio et al. (2014) [44]. Nirajan Shiwakoti et al. (2011) [45] Caesar Saloma et al. (2015) [46]. D.R. Parisi et al. (2015) [47] Yue Xu (2017) [48] Zi-Xuan Zhou et al. (2021) [42]	Flurin S. Hänseler et al. (2016) [49] Kapil Sinha et al. (2021) [50] Jia Liu and Xiaohong Chen (2019) [51] Mohamed Hussein and Tarek Sayed (2018) [4] Erfan Hassannayebi et al. (2019) [36]. Haoling Wu et al. (2016) [52]. J. Zhang et al. (2015) [53] Joshua Auld et al. (2016) [54] Hussein and Sayed (2015) [55] Kaziyeva et al. (2023) [56] C. K. Lim et al. (2018) [57]

displays the number of studies on PD conducted in built environments per year. As clearly shown in the figure, there is an increment in studies carried out in this domain.

The research analysis organizes the studies into clear categories, emphasizing their contributions to understanding and modeling pedestrian dynamics in different contexts Table 8. Additionally, the analysis of these research papers is summarized in a tabulated format below to facilitate comprehension of their complexities

Table 9. Summary of results.

	Study	Main Findings	Relevance to Pedestrian Dynamics
[41]	López Baeza et al. (2021)	ABM simulations matched real-world pedestrian behavior. Highlighted urban space modifications to influence pedestrian activity levels.	Enhances understanding of pedestrian behavior and informs urban space modifications to increase pedestrian presence.
[58]	Gabriele F. et al. (2019)	The regional algorithm led to diverse agent routes, suggesting accurate representation of pedestrian behavior.	Provides realistic simulations crucial for accurate pedestrian flow analysis in urban planning.
[14]	Kostas Cheliotis (2020)	ABM accurately simulated human and crowd behavior in public spaces, aligning with real-world scenarios.	Demonstrates ABM’s effectiveness in simulating pedestrian behavior in public spaces.
[38]	Andrew Crooks et al. (2015)	Introduced SA2-ABM with spatiotemporal data for accurate pedestrian movement representation.	Offers detailed insights into pedestrian dynamics for urban planning applications.
[39]	Nova Asriana (2021)	Leveraged ABM for analyzing pedestrian behavior in tourism areas, highlighting physical and social influences.	Provides insights into pedestrian movement patterns in tourism-heavy areas.
[59]	Gabriele F. and Judith A. V. (2021)	Inclusion of landmarks in ABM led to realistic pedestrian movement patterns.	Enhances pedestrian flow modeling by considering landmark-based navigation.
[50]	Kapil Sinha et al. (2021)	Highlighted ABM’s capability in capturing passenger behavior in airport terminals.	Informs design improvements for pedestrian safety and efficiency in transportation hubs.
[49]	Flurin S. Hänseler et al. (2016)	Explored ABM in train stations combining dynamic demand estimation with traffic assignment models.	Applicable in optimizing pedestrian flow and level of service in transportation hubs.
[51]	Jia Liu and Xiaohong Chen (2019)	Explored ABM in metro stations integrating emotional states into movement analysis.	Provides a robust framework for predicting pedestrian destination choices in metro stations.
[4]	Mohamed Hussein and Tarek Sayed (2018)	Focused on ABM calibration and validation for pedestrian dynamics in subway stations.	Ensures accurate simulation of pedestrian interactions in complex environments.
[36]	Erfan Hassannayebi et al. (2020)	Highlighted ABM effectiveness in transportation hubs focusing on live video data calibration.	Essential for accurate modeling of pedestrian traffic and emergency evacuations.
[52]	Haoling Wu et al. (2016)	Validated ABM under various pedestrian conditions, emphasizing pedestrian diversity in walkways.	Highlights the impact of pedestrian diversity on walkway capacity.
[42]	Zi-Xuan Zhou et al. (2021)—Visual Information	Integrated machine learning with ABM for realistic evacuation simulations.	Enhances accuracy of pedestrian evacuation modeling using visual information.
[57]	C. K. Lim et al. (2018)	Implemented particle-based crowd simulation capturing diverse behaviors in George Town, Malaysia.	Useful for simulating complex pedestrian dynamics in multicultural urban centers.
[53]	J. Zhang et al. (2015)	Highlighted pedestrian flow dynamics through controlled experiments, crucial for urban design and emergency planning.	Provides critical insights for designing pedestrian flows in various corridor configurations.
[43]	Ren-Yong Guo et al. (2012)	Explored pedestrian behavior in emergency evacuations under varying visibility conditions.	Highlights the necessity of considering visibility in pedestrian evacuation planning.
[45]	Nirajan Shiwakoti et al. (2011)	Investigated ant behavior for insights into human crowd dynamics in panic conditions.	Emphasizes the complexity of modeling pedestrian behavior under panic.
[46]	Caesar Saloma et al. (2015)	Explored the impact of prior training on group emergency evacuation behavior of mice.	Highlights the effectiveness of training in improving pedestrian evacuation efficiency.
[60]	A. Garcimartín et al. (2015)	Explored dynamics of sheep herds under competitive conditions, applicable to crowd dynamics.	Provides insights into pedestrian dynamics in competitive evacuation scenarios.
[47]	D.R. Parisi et al. (2015)	Examined validity of using animal models to simulate human behavior in emergency evacuations.	Critically reassesses modeling parameters for accurate human behavior simulation.
[44]	Ruggiero Lovreglio et al. (2014)	Introduced a mixed logit model to enhance ABM for emergency evacuations.	Offers a detailed approach to modeling pedestrian decisions during evacuations.
[48]	Yue Xu (2017)	Explored ABM for emergency evacuations in underground transport settings.	Informs urban safety planning and pedestrian evacuation strategies.
[54]	Joshua Auld et al. (2016)	Introduced Polaris ABM framework for urban environments integrating various urban dynamics.	Demonstrates the impact of integrated modeling on pedestrian behavior understanding.
[61]	Hussein and Sayed (2017)	Established a novel ABM for simulating pedestrian movement in urban settings.	Highlights accurate prediction of pedestrian trajectories in urban areas.
[55]	Hussein and Sayed (2015)	Developed an advanced ABM for simulating pedestrian behaviors influenced by environmental variables.	Enhances understanding of pedestrian navigation in urban contexts.
[56]	Kaziyeva et al. (2023)	Introduced a sophisticated ABM for regional pedestrian traffic analysis validated through empirical data.	Improves pedestrian traffic flow analysis and public space design.

.

Each paper is meticulously reviewed under the framework of three distinct research questions. The summaries of these reviews are presented in

Table 10. RQ 1: accuracy of ABM in simulating PD.

Study Reference	Urban Environment Context	Modeling Parameters	Software Platforms Used	Simulation Accuracy	Key Findings	Validation Methods	Data Sources	Recommendations
Asriana et al., 2021 [39]	Palembang, South Sumatra	Pedestrian sources, agents’ speed, behavior reactions	Grasshopper plugin (Rhino version 7)	High	Improved understanding of pedestrian movement patterns in tourism areas	Comparison with field observations	Field surveys, observations	Incorporate detailed agent interactions and environmental factors
Filomena and Verstegen, 2021 [59]	London city center	Road distance, angular change, landmark integration	GeoMASON simulation environment	High	Landmark-based navigation leads to more realistic pedestrian distributions compared to pure minimization models	Comparison with GPS trajectories	GPS trajectories, street segment volumes	Incorporate individual spatial knowledge differences, enhance cognitive modeling
Davidich et al., 2013 [62]	German railway stations	Waiting zones, pedestrian interactions	Cellular [50] automata	High	Standing pedestrians increase walking time by up to 20% during rush hour	Comparison with field measurements, video analysis	Field experiments	Incorporate standing pedestrians in simulations for realistic pedestrian flow models, especially in critical infrastructures.
Sinha et al. (2021) [50]	Passenger flow in terminal buildings	Agent-based modeling with subjective surveys and simulations	Anylogic (version not specified)	Limited to specific terminal layout, subjective survey bias	Demonstrates the importance of demographic attributes in ABM accuracy	field data obtained through quantitative and subjective surveys. Video footage of the check-in hall is used)	Field surveys, quantitative surveys for arrival rates, processing times, and waiting times,	need for dynamic internal heat gain estimates based on occupant density and dwell time distribution for accurate HVAC sizing and energy efficiency
Hänseler et al., 2016 [49]	Public walking areas	Macroscopic loading model for time-varying pedestrian flows	-	High	Accurate level-of-service prediction	Comparison with social force model	Train timetable, ridership information	Integration of train timetable essential for accuracy
Liu and Chen (2019) [51]	Metro stations in China	Destination choice, path planning, pedestrian dynamics	Custom ABM software	High; validated against real data	Agents choose optimal routes, impact of facility design	Comparison with video data, t-test for crowd density	Surveillance video from metro stations in China	Guide passengers to distribute between stairways and escalators, add more stairways or escalators to reduce overall consumed time
Martinez-Gil et al., 2017 [18]	Various urban environments	Path planning, congestion, lane formation	MARL-Ped	High	Developed model simulates human-like behaviors; robust in scaling scenarios by an order of magnitude	Fundamental diagrams, density maps, performance tests	Real pedestrian data with available researches.,	Further work on handling heterogeneous group sizes and mechanical responses
Sinha et al., 2021 [50]	Mid-sized airport terminal in India	Arrival rate, service time, dwell time, heat loss	Anylogic, TAITherm	91.76% (mean error 8.24%)	ABM coupled with thermo-physiological model provides realistic heat gains estimates	Comparison with field data, ANOVA, F-statistics	Field surveys, video recordings, airport management data	Consider dynamic heat gain for HVAC system optimization
Liu and Chen, 2019 [51]	Guanggu Square Station in Wuhan, China	Expected velocity, attractive force, destination choice, path planning	Not specified	Models simulate practical situation very well	Adding stairways or escalators can shorten overall consumed time; establishment of escalators increases time compared to stairways	t-test analysis, video data comparison	Surveillance video from busiest metro stations in China	Guide outbound passengers to use stairways or escalators homogeneously; inbound passengers to use escalators
Martinez-Gil et al., (2017) [63]	Simulated pedestrian groups in different urban scenarios	Learning rate, discount factor, state space features	Open Dynamics Engine	High (98.6% success in small scale)	Emergent collective behaviors such as roundabout movements; high accuracy in goal-reaching in small-scale experiments	Fundamental diagrams, density maps	Real data from previous studies (Seyfried, Weidmann)	Increase scenario dimensions for higher scalability, use RL techniques like reward shaping for improving performance
Hussein and Sayed (2019) [64]	Major street in downtown Vancouver, during a social event	Various parameters including prediction time, perception area, swerving distance, etc.	Not specified	High (Average location error: 0.28 m; Speed error: 0.06 m/s)	Model is capable of handling pedestrian interactions with high accuracy in various scenarios	Comparing actual and simulated trajectories	Video data collected during a social event in Vancouver	Continue examining model applicability in other environments and larger datasets; study group behavior and desired speed more precisely
Wu et al., (2016) [52]	Subway station walkway	Walking speed, occupied space, pedestrian types (P, F, O)	Custom simulation	High	P-pedestrians negatively affect flow; F-pedestrians positively affect flow until they exceed 80% of the crowd	Comparison of observed data with simulation results under homogeneous and heterogeneous conditions	Field data from Beijing Xizhimen subway station	Consider heterogeneity in pedestrian attributes for better capacity management and emergency planning
Zhou et al., 2021 [65]	Evacuation scenarios with visibility conditions	Visual information perception, path planning, obstacle detour	Various ML algorithms	High accuracy	Improved evacuation efficiency with global visual information by 6.3%	Experimental data	Pedestrian trajectory and social attributes data from evacuation drills	Increase guide resources near exits to divert crowd efficiently
Lim et al., 2018 [57]	Multi-ethnic trading port in 19th century	Neighborhood model, vision models, density-speed control model	Unity3D	Moderate to high	Cooperation among soldiers, competition among vendors, improved realism in multi-ethnic crowd simulation	Scenario-based visual observations	Historical records	Apply parameter adaptation through high-level controller to manage real-time changes in simulation
Zhang et al., 2015 [53]	Straight corridors and T-junctions	Density, flow, velocity	PeTrack	High accuracy for ρ < 3.5 m2	Measurement method influences results; Voronoi method provides fine structure	Empirical experiments	Series of controlled laboratory experiments	Fundamental diagrams can unify corridor widths into a single specific flow diagram. Tailor models to specific facility geometries. Consider entrance and exit widths to manage flow rates effectively.
Guo et al., 2012 [43]	Classroom with internal obstacles	Route distance, pedestrian congestion, route capacity	Not specified	High	Pedestrians prefer routes unoccupied by seats even if longer; efficiency improves with repeated exercises	Experiments, numerical simulations	Video recordings, experimental data	Incorporate dynamic learning and adaptation in evacuation drills; improve layout and exit positions
Shiwakoti et al., 2011 [45]	Various urban settings including panic conditions	Attraction and repulsion forces, impulsive forces, local interactive forces, collision and pushing forces	Custom simulation software	High accuracy for both non-panic and panic scenarios	Scaling of ant dynamics to human crowds shows consistent results	Empirical validation with experiments on Argentine ants and pedestrian flow data	Experiments with ants, pedestrian flow data from Duisburg-Essen University	Use of biological scaling concepts to improve ABM accuracy
Saloma et al., 2015 [46]	Group emergency evacuation using mice	Pool occupancy rate, individual training, group training	Not specified	High	Trained mice escaped 7× and 5× faster than untrained at occupancy rates of 11.9% and 4%, respectively.	Empirical experiments	Laboratory of Molecular and Cell Biology, UP Diliman	Prior individual training enhances evacuation efficiency; smaller groups are more effective for training
Garcimartín et al., 2015 [60]	Sheep herd passing through a bottleneck in a farm setting	Door size, presence of obstacle	Not specified	High	Widening doors and placing obstacles reduced clogging probability	Video recording and statistical analysis	Real-time video footage	Implement similar strategies in human crowd management to reduce clogging risks
Parisi et al., 2015 [47]	Controlled lab environment (ant arena)	Time lapses, velocities, densities	Custom software for image processing	High	Ants distribute uniformly over available space, no jamming or clogging observed. Faster-is-slower effect due to backward steps, not friction.	Comparison with Social Force Model simulations	Video recordings of ant experiments	Ants should not be used to model human behavior under emergency egress. Focus on human-specific models.
Lovreglio et al., 2014 [44]	Emergency evacuation in urban buildings	Exit choice, crowd behavior, proximity	FDS + Evac	High	Influence of group dynamics, herding behavior, cooperative/selfish behavior	Sensitivity analysis, behavioral analysis	Online survey	Further experimental research to understand psychological and environmental factors
Auld et al., 2016 [54]	Chicago metropolitan area	Dynamic activity generation, within simulation activity attribute planning, and detailed activity scheduling model	Polaris, Medina, MN, USA	High	The POLARIS ABM effectively models large-scale transportation networks and integrates demand and network modeling aspects.	Calibration against observed data; comparison of network loading characteristics	Chicago travel survey data; historical traffic incident data	Improve computational efficiency; Enhance the model for policy analysis
Kaziyeva et al., 2023 [56]	Salzburg city and adjacent municipalities	Activity type, mode, route choices	GAMA, Brussels, Belgium	Moderate to high	Walkability-based routing improves traffic distribution; model under-represents central traffic	Comparison with empirical data, Spearman’s and Pearson’s correlation, MAE	GNSS trajectories, mobility surveys, OpenStreetMap	Further focus on spatial psychology and sociodemographic differences
Lei Ma et al., 2023 [66]	Campus of University of Gävle, Gävle Hospital	Angle and depth of vision, affordance, visit frequency	Not specified	High	Paths emerged from interactions, angle impacts path pattern	Comparison with observed paths	Field survey, observed footprints	Incorporate visual parameters and environmental heterogeneity

,

Table 11. RQ 2: challenges and limitations of ABM.

Study Reference	Key Challenges	Limitations Identified	Suggested Solutions	Modeling Accuracy	Implementation Issues	Case Study/Scenario
Asriana et al., 2021 [39]	Complexity in simulating diverse pedestrian behavior	Limited real-time data for validation	Integrate more real-time data sources, enhance agent interaction models	Medium	Handling diverse tourist behaviors	Palembang, South Sumatra
Filomena and Verstegen, 2021 [59]	Cognitive complexity, data availability	Difficulty in modeling cognitive representations, data integration challenges	Enhance cognitive modeling, integrate varied data sources	High	Computational effort, data quality	London city center
Davidich et al., 2013 [62]	Inclusion of waiting pedestrians, model validation	Limited empirical data for waiting zones, computational complexity	Use empirical data for calibration, optimize model algorithms	High	Data collection and processing	German railway stations
Hänseler et al., 2016 [49]	Data variability in pedestrian dynamics	Limited data availability, heterogeneous pedestrian behavior	Use of multiple data sources for reliability	High for dimensioning purposes	High cost of data collection, sensor placement challenges	Lausanne railway station
Liu and Chen (2019) [51]	High crowd density, realistic modeling of pedestrian behavior	Limited by video data quality, legal constraints on site shooting	Use improved models considering multiple factors such as convenience and queuing	Better than classical models high accuracy in practical simulation	Data extraction and processing from surveillance videos	Metro stations in China
Sinha et al., 2021 [50]	Dynamic passenger behavior, variable heat gains	Standard models overestimate/underestimate heat gains	Integrate dynamic activity sequences into ABM	High (mean error 8.24%)	Requires detailed passenger data	Mid-sized airport terminal in India
Martinez-Gil et al., 2017 [63]	Scaling up the number of agents, emergent behaviors	Low percentage of agents reach goals in large scales	Learning by examples, reward shaping, policy shaping	Reduced in high-density scenarios	Ensuring consistency in successful simulations	Multiple scenarios
Hussein and Sayed (2019) [64]	Complex pedestrian movements and interactions, frequent speed and direction changes	Complexity in calibrating model parameters	Use of Genetic Algorithms for calibration	High (accuracy varies from 87% to 100%)	Validation limited to one location	Downtown Vancouver during a social event
Wu et al., 2016 [52]	Modeling heterogeneity in pedestrian dynamics; managing large-scale simulations	High proportion of pedestrians decreases capacity; oversimplification of individual behaviors	Improved floor field model incorporating heterogeneity parameters	High	Difficulty in data collection for accurate heterogeneity parameters	Subway station pedestrian flow
Zhou et al., 2021 [42]	Visual occlusion by obstacles, data dependency	Limited real-time data on pedestrian movements	Collect more detailed pedestrian data	Affected by visual occlusion	Lack of real-time pedestrian movement data	Pedestrian evacuation with various visibility conditions
Lim et al., 2018	Real-time parameter adaptation	Computational overhead	Use high-level controller	High	Real-time simulation challenges	Multi-ethnic trading port
Zhang et al., 2015 [53]	Measurement method variability	High fluctuations with some methods	Use Voronoi method	High for Voronoi method, less for others	Differences in measurement methods affect results	Pedestrian flow in corridors and T-junctions
Guo et al., 2012 [43]	Route-choice behavior under low visibility	Limited to specific classroom setup	Improve model generalizability and flexibility	Moderate	Complexity in modeling pedestrian interactions	Classroom with internal obstacles
Shiwakoti et al., 2011 [45]	Lack of human panic data, complexity of human interactions	Scarcity of panic data, difficulty in measuring certain parameters	Use of ant behavior as a model, empirical validation with ants	High for panic scenarios based on biological scaling	High computational requirements, parameter estimation challenges	Panicking Argentine ants, human crowd simulations
Saloma et al., 2015 [46]	Ethical issues with human participants	Small-scale experiments may not capture large crowd dynamics	Use of animal models like mice to simulate human behavior	High	High effort in training animals	Emergency evacuation in a controlled environment
Garcimartín et al., 2015 [60]	Collecting real-world data for validation of ABM models	Ethical concerns in conducting human experiments	Use of animal models (e.g., sheep) as proxies	High	Feasibility of data collection	Sheep herd in farm setting
Parisi et al., 2015 [47]	Differences between ant and human behavior in egress situations	Ants do not jam or clog like humans	Avoid using ants to model human egress	High	Citronella concentration affecting sensory and motor systems of ants	Egress in controlled ant arena experiments
Lovreglio et al., 2014 [44]	Modeling heterogeneous decision-maker behavior	Limited by homogeneity in sample demographics	Integration of revealed preferences into real/simulated emergencies	Moderate	Online surveys may not replicate real emergency stress	Emergency evacuation
Auld et al., 2016 [54]	Scalability for large-scale systems	High computational resource requirement	Use of fast shared memory approach; Multi-threading	High	High demand for allocations/deallocations of homogeneous objects	Chicago metropolitan area
Kaziyeva et al., 2023 [56]	Under-representation of central traffic, lack of spatial psychology data	Insufficient representation of small-scale mobility, absence of pedestrian access information in OSM	Incorporate walkability scores, detailed spatial psychology indicators, better data on pedestrian access	Moderate to high	Lack of high-quality, up-to-date input data, computational intensity	Regional traffic in Salzburg city and adjacent municipalities
Lei Ma et al., 2023 [66]	High computational complexity, integrating granular visual parameters	Difficulty in incorporating detailed visual parameters	Simplify models while retaining critical visual factors	High	Computational demands, parameter sensitivity	University of Gävle, Gävle Hospital

and

Table 12. RQ 3: insights from ABM simulations for urban planning.

Study Reference	Case Study/Application	Simulation Objectives	Key Outcomes	Impact on Urban Planning/Design	Policy Implications	Key Metrics/Indicators	Tools/Techniques Used
Asriana et al., 2021 [39]	Palembang, South Sumatra	Develop design strategy for pedestrian behavior in tourism areas	Better understanding of pedestrian movement, improved walkability	Inform urban design and planning for tourism areas	Recommendations for pedestrian zoning, facility placement	Pedestrian flow, density, connectivity patterns	Grasshopper plugin for ABM simulation
Filomena and Verstegen, 2021 [59]	London city center	Evaluate effect of landmarks on pedestrian dynamics	More realistic pedestrian distribution, enhanced urban design	Supports integrated urban design incorporating landmarks	Recommendations for integrating landmarks in planning	Pedestrian volumes, route diversity, landmark usage	GeoMASON simulation environment
Davidich et al., 2013 [62]	German railway stations	Assess impact of waiting zones on pedestrian flow	Waiting zones increase walking time by up to 20% during rush hour	Identify critical areas for infrastructure improvement	Recommendations for infrastructure design, congestion management	Walking time, pedestrian density, flow disruption	Cellular automata
Hänseler et al., 2016 [49]	Lausanne railway station	Estimate pedestrian origin–destination demand	Accurate prediction of level of service	Improved design and dimensioning of facilities	Guidelines for infrastructure development	Level-of-service, walking times	Pedestrian traffic assignment model
Liu and Chen (2019) [51]	Guanggu Square subway station, Wuhan, China	Optimize passenger flow, reduce overall consumed time	Improved passenger distribution, reduced congestion	Better facility design, enhanced passenger guidance	Improve infrastructure to handle high density	Overall consumed time, crowd density	ABM simulation, social force model
Martinez-Gil et al., 2017 [63]	Various urban scenarios	Assessing robustness and scalability of MARL-Ped	Emergent behaviors consistent with real data	Potential for designing better pedestrian flow systems	Evaluating new urban designs based on realistic simulations	Speed, density, goal-reaching success rates	MARL-Ped, fundamental diagrams, density maps
Sinha et al., 2021 [50]	Airport terminal building	Estimate dynamic heat gains from passengers	Realistic heat gain estimates, impact of activity sequences	Improved HVAC sizing, optimized energy usage	Consideration of dynamic activity sequences in HVAC standards	Sensible and latent heat loads, occupancy profiles	Anylogic, TAITherm
Martinez-Gil et al., 2017 [63]	Four-way intersection (4WI), free field (FF)	Analyze emergent behaviors, assess scalability	Emergent behaviors like roundabout movement, high accuracy in small-scale simulations	Provides insight into pedestrian flow management in complex scenarios	Supports development of more efficient pedestrian facilities	Number of agents reaching goals, density maps	Multi-agent reinforcement learning, Open Dynamics Engine
Hussein and Sayed (2019) [64]	Pedestrian movement in downtown Vancouver during a social event	Simulate pedestrian interactions in a crowded environment	High accuracy in reproducing pedestrian behavior during different interactions	Useful for pedestrian safety studies and large event planning	Enhance pedestrian facilities for better safety and satisfaction	Average location and speed errors	Genetic Algorithm, Computer Vision
Wu et al., 2016 [52]	Subway station walkway	Analyze the effects of pedestrian heterogeneity on flow dynamics	Pedestrians reduce flow capacity; pedestrians increase capacity until saturation point	Understanding pedestrian heterogeneity helps design walkways to optimize flow and prevent bottlenecks	Guidelines for pedestrian management in public transit facilities	Capacity (pedestrians/m2·s)	Improved floor field CA model incorporating heterogeneity
Zhou et al., 2021 [65]	Evacuation scenarios with visibility conditions	Improve evacuation efficiency	Efficiency increased by 6.3%	Better design of evacuation routes	More efficient crowd management policies	Evacuation time, pedestrian distribution	Machine learning algorithms, visual information perception
Lim et al., 2018 [57]	Multi-ethnic trading port simulation	Recreate historical interactions	Realistic multi-ethnic behaviors	Improved understanding of historical interactions	Insights for cultural heritage	Interaction frequencies	Unity3D, high-level controller
Zhang et al., 2015 [53]	Pedestrian dynamics in corridors and T-junctions	Analyze flow and density relationships	Fundamental diagrams differ by geometry	Different planning needed for varying corridor widths	Ensure adequate corridor widths to prevent flow issues	Density, flow, velocity	PeTrack, Voronoi diagrams
Guo et al., 2012 [43]	Classroom evacuation	Evaluate pedestrian route choice under various visibility conditions	Pedestrians follow shortest path; prefer unoccupied routes	Improve internal layout designs for better evacuation efficiency	Design evacuation plans that consider visibility	Evacuation time, route selection, pedestrian density	Microscopic pedestrian model, cellular automata
Shiwakoti et al., 2011 [45]	Simulation of pedestrian egress under panic conditions	To model collective pedestrian dynamics, validate with non-human entities	Effective scaling from ants to humans, consistent evacuation patterns	Improved design strategies for emergency egress, insights into structural influences on flow	Potential for enhanced safety regulations and building codes	Evacuation times, flow rates, headway distributions	Custom simulation framework, empirical data integration
Parisi et al., 2015 [47]	Ant egress in controlled lab environment	Study the distribution, velocities, and densities of ants under stress	Uniform distribution of ants leads to efficient evacuation without jamming	Highlight differences between ant and human behavior in emergencies	Reconsider the use of ants for human egress modeling	Density maps, time lapses, velocities	Custom image processing software
Lovreglio et al., 2014 [44]	Emergency evacuation modeling	Understanding exit choice behavior	Influence of exit proximity and crowd behavior	Insights into designing safer evacuation routes	Evacuation policy	Decision-maker characteristics (age, height, education)	FDS+Evac
Auld et al., 2016 [54]	Chicago metropolitan area	Evaluate the benefit of ITS infrastructure	Improved network performance	Enhanced capability for evaluating network operational improvements	Evaluation of human-in-the-loop TMC operational strategies	Traffic density; Average speed; Flow rate	POLARIS; Newell’s simplified kinematic waves traffic flow model
Kaziyeva et al., 2023 [56]	Salzburg city and adjacent municipalities	Simulate pedestrian traffic flows over a day	Improved traffic distribution with walkability-based routing, moderate to high accuracy	Supports planning strategies with insights on traffic patterns and pedestrian flows	Design of urban spaces for better walkability	Daily and hourly counts, trip distances	Agent-based modeling, probabilistic rule sets
Lei Ma et al., 2023 [66]	University of Gävle, Gävle Hospital	Understand path emergence	Paths align with real-world data, angle impacts efficiency	Helps in predicting pedestrian paths, improving design	Promotes consideration of pedestrian behaviors	Accuracy of simulated paths	Agent-based modeling software

to address the research questions comprehensively.

4. Discussion

This discussion shed lights on the effective utilization of ABM in the urban panning spectrum with its pros and cons, including methodologies, limitations, and future research possibilities. Simulation capacities basically lie in the pedestrian dynamics-related context in an urban context, where the most advanced technology has influenced robust interventions in fields such as transportation hubs, emergency evacuation, public spaces, pedestrian behavior under panic situations, and various corridor configurations.

ABM is capable of handling agents with clear rules, and their interaction has proven very effective in capturing pedestrian movements in urban environments. López Baeza et al. (2021) [41], Gabriele F et al. (2019) [58], and Kostas Cheliotis (2020) [14] ascertained precise real-world results in pedestrian flaws related to pedestrian activity and safety. Moreover, highly accurate urban planning decisions have been made by integrating sophisticated algorithms with route choice-based landmarks.

This study analyzed the number of innovative methods that led the advancement of ABM with better applicability and capability. Ref. [41] simulated the pedestrian behavior in an urban environment and proposed required modifications to increase the activity levels in the same environment. This was enhanced to simulations of real-world scenarios by facilitating urban planning options. Therefore, it is clearly proven that the various space configurations can lead designers to identify flows and make decisions accordingly. When it comes to the pedestrian flow in transportation hubs, ABM has been used to identify demand estimates and traffic assignment models related to pedestrian dynamics in train stations, giving more priority to understanding pedestrian flow-based designing and planning [49]. Moreover, it is very interesting to note the integration of ML and ABM to simulate pedestrian flow in evacuation planning. Visual information and environmental cues in identifying evacuation patterns have become instrumental in saving lives in these scenarios [65].

The empirical basis of ABM has been truly enhanced by the use of spatiotemporal trajectory data and pedestrian volume measurements. Furthermore, the importance of various other environmental factors, including individual behaviors and demographic attributes, have taken pedestrian dynamics to the next level.

Researchers have identified key practical implications related to the urban planning and designing perspective, as follows.

Efficient and sustainable public space: Planners receive better insight on pedestrian behavior, which can practically support creating better spaces to improve overall user experience and accessibility, while maintaining security aspects, improving the flow with efficient pedestrian pathways to reduce congestion and to maintain the smooth functioning of enhanced public areas [67].

Crowd Management: Planning of large events with ABM is commendable from individual and group perspectives. Hence, urban planners and urban managers utilize the simulations to control the overall city management with crowd control and also emergency evacuation-related planning [68].

Policy Development: During policy formulation, it is very important to understand the walkability, traffic conditions, and overall urban mobility to formulate necessary policies such as use of public transportation, promoting walkability, or promoting cycling, etc., based on ABM simulations [69].

Sustainable Urban Planning: ABM mostly enables trial and error without any cost or risk in different scenarios. Therefore, modeling and implementing concepts in real situations with environment-friendly and sustainable options have become possible [70].

4.1. Limitations

While ABM demonstrates substantial potential, certain limitations and challenges remain. These include the calibration and validation of models, the computational intensity of simulating large populations, and the need for accurate and comprehensive datasets for model parameterization. Moreover, as highlighted by studies such as those by Zi-Xuan Zhou et al. (2021) [65] and Mohamed Hussein and Tarek Sayed (2019) [64], the variability in human behavior and the dynamic nature of pedestrian flows in complex environments pose ongoing challenges to ABM’s predictive accuracy. Further, J Zhang et al. (2015) [53] highlighted the challenges of model scalability.

The availability of data related to targeted journals and databases is the most significant limitation of this SLR due to diverse built environment implementations in constructions and human psychology, which necessitated some rules in selecting related articles. Another disadvantage is the bias in selecting articles, SLRs, and surveys. However, the selection of studies published in English was a minor constraint.

4.2. Future Recommendations

This section outlines several key future research directions aiming to enhance the scope of ABM in the field of PD. The researchers’ understanding about urban environment and PD explored in this paper showcases significant potential in practical urban planning decisions. As cities continue to evolve, it is imperative to explore new research avenues that can further harness the capabilities of ABM.

Proposed future research directions suggested by the researchers include integrating advanced data analytics and real-time data, comprehensive calibration and validation techniques, exploring human-centric urban design, incorporating cognitive and behavioral modeling, leveraging machine learning for enhanced predictive capabilities, addressing the impacts of micro-mobility and pandemics, and developing real-time decision support systems, explained in separate sections.

In addition, the researchers also identified some emerging possibilities to integrate ABM in the fields of urban digital twins, application of augmented reality (AR) and virtual reality (VR) technologies, exploring social and ethical implications of ABM in PD, environmental impact modeling, and traffic management systems, and adopting of cross-disciplinary approaches concerning ABM and PD.

The following

Table 13. Summary of Future Recommendation.

Research Direction	Current Focus	Expanded Suggestions	Examples	Suggested Tools/Processes
Enhancing ABM Simulations through Advanced Data Analytics and Real-Time Data	Integrating advanced data analytics and real-time data for robust simulations.	Utilize sensor networks and IoT for real-time updates. Incorporate dynamic environmental factors (e.g., noise, temperature).	Integrating real-time pedestrian data from GPS tracking enhances simulation accuracy [49]. Spatial simulations considering noise and temperature comfort can provide detailed insights into pedestrian dynamics.	IoT Platforms (e.g., AWS IoT Core v1.0.2, Azure IoT Hub v2.8.0) for collecting real-time data. GIS Software (e.g., ArcGIS Pro 2.8) for spatial data integration. Real-Time Data Analytics Tools (e.g., Apache Kafka 2.7.0) for processing streaming data.
Comprehensive Calibration and Validation Techniques	Calibrating and validating ABM using real-world data and site surveys.	Develop standardized validation protocols for different settings. Use of video surveillance for behavior analysis.	Validating ABM with observational data from urban parks improves model accuracy [41]. Site surveys can be used to enhance calibration and validation of ABM.	Validation Frameworks (e.g., PRISMA 2020 for systematic reviews). Computer Vision Software (e.g., OpenCV 4.5.1) for analyzing video surveillance data. Survey Tools (e.g., SurveyMonkey v3.5) for collecting site-specific data.
Exploring Human-Centric Urban Design	Using ABM to assess impacts of built environment changes on pedestrian behavior, safety, and well-being.	Evaluate the impact of specific urban design interventions. Adapt urban designs to changing demographics using ABM.	ABMs predicting pedestrian responses to urban design changes enhance safety and accessibility [65]. Dynamic adaptation of urban plans can be informed by real-time pedestrian data and ABMs.	Urban Design Simulation Tools (e.g., Rhino 7, AutoCAD 2023) for modeling design changes. Demographic Analysis Tools (e.g., SPSS 27, R 4.0.3) for studying population impacts. Decision Support Systems (e.g., GIS-Pro 2.8, QGIS 3.18) for real-time urban planning.
Incorporation of Cognitive and Behavioral Modeling	Developing ABMs that include cognitive and behavioral aspects to simulate realistic decision-making.	Integrate psychological models to simulate emotions and cognitive processes. Model social interactions and group dynamics.	Behavioral models in ABM simulate pedestrian decisions during emergencies, providing insights into stress responses [65]. ABMs can simulate decision-making processes influenced by individual preferences and social influences.	Behavioral Simulation Software (e.g., AnyLogic 8.7, NetLogo 6.2.0) for modeling cognitive processes. Social Network Analysis Tools (e.g., Gephi 0.9.2) for understanding interactions. Psychological Modeling Frameworks (e.g., PECS 2.0) for integrating human behavior.
Leveraging Machine Learning for Enhanced Predictive Capabilities	Enhancing ABM’s predictive capabilities of pedestrian dynamics through machine learning.	Automate pattern recognition with machine learning. Use historical data to improve ABM predictive accuracy.	Machine learning algorithms refine ABM predictions in complex evacuation scenarios [65]. ML can assist in capturing complex behavioral patterns and improving simulation accuracy.	Machine Learning Libraries (e.g., TensorFlow 2.4.1, PyTorch 1.8.0) for developing predictive models. Data Analysis Tools (e.g., Pandas 1.2.1, NumPy1.20.1) for processing historical data. Pattern Recognition Software (e.g., MATLAB R2021a) for detecting and modeling behavioral patterns.
Addressing the Impacts of Micro-Mobility and Pandemics	Assessing how micro-mobility solutions and health crises like pandemics impact pedestrian dynamics.	Develop ABM simulations for pandemic conditions (e.g., social distancing). Evaluate micro-mobility’s impact on pedestrian dynamics.	ABM during COVID-19 shows the impact of social distancing on pedestrian movement (pandemic research). Simulating micro-mobility options like scooters and bike-sharing (micro-mobility research).	Pandemic Simulation Models (e.g., SEIR 1.1.0models) for studying health crises. Micro-Mobility Planning Tools (e.g., Bike-Share Analysis Tools 3.2) for evaluating impact on pedestrian dynamics. Public Health Data Integration (e.g., using data from WHO, CDC) for accurate modeling of pandemic impacts.
Development of Real-Time Decision Support Systems	Developing real-time decision support systems for urban planners and emergency responders leveraging ABMs.	Implement adaptive response mechanisms for urban events and emergencies. Develop crisis management tools for real-time insights.	Real-time ABM assists in managing pedestrian flows during peak hours in transportation hubs. Decision support systems leveraging ABMs can provide immediate insights during critical events.	Decision Support Software (e.g., DSS tools like ArcGIS Pro 2.8 for real-time urban planning). Crisis Management Platforms (e.g., Everbridge 2023, RapidSOS 2023) for emergency response. Adaptive Control Systems (e.g., using AI-based adaptive traffic management systems) for real-time event handling.
Integration with Urban Digital Twins	ABM used independently to simulate pedestrian dynamics.	Integrate ABM with urban digital twins for dynamic, real-time city models.	Urban digital twins enhance the realism and applicability of ABM simulations.	Digital Twin Platforms (e.g., Bentley Systems 2023, Siemens CyPT 3.1) for real-time data integration. Data Integration Tools (APIs 2023 and data services for syncing real-world data).
Augmented and Virtual Reality (AR/VR) Applications	Traditional visualization methods (e.g., 2D maps, graphs).	Use AR/VR to visualize ABM simulations in immersive, 3D environments.	AR/VR provides an intuitive understanding of pedestrian dynamics ([AR/VR simulation examples]).	AR/VR Platforms (e.g., Unity 2023, Unreal Engine 5) for developing immersive experiences. Visualization Tools (e.g., Autodesk Revit 2023) for 3D modeling.
Ethical and Social Implications of ABM	Focus on technical aspects and practical applications.	Explore ethical and social implications of ABM in urban planning.	Considering privacy, data security, and social equity ensures responsible use of ABM.	Ethical Frameworks (e.g., IEEE Global Initiative on Ethics) for guiding responsible AI and data use. Social Impact Assessment Tools for evaluating urban planning decisions.
Sustainability and Environmental Impact Modeling	Studies focus on pedestrian dynamics without strong sustainability emphasis.	Integrate sustainability metrics and environmental assessments into ABM.	Sustainability metrics in ABM support eco-friendly urban planning.	Sustainability Assessment Tools (e.g., SimaPro 9.3) for environmental impact analysis. Green Infrastructure Modeling (e.g., ENVI-met 4.4) for simulating environmental benefits.
Adaptive and Predictive Traffic Management Systems	ABM simulates pedestrian behavior in static or controlled environments.	Develop systems that use ABM to predict and respond to real-time traffic and pedestrian flow changes.	Adaptive traffic management systems enhance urban mobility.	Adaptive Traffic Control Systems (e.g., IBM’s Traffic Prediction Tool 2.1) for dynamic traffic management. Predictive Analytics Platforms (e.g., SAS 9.4) for forecasting flow patterns.
Cross-Disciplinary Approaches	ABM applied within specific domains like urban planning or transportation.	Explore cross-disciplinary applications of ABM, integrating insights from sociology, economics, public health, etc.	Cross-disciplinary approaches provide a holistic understanding of pedestrian dynamics.	Cross Disciplinary Collaboration Platforms (e.g., ResearchGate 2023) for interdisciplinary research. Integrated Simulation Environments (e.g., AnyLogic 8.7) for multi-domain simulations.

provides a comprehensive overview of these future research directions, including the current focus, expanded suggestions, practical examples, and suggested tools or processes to explore these avenues. This comprehensive guide aims to inspire and direct future studies, encouraging the continued advancement of ABM in creating safer, more efficient, and more inclusive urban environments.

5. Conclusions

The advancement of technology and the evolution of human psychology and behaviors significantly influence industrial and commercial projects, underscoring the need for dynamic and nuanced approaches to urban planning and PD analysis. This study embarked on a comprehensive SLR with the primary objective of examining pedestrian dynamics within the built environment, focusing on pedestrian modeling, computational techniques, and the tools utilized in model development. The investigation into ABM revealed its profound efficacy in understanding and simulating PD, marking a significant contribution to the field by offering insights into pedestrian behavior in microscopic environments and demonstrating the potential for achieving optimal outcomes from the datasets used for model development.

Contrary to previous studies that explored PD and the built environment using alternative modeling approaches, this SLR uniquely underscores the utility of ABM in developing detailed pedestrian dynamics simulations. It brings to light the adaptability of ABM in calibrating, validating, and verifying models to meet high accuracy levels, a critical aspect in modeling pedestrian behaviors in various urban settings. This study thus stands out for providing an extensive insight into the application of ABM, showcasing its unparalleled ability to model complex pedestrian interactions and movements with precision.

For researchers and practitioners engaged in a wide array of industrial and commercial projects, this SLR offers invaluable guidance on the latest tools and computational techniques best suited for different environments, scales, and observation contexts using ABM. It acknowledges, however, the challenges associated with employing specific modeling tools across varied instances of the built environment. For example, while NetLogo presents difficulties in micro-level application, CA models show limitations in accommodating multi-agent pedestrian simulations, and Unity 3D may hinder the data fetching process due to its computational intensity.

Despite these challenges, the rigorous process and broad scope of this SLR emphasizes the value of ABM in studying, designing, implementing, and visualizing pedestrian dynamics across any built condition or place. It highlights the necessity of continuing to refine and adapt ABM techniques to overcome existing limitations and fully harness the potential of this modeling approach in contributing to more informed, effective, and human-centric urban planning and design. The integration of ABM into the analysis of pedestrian dynamics within the built environment promises to enrich understanding and management of urban spaces, ensuring environments better tailored to the evolving needs and behaviors of their inhabitants. Despite its challenges, the continued evolution of ABM methodologies holds the promise of even more sophisticated and accurate simulations, guiding the future of urban design towards more human-centric environments.