1 Introduction

Traffic congestion is one of the main factors we need to pay attention in order to maintain sustainable transportation in urban areas, as it has already led to a slew of other significant issues, such as air pollution, climatic changes, high energy consumption, and waste of time. Therefore we can identify that traffic congestion has substantial impacts on society and the economy. Promoting sustainable mobility is one of the most widely held goals in the world of transportation policy at the moment [1]. Micro-mobility can be considered a rapidly expanding trend in urban mobility, which can help address many of the current transportation-related issues affecting cities worldwide [2].

With the rise of micro-mobility, urban transportation systems have undergone rapid changes in recent years. Micro-mobility has the potential to be a great sustainable transportation solution for traffic congestion. According to research data, the majority of micro-mobility trips appear to be short in distance and duration [3]. Micro mobility is defined as a lightweight, small-scale transport device weighing less than 350 kg and moving at a speed less than 45 kmph [4]. Examples of micro-mobility devices include bicycles, electric scooters, electric bikes, hoverboards, segways, and electric skateboards, all of which can be either human or electric-powered and are either privately owned or shared. It is expected that encouraging micro-mobility devices and systems will aid in the resolution of traffic congestion and air pollution. Furthermore, micro-mobility modes are well-known for their ease of use, less-noise, ease of finding parking space, energy efficiency, and, most significantly, environmental friendliness with no toxic vehicle exhaust. Micro-mobility has grown in popularity in recent years as a shared mode of transportation that can be reserved using applications on connected mobile phones and other mobile devices thanks to advances in mobile computing. The popularization of micro-mobility, particularly for first and last kilometer travel, has also aided in the publicity of active modes of transportation, which have health benefits for city dwellers [5]. People can now find both traditional micro-vehicles like bicycles and modern ones like e-scooters, e-bikes, and so on in many areas across the world.

The main advantages of using micro-mobility include reducing traffic congestion, being one of the most environmentally friendly modes of transportation, and providing low-cost personal transportation. Micro-mobility devices such as e-scooters and e-bikes are significantly less expensive than purchasing or renting a car. Furthermore, e-scooter owners save money on maintenance and fuel because e-scooters and other micro-mobility devices have fewer parts and do not use fuel [6]. Even though micro-mobility has so many benefits as mentioned above, in real-life scenarios, micro-vehicles can cause issues that affect the safety on roads where the traffic flows are not properly handled since it can increase the complexity of interactions between pedestrians, cyclists, and other transport modes such as motor vehicles [4]. There are numerous safety precautions to take when using micro-mobility because micro-mobility devices pose significant safety risks, which is why many cities and governments are still hesitant to promote them. Since micro-mobility devices do not provide in-built protection, such as car airbags, some consider using those devices might be risky. Therefore, it is important to emphasize that it is necessary to estimate the traffic flows and configuration prior to introducing such a micro-mobility solution to a city. For this purpose, traffic simulation tools play a significant role in traffic management, considering the demand and the road network infrastructure. As a result, it appears that new policies and infrastructure will be required in society when introducing micro-mobility as a solution to traffic congestion in urban areas.

1.1 Our contribution

Sustainable mobility is a top transportation policy priority worldwide and remains an ongoing challenge. Agent-based modeling has revealed numerous practical applications in the field of traffic and transportation. It is a modeling and computational framework for simulating dynamic processes involving autonomous agents and offers previously unavailable tools for modeling individual behaviors and how these behaviors interact with their environment. Thus, agent-based modeling provides an effective means of representing various traffic elements, whether they are pedestrians or vehicles, making it particularly adept at capturing traffic dynamics [7].

This paper provides an overview of the literature on agent-based models for the traffic simulation including micro-mobility. Many studies have been conducted to control traffic in urban areas utilizing micro-mobility and agent-based traffic simulation tools. This review aims to analyze and evaluate some of those case studies and experiments conducted in various locations across the world to simulate different types of traffic congestion. To the best of our knowledge, a thorough review of agent-based models for traffic simulation, particularly in the context of micro-mobility, has been notably absent in the existing body of literature.

1.2 Organization of the paper

Section 2 discusses the fundamental concepts of agent-based models as well as associated keywords of the review paper, such as what an agent-based model is and its elements, and how to apply agent-based models to transportation systems and micro-mobility. Section 3 discusses the methodology of the research, where the Table 1 divides the selected papers into three categories, considering the purpose of the study; (1)– mixed traffic simulation, (2)– micro-mobility traffic simulation, and (3)– other applications. Section 4 discusses the methods of analysis, where the Table 2 summarizes the selected case studies and provide a comprehensive analysis. Section 5 presents a comprehensive discussion. Section 6 discusses on how to model the “last-mile”, highlighting the advantages and limitations of the agent-based models for the micro-mobility simulations, the mixed traffic simulations and the other applications. Finally, section 7 concludes the paper, identifying the main gaps in knowledge and potential future research directions.

2 Agent-based modeling (ABM)

Agent-based modeling (ABM) is a modeling and computational framework for simulating dynamic processes involving autonomous agents. ABMs offer previously unavailable tools for modeling individual behaviors and how those behaviors interact with it’s environment. The rise in interest in ABM is evident in the number of articles published in modeling and applications journals. There has also been a rise in demand for courses and training programs in this area [8].

The conceptual aspect of many earlier agent-based models made them ideal for theory development and illustration. They were deterministic in nature, disregarding stochastic components, and not calibrated to real-world data. The field is now moving towards building large-scale systems designed to supply data for analysis and decision making [9]. Some of the applications of ABM include forecasting the spread of diseases and analyzing the performance of marketing campaigns [8].

In ABM, the system is developed with a bottom-up approach. One of the key components of an ABM is modeling a population of autonomous agents that interact frequently, and each has its own attributes. Agents can be programmed to act in a variety of human-like, autonomous ways, interacting with one another and with real-world actors [10].

2.1 Elements of an agent-based model

According to Macal and North [8], there are three elements in a typical agent-based model:

  1. 1.

    Agents,

  2. 2.

    Agent relationships, and

  3. 3.

    Agents’ environment.

In order to design an agent-based model, a model developer has to recognize, simulate, and implement all the aforementioned components.

2.1.1 Agents

The precise definition of the term “agent” in the context of ABM has been a subject of ongoing debate and discussion, as noted by Chen in his work [7]. This debate extends beyond academic circles, often emerging when individuals assert that their models are agent-based or when there’s a need to assess the validity of similar claims made by others.

Agents are distinct entities that are self-contained, capable of making autonomous decisions, and capable of engaging with other agents and their surroundings. They are regulated by a set of behaviors known as rules, which describe how they engage with one another and with the surroundings. Real-world behaviors can be used to generate these rules [11].

In practical modeling terms, Macal and North [8] have put forth a framework that defines agents based on specific properties and attributes, providing a more concrete perspective.

Autonomy An agent possesses autonomy and self-guidance. It has the capability to operate independently within its environment and engage with other agents, typically within predefined circumstances. When discussing an agent’s behavior, we refer to the mechanism that links the information it acquires from its surroundings and interactions to its decision-making and subsequent actions.

Modularity An agent represents a distinct, discrete entity with its own set of attributes, behaviors, and the ability to make decisions. The concept of modularity underscores the existence of a clear boundary for an agent, making it straightforward to determine whether a given element belongs to an agent or constitutes a shared trait among all agents.

Sociality An agent is a social entity that engages with other agents through various protocols such as competing for space, avoiding collisions, recognizing other agents, exchanging information and communication, exerting influence, and using specific mechanisms related to a particular domain or application.

Conditionality An agent’s state is a representation of its condition that is defined by the critical variables associated with its present circumstances, that changes over time. The behavior of an agent is influenced by its state. If an agent can assume a wide range of potential states, it can exhibit a more diverse set of behaviors.

Furthermore, agents typically possess other properties, which may or may not be necessary for the agency to function effectively. As an example, an agency may have explicit objectives that direct its behavior, and an agent may possess the capacity to learn and adjust its behaviors based on its experiences. At the individual level, learning and adaptation can be modeled as agent behaviors. Adaptation can also be modeled at the population level, or by allowing agents to join or leave the population.

2.1.2 Agent relationships

ABM concerns itself with modeling agent relationships and agent interactions. Identifying who is or may be connected to whom is the main challenge in simulating agent interactions. The fundamental principle of an agent-based model is that agents can only interact with a finite number of other agents at any given moment, regardless of the agent-interaction topology used to link the agents. This idea is put into practice by designating a local neighborhood and restricting agent interaction to a few [8].

2.1.3 Agents’ environment

The environment establishes the conditions surrounding agents’ existence, where agents engage both with their surroundings and fellow agents [12]. According to Macal and North [13], the environment can provide valuable information regarding the spatial interactions among agents, including access to comprehensive geographical data akin to a Geographic Information System (GIS). In certain scenarios, it becomes essential to track dynamic changes in agent positions, especially when assessing their movements across landscapes, competition for resources, and various situational contexts. Additionally, the environment may impose constraints on agent activities. For instance, in an agent-based transportation model, the environment encompasses factors like road network node and link capacities, which influence travel speeds and regulate the number of agents permitted to traverse the network simultaneously, as discussed by [14]. According to Bandini et al. [14], in the specific context of simulation, the environment is typically responsible for,

  • Depicting, actualizing, and overseeing the layout of the physical or social configuration of the entire system,

  • Incorporating and facilitating controlled access to elements and components of the system not represented as agents,

  • Facilitating agent perception and actions in specific contexts,

  • Sustaining internal dynamics within the system, and

  • Establishing and enforcing rules governing the system’s behavior.

2.2 ABM and transportation systems

ABM has discovered numerous practical applications in the realm of traffic and transportation. Urban traffic plays a pivotal role in city functions and significantly influences the quality of life and economic vitality of urban areas. ABM offers an effective means of representing various traffic elements, whether they are pedestrians or vehicles, making it particularly adept at capturing the dynamics of traffic, as emphasized by Chen [7].

Within the domain of traffic simulation, a multitude of models exist, which can be categorized into the following four levels of detail, as outlined by Nguyen [15]:

Macroscopic simulations employ intricate mathematical models to predict traffic patterns, serving as valuable tools for analyzing extensive systems where a granular model is unnecessary, such as highway traffic simulations. These simulations are characterized by their speed and efficiency, as they demand minimal computational resources owing to their simplified level of detail.

Microscopic microscopic simulations concentrate on replicating individual entities, such as passengers, vehicles, and traffic signals, with a high level of precision. These simulations are commonly employed for investigating urban traffic scenarios, facilitating the analysis of both macroscopic and microscopic aspects, such as multi-modal traffic patterns and the performance of traffic light algorithms. However, it’s worth noting that due to their intricate level of detail, microscopic simulations typically demand more processing time compared to macroscopic simulations.

Mesoscopic hybrid simulations encompass elements from both macroscopic and microscopic modeling approaches. While they offer a higher level of detail when modeling traffic entities compared to macroscopic methods, the interactions and behaviors of individuals tend to be portrayed with less intricate detail.

Nanoscopic these simulations surpass the level of detail found in microscopic approaches, particularly in domains like autonomous driving, where in-depth exploration of internal vehicle functions such as gear shifting and vehicle vision is crucial.

Agent-based models can be likened to tiny-scale simulations that can serve research objectives with varying levels of intricacy, encompassing mesoscopic and macroscopic perspectives. The extent to which different elements of the transportation system are considered depends on this level of detail, a pattern mirrored in the data necessary for modeling. Incorporating real-world data is expected to heighten the authenticity and precision of simulations, necessitating researchers to align the simulation’s purpose with an appropriate model complexity to avoid unnecessary intricacy and an increased data demand [15].

3 Methodology

In order to obtain a comprehensive overview of the literature in ABM practices in the domains, micro-mobility and urban traffic, a well defined search strategy was followed. The methodology is described in details below.

Recent sixteen case studies published between 2014 and 2022, were examined and they were collected from academic research databases such as ScienceDirect, IEEE Xplore, and Scopus. In the selection of case studies, we followed a rationale to ensure the comprehensiveness and relevance of the content included:

  • Scope: The primary aim was to highlight and discuss the key concepts, methods, and trends in ABM for transportation systems. The selected publications were chosen primarily based on the purpose of the each case study. The phrases “agent-based models for micro-mobility” and “agent-based models for traffic simulation” were used to search papers from the repositories. The results retrieved from the search were filtered using the type of traffic, simulation tool, mode of transport and sources of data sets used by the studies and the final sixteen papers were selected, and

  • Contribution: The selection of papers for this work was primarily based on their contributions to the conceptual framework and advancements in the field. Papers were chosen based on the quality of their research, the novelty of their approaches, and their relevance to the overall narrative.

Moreover, we meticulously scrutinized author profiles in both Google Scholar and SCOPUS, while also evaluating the journal’s standing in prominent databases including SCOPUS, Scimago, and Web of Science; this approach was undertaken to ensure the reliability of the content.

Considering the purpose of the study, the selected papers were divided into three categories as shown in Table 1.

Table 1 Categorization of purpose of the study
Full size table

After finalizing the case study, they were analyzed based on six criteria as mentioned below.

  1. 1.

    Location associated with the study,

  2. 2.

    Purpose of the study,

  3. 3.

    Traffic simulation tool used,

  4. 4.

    Transportation modes used as the agents,

  5. 5.

    Data set used and how it was collected, and

  6. 6.

    Model validation and evaluation of results.

3.1 Location associated with the study

The geographic location is an important factor to be considered when analysing the distribution of research work on a certain domain over the world. Since the ABM for micro-mobility is an emerging research area, it is important to identify the places and countries which are interested in implementing micro-mobility transport modes and how ABM is used to enhance the systems. The geographic information related to a certain research topic will help to get an understanding about the feasible locations and specific characteristics of them for the future researchers.

3.2 Purposes of the study

Most of the time purpose of a research is to apply the findings of the research in a real world situation and enhance the existing systems. Here the concentration was to identify in which parts of the transportation domain, the developed models using agent-based simulation were tested. According to the area of application, agents, agent rules, simulation tool can be changed. Therefore purpose of a research study is an important factor which can be analyzed further.

3.3 Traffic simulation tools

There are numerous agent-based simulation tools available for traffic simulation. MATSim, POLARIS, SimMobility are some examples for them. Each simulation tool is comprised with different architectures and selecting the simulation tool depends on the area of application [15].

According to the literature MATSim has the largest user community when compared to other applications. MATSim is an agent-based software framework implemented in Java and licenced under GPLv2 and the framework has a general focus and is designed for the simulation of large-scale transportation scenarios. The framework consists of five components as Initial Demand, Execution, Scoring, Replanning and Analysis [15].

SimMobility is a simulation platform written in C++ and published under an own open-source licence. This simulator is capable of simulation in different time scales (short-term, mid-term, long-term). GAMA (GIS Agent-based Modeling Architecture) is another platform used in traffic simulation and it integrates powerful tools coming from Geographical Information Systems (GIS). There are more simulation platforms which are used considering the problem, application area and the time scale.

3.4 Transport modes used as agents

Two types of agents can be identified based on three types of basic components in ABM. Strategic agents are the rules that govern various interactions in the environment. Agents can be given activity schedules, activity types, activity locations, modes of travel between activities (for travelers), and so on, based on these various rules, and agents: These can be people or vehicles (occupied by individuals). Our main goal in this section is to identify the various types of vehicle agents use for case studies.

3.5 Data sets

The input data used in an agent-based simulation framework is critical to the outcomes. Different errors may occur as a result of input data; it is necessary to understand how these errors affect the simulation model. Data collection is a major process that can introduce errors into the model. While it may be desirable to model all aspects of people’s behavior and the transportation system, the data needed to do so is not easily accessible and may be difficult to obtain. This leaves the option of being selective and selecting the required data based on availability. Data, on the other hand, is a statistical representation of real life. Identifying the critical data to use that would best suit the model is essential for minimizing errors in the model’s output. The available data may be in various forms that require cleaning or transformation before being used in the agent-based model [11].

3.6 Model validation and evaluation of results

Model validation is the process of determining whether the constructed model accurately represent the behavior of the system. Models can be validated by comparing output with experimental data sets that align with the simulated scenario. There are qualitative as well as quantitative techniques in model validation.

In order to apply different techniques to make comparisons, it is important to define measures of performances. Under qualitative validation techniques comparison using graphical displays and exploration of model behavior can be used for observable systems. Quantitative validation techniques use statistical tests and procedures.

Table 2 Summary of case studies
Full size table

4 Analysis

The details about the analysis are discussed in this section.

4.1 Location associated with the study

The summary of case studies is illustrated in the Table 2. The research included various case studies conducted in different regions, with a focus on the North America (specifically San Francisco [19, 28], Boston [20], Sioux Falls [29], and Austin [17]), Europe (including Sulzburg [18, 23, 26], Munich [25], Stuttgart [22], Amsterdam [30], Mielec [21] and Corsica [32]), and two Asian countries, Vietnam (Hanoi [16]) and Singapore [24]. However, the study noted a lack of representation from South America, Africa, Antarctica, and Australia, despite these regions having larger populations. A graphical view of geographical locations of selected research papers for the review generated using PGF/TikZFootnote 1, worlflagsFootnote 2 and WorldMap-TikzFootnote 3 is shown in Fig 1.

Fig. 1
figure 1

Geographical locations of the selected research papers for the review

Full size image

4.2 Purposes of the study

Among the experiments done based on mixed traffic simulation, the case studies done in Vietnam [16] and San Francisco [19] mainly focus on reproducing the traffic patterns of the city to identify the problems and methods to address them. Similarly, Ben-Dor et al. [29] addresses a specific problem by observing the impact on the Dedicated Bus Lanes (DBLs) on urban road traffic. Some of the mixed traffic simulation experiments have been used for resource management planning (refer to the Table 2). For example the case study done in Singapore [24] aims to improve mobility by understanding the interaction between transport and land use in the country.

Some of the objectives identified while analysing previous work are checking the feasibility to establish a micro-mobility transport method, developing existing systems and developing the road infrastructure. The Sulzburge case study [26] mainly focuses on developing an Agent-based Bicycle Model to simulate and capture the spatio-temporal distribution of cyclists across the Sulzburg Road network, and the Berlin scenario [23] try to identify the cyclists’ decisions through an ABM simulation and take them into account to improve the infrastructure of the road network. A case study that was conducted in the city of Amsterdam [30] also had a goal that was more like a combination of the objectives of the first two cases. Another important objective of micro-mobility ABM experiments is to identify where the micro-mobility ride-sharing systems are crowdsourced and where those systems should be established [17]. Moreover, micro-mobility ABM experiments have been used for forecasting the impacts of emerging micro-mobility modes [25], and for understanding of how the shared mobility systems should be re-balanced based on the demand predictions for realistic scenarios [20].

The literature states that ABM simulations have also been used to put automatic traffic signal generators in place in the Greater Berlin Area [18] and Germany has used an ABM approach to obtain useful decisions for managing their long-haul freight traffic [27].

4.3 Traffic simulation tools

The considered case studies have used different ABM simulation tools for their experiments (refer to the Table 2). Most of them have used existing ABM simulation tools such as GAMA [16, 30], NetLogo [26], MATSim [18, 21,22,23, 27,28,29, 32], SimMobility [24], GEMSim [25], and OD Matrix [17]. In Zhao et al. [19], they have used their own tools such as SF ABM [19]. The distribution of the simulation tools used is visualized in Fig 2.

ABM in transportation systems may involve a wide array of tools and softwares, and there are numerous alternatives available; some of them are VISSIM, Anylogic, AequilibraE and Emme. However, our aim is to present a balanced overview of the various tools and not focus exclusively on any specific ones. Thus, the selection of case studies for this section was primarily based on their contributions to the conceptual framework and advancements in the field, rather than their choice of specific software tool, as mentioned in section 3. Therefore, omission of other alternative tools from this work is not indicative of their lack of significance.

Fig. 2
figure 2

Distribution of simulation tool used among the selected studies

Full size image

MATSim can be regarded as the most widely used ABM simulation tool among the selected studies. In MATSim, each agent has plans that include a sequence of activities such as home-work-shop-home, together with their locations and end times. Moralıoğlu et al. [23] has used MATSim for bicycle traffic simulation. The iterative MATSim loop is made out of the following components: In network loading (also known as mobility simulation), all chosen plans are executed in a synthetic reality at the same time. Following that, all executed plans are scored. Finally, all synthetic people are permitted to re-plan. This loop is repeated till the system is reasonably “relaxed” as determined, for example, based on the progress of agents’ plan scores. Ziemke and Braun [18] used MATSim to model inner-city traffic more realistically by identifying the same features; this study extends MATSim input generation. As can be seen, MATSim has created a variety of extensions that enable the simulation or analysis of various transport services such as automated vehicles, demand responsive transport, and car sharing. Not only that, MATSim is freely available open-source software with the following features. The ability for simulation of both human and engine-powered vehicles (bicycles, scooters, and cars); simulation of docked and undocked services (for station services, each station is distinguished by the vehicles and parking spaces that are currently available); simulation of multiple operating systems providing services in the simulated area; and definition of each operating system’s service area.

In addition to MATSim, the GAMA framework is the most commonly used traffic simulation tool. MATSim was used in eight of the sixteen case studies, and GAMA was used in two. The GAMA platform is built on the Eclipse software package and can be fully customized for any application. GAMA can also handle shape file input, generate spatially referenced output files, and provide a variety of 2D and 3D visualization options [30]. In Dang-Huu et al. [16], GAMA platform was used because the model is designed to cope with mixed traffic without depending on the expectation of lane regards. They needed to implement a 2-dimensional model to simulate mixed traffic, which included at least motorcycles, cars, and buses.

In Wallentin and Loidl [26], NetLogo was used as the ABM framework. They created two scenarios to test the hypothesis that the surrounding area of Sulzburg is important for the spatial distribution of cyclists. The first situation - Sulzburg city, simulated one day of cycle traffic in Sulzburg’s administratively defined city borders (65 \(\hbox {km}^2\)). Commuters to and from places outside of the city limits were not taken into account. As a result, the city was regarded as a closed system. The second situation - Sulzburg city region, included the greater Sulzburg region, 330 \(\hbox {km}^2\). In the both situations, the developed framework and all of its parameters, along with the total number of simulated cyclists, street network characteristics, and routing preferences, were the same. Therefore, it is clear that there are various purposes that ABM simulation tools can be used for.

4.4 Transport modes used as agents

The various entities moving on the road are represented by vehicle agents. They can be of numerous types, such as a bus, a car, or a motorcycle etc. The majority of the models merge vehicles and drivers into a single agent. As we can see from the selected case studies, 8 of the 16 have attempted to simulate mixed traffic in their cities [16, 18, 19, 22, 25, 29, 32]. As a result, they have used various modes of transportation as agents such as cars, motorbikes, buses, public transport modes and trucks. Since the goal of Bischoff and Maciejewski [21] was to simulate a battery-powered electric taxicab fleet in a small city scenario, and the goal of Lu et al. [27] was to generate long-haul freight traffic for Germany, they used electric taxicabs and freight vehicles such as trucks as their agents respectively. The remaining six studies [17, 20, 23, 26, 28, 30] used micro-mobility agents, primarily bicycles or e-scooters as their main goal was to simulate an agent-based model for micro-mobility ride sharing.

4.5 Data sets

In most of the experiments, they used data from available research and resources from the selected case studies (refer to the Table 2). For example, in the Vietnam model [16], available traffic video footage was used as the primary data source. One of the benefits of using data from publicly available sources is that it is inexpensive. Using this type of data is also critical when building and evaluating models.

Because it provides geospatial information, the majority of the experiments used OpenStreetMap road network data as their data source [18,19,20, 23, 25, 27]. The geospatial data module includes (a)–the road graph, (b)–building data, and (c)–the geospatial indexing system, provides geospatial information [36]. The road graph is the only one that is strictly required for the simulation. This Geospatial data can be obtained by using the OpenStreetMap service. The remaining case studies have used data from previously gathered statistical data and findings from the literature [24, 26, 28, 30, 32] used publicly available data and the data which were collected from a Singapore travel survey. It is clear that almost every case study makes use of previously available data for their models.

4.6 Model validation and evaluation of results

Model validation is a critical component of traffic simulation using ABM tools which indicates how well the model can reproduce traffic in reality. According to the literature, various validation techniques have been used for this.

Some case studies [16, 24] compared the model’s output to the calculated number of vehicles per second from traffic videos. Three case studies [26, 30, 32] used the method of verification by comparing the resulting bicycle count to real-world traffic counts. Another method found in the literature was to compare the results to those obtained from previously published models. Apart from that, mode distribution has been used as a verification method for the model of Sanchez et al. [20].

A sensitivity analysis with the simulation model shows how the model’s results varies as the inputs change, and we can determine the influence of variables on the results. Furthermore, predicting future travel scenarios is an important aspect of agent-based models. However, no previous reports have validated the forecasting ability of agent-based models. Conducting Before and After studies is one possibility. For example, these studies can demonstrate whether an agent-based model can predict traffic demand based on infrastructure changes that occurred within a specific time frame.

5 Discussion

Alongside with the subsections of section 3 and section 4, now we present a comprehensive discussion on our study.

5.1 On location associated with the study

The summary of case studies provided in Table 2 offers a comprehensive overview of research conducted in the field of transportation modeling. These studies have explored a variety of transportation-related topics, shedding light on innovative approaches and potential solutions for transportation challenges. However, it is crucial to recognize both the strengths and limitations of the geographical representation within these case studies.

A significant focus of this research has been on North America, particularly in cities like San Francisco, Boston, Sioux Falls, and Austin. These case studies have examined a wide range of transportation issues and have contributed valuable insights into micro-mobility, mixed traffic models, and other applications. They serve as essential references for policymakers and urban planners striving to enhance transportation systems in North American cities. Europe has also been a prominent region in this research, with multiple case studies conducted in cities such as Sulzburg, Berlin, Munich, Stuttgart, Amsterdam, Mielec, and Corsica. These studies have tackled various transportation challenges, providing valuable information on subjects like micro-mobility and urban transportation planning. The European context is rich in diversity, and the case studies conducted there serve as benchmarks for sustainable and efficient transportation solutions. The inclusion of case studies from two Asian countries, Vietnam and Singapore, further enhances the global perspective of this research. These studies offer unique insights into transportation issues specific to their respective regions, and they highlight the importance of considering local factors when developing transportation models and solutions. However, the notable gap in this research lies in the underrepresentation of certain regions, specifically South America, Africa, Antarctica, and Australia. These areas face distinct transportation challenges influenced by factors like diverse terrains, climate conditions, and varying levels of urbanization. The absence of case studies from these regions underscores the need for a more inclusive global perspective in transportation modeling research.

In conclusion, the case studies summarized in Table 2 are essential contributions to the field of transportation modeling. They offer valuable insights and solutions for transportation challenges in various regions, particularly in North America, Europe, and select Asian countries. Nevertheless, to provide a more comprehensive and globally applicable understanding of transportation dynamics, it is imperative that future research endeavors consider broader geographical representation, encompassing the unique challenges and opportunities found in South America, Africa, Antarctica, and Australia. Expanding the scope of transportation modeling research to these regions will contribute to a more holistic and well-informed approach to transportation solutions on a global scale.

5.2 On purpose of the study

The experiments conducted within the realm of mixed traffic simulation have provided valuable insights and solutions for a variety of transportation-related challenges. Several case studies stand out for their specific objectives and applications, as outlined in the discussion below.

Reproducing traffic patterns and identifying issues The experiments in Vietnam [16] and San Francisco [19] have primarily focused on reproducing the traffic patterns in these cities. By doing so, they aim to identify existing traffic problems and develop potential solutions. These studies provide a foundation for addressing the unique challenges of each city’s transportation system.

Impact of DBLs Ben-Dor et al.’s study [29] addresses a specific problem by examining the impact of DBLs on urban road traffic. It investigates how the implementation of DBLs affects traffic flow and public transit use, providing insights into resource management and public transportation planning.

Resource management planning Some mixed traffic simulation experiments, as indicated in Table 2, have been employed for resource management planning. For example, the case study in Singapore [24] focuses on improving mobility by understanding the complex interaction between transportation and land use within the country. This approach assists in optimizing resource allocation and urban planning.

Developing micro-mobility transport methods and road infrastructure Several case studies aim to develop micro-mobility transport methods and enhance existing road infrastructure. The case study in Sulzburg [26] concentrates on creating an Agent-based Bicycle Model to simulate the distribution of cyclists across the Sulzburg Road network. Similarly, the Berlin scenario [23] uses ABM to understand cyclists’ decisions and improve road network infrastructure. In Amsterdam [30], the objective combines elements of both, focusing on enhancing existing systems and developing new micro-mobility transport methods.

Establishing micro-mobility ride-sharing systems Some experiments aim to determine the feasibility of establishing micro-mobility ride-sharing systems. The case study in San Francisco [17] explores where these systems should be established, taking into account various factors to optimize their deployment.

Forecasting the impacts of emerging micro-mobility modes Micro-mobility ABM experiments have been employed to forecast the impacts of emerging micro-mobility modes, as demonstrated by the study in Munich [25]. This type of research is vital for staying ahead of the curve and adapting to new transportation trends.

Re-Balancing shared mobility systems Experiments, such as the one conducted by Coretti and Sanchez [20], use ABM to understand how shared mobility systems should be re-balanced based on demand predictions for realistic scenarios. This optimization of shared mobility services is crucial for providing efficient and convenient transportation options.

Implementing automatic traffic signal generators ABM simulations have also been utilized to implement automatic traffic signal generators, as seen in the Greater Berlin Area [18]. This technology aids in optimizing traffic signal management and improving traffic flow.

Long-haul freight traffic management Germany’s approach to employing ABM for managing long-haul freight traffic [27] showcases how this modeling technique can assist in making informed decisions regarding resource allocation and infrastructure improvements, especially for freight transportation.

In conclusion, the various objectives and applications of mixed traffic simulation experiments underscore the versatility and relevance of ABM in addressing a wide range of transportation challenges. From replicating traffic patterns and assessing the impact of infrastructure changes to resource management, forecasting, and decision-making support, these studies offer valuable contributions to transportation planning and management.

5.3 On traffic simulation tools

The selection of ABM simulation tools for the considered case studies, as presented in Table 2, illustrates a diversity of software options employed to address transportation challenges. Notably, various existing ABM simulation tools have been utilized, each suited to the specific requirements of the studies. These tools include GAMA, NetLogo, MATSim, SimMobility, GEMSim, and OD Matrix.

GAMA, employed in the case studies conducted in Vietnam [16] and Amsterdam [30], provides a flexible platform for developing and implementing agent-based models in diverse contexts. NetLogo, used in the Salzburg case study [26], is well-known for its user-friendly interface and suitability for modeling complex systems. MATSim is a widely adopted choice for simulating multi-agent transportation scenarios and is featured in numerous case studies, including Berlin [23], Munich [18], Corsica [32], Mielec [21], San Francisco [19], Austin [17], and Stuttgart [22]. SimMobility was utilized in the case study conducted in Singapore [24], offering a platform for comprehensive urban mobility modeling. GEMSim, applied in the study of emerging micro-mobility modes [25], caters to the specific needs of modeling these new transportation trends. The OD Matrix tool was utilized in the case study investigating micro-mobility in San Francisco [28].

It is noteworthy that Zhao et al. [19] developed their own tool, SF ABM, for their specific case study in San Francisco. This customization allowed them to address the unique requirements of their experiment effectively.

It is essential to acknowledge that the field of ABM in transportation systems offers a multitude of tools and software options, such as VISSIM, Anylogic, AequilibraE, Emme, and many others, each with its strengths and capabilities. The omission of specific tools from this discussion should not be interpreted as a diminishment of their importance; rather, the focus here is on highlighting the diverse range of tools that have been applied in the considered case studies.

The choice of simulation tools in each case study was guided by their appropriateness for the research objectives and the modeling needs of the specific transportation challenges under investigation. As such, the diverse selection of tools showcases the adaptability of ABM simulation in addressing transportation issues across different contexts. The key criterion for the inclusion of case studies in this discussion was their contributions to the conceptual framework and advancements in the field, rather than their choice of software tool. This approach ensures a balanced overview of the various tools and their applicability, emphasizing the significance of ABM in transportation research.

5.4 On transport modes used as agents

In the realm of ABM for transportation systems, the representation of various entities on the road, often referred to as agents, plays a pivotal role in capturing the complexity of mixed traffic scenarios. These entities encompass a diverse array of vehicle types, including cars, motorcycles, buses, public transport modes, trucks, electric taxicabs, freight vehicles, bicycles, and e-scooters. The choice of agents is closely aligned with the research objectives and the specific transportation challenges addressed in each case study.

It is noteworthy that the majority of the selected case studies adopt a model where both vehicles and drivers are merged into a single agent. This amalgamation streamlines the modeling process and facilitates a holistic representation of the transportation system. As a result, eight out of the sixteen chosen case studies have undertaken simulations of mixed traffic in their respective cities. These studies, including those in Vietnam [16], San Francisco [19], Berlin [18], Corsica [32], Munich [25], and other regions, encompass a wide range of transportation modes within their agent-based models. These modes include cars, motorbikes, buses, public transport, and trucks. Such comprehensive modeling allows for a more accurate portrayal of the complex interactions among various transportation entities in urban environments.

Two case studies deviate from the common trend by using specialized agents based on their research objectives. Bischoff and Maciejewski [21] sought to simulate a battery-powered electric taxicab fleet in a small city scenario, leading to the utilization of electric taxicabs as agents. Meanwhile, Lu et al. [27] aimed to generate long-haul freight traffic for Germany, necessitating the use of freight vehicles such as trucks as the primary agents.

The remaining six case studies have adopted micro-mobility agents, primarily bicycles or e-scooters, as their core focus revolves around simulating agent-based models for micro-mobility ride-sharing services. The inclusion of these studies acknowledges the increasing relevance of micro-mobility solutions in contemporary urban transportation systems, underscoring the importance of tailored agent-based models for this specific domain.

In summary, the selection of agents in agent-based transportation models is driven by the research objectives and the specific transportation scenarios being examined. These agents, whether they represent mixed traffic, electric taxicabs, freight vehicles, or micro-mobility modes, are integral to capturing the intricate dynamics of transportation systems and enabling more informed decision-making in urban planning and resource management. The diversity of agent types reflects the evolving landscape of transportation and underscores the adaptability of ABM in addressing a broad spectrum of transportation challenges.

5.5 On data sets

The effective development and evaluation of agent-based transportation models heavily rely on the availability and quality of data sources. In the majority of the experiments outlined in Table 2, researchers utilized data from various existing research studies and publicly available resources. This approach not only reduces costs but also ensures the models are grounded in real-world data, enhancing their accuracy and applicability.

One notable example is the Vietnam model [16], which leveraged available traffic video footage as a primary data source. The accessibility of such data from public sources proves invaluable, offering researchers a cost-effective means to gather essential information. Utilizing real-time traffic video footage provides a dynamic and authentic basis for modeling traffic patterns and behaviors.

Additionally, many experiments incorporated geospatial information obtained from OpenStreetMap road network data. This comprehensive dataset includes essential components such as road graphs, building data, and geospatial indexing systems. Among these, the road graph is particularly crucial for simulations. OpenStreetMap’s geospatial data serves as a valuable resource for researchers, offering detailed and up-to-date information about road networks and urban infrastructure. The availability of such data ensures the models accurately reflect the real-world layout of cities and their transportation systems.

Moreover, some case studies drew from previously gathered statistical data and literature findings. For instance, the experiments in Salzburg [26], Corsica [32], and Amsterdam [30] utilized publicly available data and information from existing studies. Similarly, the Singapore case study [24] incorporated data from a Singapore travel survey. This reliance on existing data highlights the importance of collaborative efforts within the research community and the utilization of established knowledge to inform new studies.

In summary, the utilization of publicly available data, whether from traffic video footage, OpenStreetMap, or existing research studies, underscores the pragmatic and resourceful approach of researchers in the field of agent-based transportation modeling. Leveraging these data sources enhances the validity of the models, ensuring they are rooted in real-world observations and are capable of providing meaningful insights into urban transportation dynamics. The consistent use of previously available data across these case studies showcases the critical role such information plays in advancing the field and addressing complex transportation challenges.

5.6 On model validation and evaluation of results

Model validation is a crucial step in agent-based traffic simulation, as it gauges the model’s ability to accurately replicate real-world traffic dynamics. The literature reveals a variety of validation techniques used in agent-based transportation modeling, each tailored to the specific objectives and features of the case studies.

Comparison to real traffic data Some case studies, such as those in Vietnam [16] and Singapore [24], adopted a validation approach that involved comparing the model’s output to the actual number of vehicles per second observed in traffic videos. This direct comparison to real-world data serves as a robust method for verifying the model’s ability to reproduce traffic patterns faithfully.

Comparison to real bicycle traffic counts Other experiments, like the ones in Corsica [32], Salzburg [26], and Amsterdam [30], verified their models by comparing the resulting bicycle counts from the simulations to real-world bicycle traffic counts. This validation technique specifically addresses the accuracy of modeling bicycle traffic, which is of particular importance in micro-mobility-focused studies.

Comparison to previously published models The literature also includes instances where model validation involved comparing the results of an agent-based model to those obtained from previously published models. This approach provides a basis for assessing the model’s performance relative to established benchmarks and existing research.

Mode distribution verification In the case of Coretti and Sanchez [20], they employed mode distribution as a verification method. This approach involves comparing the distribution of transportation modes (e.g., walking, cycling, driving) in the model’s results to observed real-world mode distribution. This verification method is particularly relevant for studies that focus on understanding and optimizing mode choices in transportation systems.

Sensitivity analysis Sensitivity analysis is an additional tool used in model validation, which assesses how the model’s results vary as input parameters change. This analysis allows researchers to understand the influence of various variables on the model’s outcomes, enhancing its robustness and reliability.

Forecasting future travel scenarios The ability of agent-based models to predict future travel scenarios is crucial for urban planning and decision-making. Although not commonly reported in the literature, one potential method for validation is conducting “Before and After” studies. These studies involve comparing the model’s predictions to real-world traffic demand data before and after specific infrastructure changes. This approach helps ascertain the model’s forecasting accuracy and its capacity to anticipate the impacts of alterations in the transportation system.

In conclusion, model validation in agent-based transportation modeling is a multifaceted process that can take various forms, depending on the research objectives and the specific aspects of traffic being studied. These validation techniques ensure that agent-based models accurately replicate real-world traffic behaviors and dynamics, making them reliable tools for informing transportation planning and management decisions. The ongoing refinement and diversity of validation methods contribute to the continued advancement of agent-based transportation modeling.

6 How to model the “last mile”

According to the areas of application of the evaluated case studies, we can summarize the findings under three topics as (a)–Micro mobility models (b)–mixed traffic models (c)–other applications. Several significant findings and potential future improvements that could be applied to the new models were identified.

As described in the papers micro-mobility is an efficient method of transport which can be used for the first and last mile trips. Reducing environmental pollution while introducing sustainable transportation, low operational cost are the main concerns of the authorities which are interested in implementing micro-mobility for inner city transportation.

In the study of Salzburg [26], which analyzed the performance of autonomous bicycles in comparison to station-based and dockless bicycles, valuable insights were gained for multiple stakeholders regarding the design, implementation, and operation of autonomous bike-sharing systems. The study measured the impact of various configuration variables on each system’s performance and validated the impact of nominal state values by running batch simulation models with a range of values for each variable. The key finding of this study was the relationship between wait times and fleet size, which is a fundamental aspect of shared autonomous systems. Furthermore, the study highlighted the effects of metrics related to bicycles, such as autonomous speed and battery autonomy, which can inform vehicle design.

The presented case study of Balac and Horl [28] looked into the capacity of the Bay-Wheels bike sharing system to decrease transportation travel times through the first and last mile integration. Provided supply and infrastructure resource constraints, it proposed that the service minimize transportation travel time by up to 1.4% of trips, but noted that with some re-balancing, this might be increased.

The comparative study of Sanchez et al. [36] done on autonomous bicycles vs station-based bicycles and dockless bicycles provides insights for many stakeholders on how to design, implement and operate autonomous bike-sharing systems. It directly measured the effect of the various configuration variables on the performance of each system and validated the impact of nominal state values by having to run batch simulation models with a range of values for each variable. These findings provide a response to one of the most fundamental questions in shared autonomous systems: the relationship between wait times and fleet size. Other important things we’ve learned from this analysis include the effects of metrics related to bicycles, such as autonomous speed and battery autonomy, because these outcomes can mentor vehicle design.

The model of Veldhuis [30], could be a useful addition to the tools and features of spatial planners, traffic flow engineers, and GIS experts for evaluating and recreating Amsterdam’s bicycle network, but only under two very specific restrictions:

  1. 1.

    More research is needed to back up the classification of influential factors in cyclist behavior, and

  2. 2.

    More data on the number, motivation, origin, and destination of cyclists must be collected.

If such two requirements are met, this framework can able to collect results when it comes to the effects of cyclist behavior and infrastructure changes on network loadings.

As proposed by the Vietnam model [16], their model demonstrated its ability to reproduce an observed traffic jam situation. They also discovered that in order to use the model in a decision support tool, it must be improved before it can be applied to a real road network created from actual GIS data. In order to gain a better understanding of and forecast complex issues, they proposed improving individual behaviors and comparing them to existing simulation methods.

The case study of San Francisco [19] is about a model that was developed is a macroscopic agent-based model for traffic simulation, and the simulation was based entirely on inputs processed from publicly available datasets, including a detailed road network graph from openStreetMap. The model of Ziemke and Braun [18] demonstrated good computational performance as well as the ability to generate reasonable traffic simulation results. Based on OpenStreetMap data, the study of Ziemke and Braun [18] introduced the implementation of an automatic generation of traffic signals and lane details for the agent-based transport modeling MATSim.They discussed how to enhance the implementation of traffic signals for pedestrians, driveway length, and road-specific flow capacity values in the future. Moreover, the paper shows that OSM does not offer any details about which lanes are signalized separately and proposes an extension of OSM traffic signal tags that can capture all of the information.

The model of Lu et al. [27] can be identified as the first version of the long-distance freight trips model for Germany. As a result, several factors are illustrated for future improvements and development. The existing system, for example, simplifies the departure time of freight trips, the freight vehicle structure, and the loading condition of freight vehicles for various kinds of goods.

Moreover, the model of Ben-Dor et al. [29] shows that the impact of DBLs is highly dependent on the city’s level of traffic jams, which is also impacted by the size of the population. A 20% rise in public transit use is observed for a high but accurate level of traffic jams.

7 Conclusions and future works

ABMs have been applied for urban traffic management, mainly in three types of applications; Micro-mobility, Mixed Traffic and Other.

The public interest in micro-mobility has been rapidly increasing while transforming the transportation system in many cities. The results of ABMs for micro-mobility ride sharing systems reveal that these models are very much suitable to apply to a crowded city center. The simulation studies have been used to identify the possible impacts of introducing novel transportation methods such as autonomous driving technology.

MATSim can be regarded as the mostly used simulation tool in traffic simulation applications. MATSim has been used for a variety of transportation studies, including analyzing the impact of road pricing policies, evaluating public transportation systems, and predicting the effects of new infrastructure projects. Its open-source nature and active development community make it a flexible and powerful tool for transportation modeling and analysis. Shared micro-mobility systems can be enriched by incorporating discrete mode choice models in MATSim. Agent-base modeling can be used to simulate long-haul freight traffic which has less attention compared to commuter traffic. Apart from the micro-mobility transport modes, electric vehicles also play a major role in contributing towards the development of sustainable transportation. Simulation studies have proved that electric taxi cabs provide taxi operations with a high service level.

Almost all the studies have used publicly available data for the simulation tasks which are poor in quality. This can be identified as a limitation in our work. Therefore, as a future work we suggest that the responsible parties must make necessary arrangements to collect quality data for future research activities in order to obtain more reliable results. Use of agent-based models to study urban traffic and micro-mobility traffic simulations help authorities to take necessary decisions and accomplish sustainable transportation goals. Moreover, the ABMs can be used to observe the alterations in the urban mobility system with the increase of the population due to development projects. Improving the individual behavior of agents is recommended to observe and forecast more complex issues such as crossroad deadlocks.