1 Introduction

The introduction of AVs will become a challenge in the coming years. Given the rapid technological advances in autonomous driving, the question is no longer how, but when AVs will be introduced for full commercial use [9]. It is claimed that 75% of vehicles will be autonomous by 2040, while conservative reports suggest that this will be the case by 2060 [76]. AVs are likely to have significant impacts on trip patterns, traffic, and transportation operations [76]. Indeed, if AVs dominate, congestion is expected to worsen [45, 96], urban sprawl will be encouraged [31, 118], and the culture of “automobility” will be maintained [92].

On the contrary, the option of using AVs in the context of shared services (Shared Autonomous Vehicles – SAVs) is a case of promoting sustainable mobility, as SAVs are envisaged to help reduce traffic congestion [32, 72], limit the need for parking, and free more public space for other activities [55]. Furthermore, since most traffic accidents are attributed to human behavior [56, 97], SAVs will have a positive impact on road safety [6, 103]. What is more, SAVs will compete with conventional, taxi-type and ridesharing services [10, 32], thus redressing the way transport providers operate in an urban environment. It is therefore evident that the development of suitable methodological tools to study and assess that impact is essential for the efficient planning and design of future transportation systems. Indeed, emerging technologies and new modes, (such as SAVs, Mobility-as-a-Service (MaaS) and so on), set new challenges for transportation planning, as they create the need to investigate travel patterns at a more refined, microscopic scale [62].

Agent-based models (ABMs) can replicate transportation systems at a fine-granular level and as such, they are widely used for modeling SAV systems. ABMs are dynamic simulation processes that converge after a series of iterations; they consist of agents, the environment, and rules [62, 133]. Agents are the actors in a system (either travelers or vehicles), the environment is the place where agents act and/or interact (the road network, land uses, and transportation services), and rules describe the behavior and interaction of agents. Since ABMs represent travel behavior at a highly disaggregated level, it is possible to analyze the effects of transport policies on travel behavior and traffic in detail [66]. Given this inherent advantage, the research community has adopted such models to study the role, behavior, and impact of SAV systems.

Considering the extensive use of ABMs in SAV-related studies, this paper systematically reviews and organizes the literature on agent-based modeling of SAVs. The paper goes beyond the previously published state-of-the-art reviews on AVs [60, 73, 93] and critically analyzes and assesses the application of ABMs and their attributes, in the case of SAVs. Also, this paper provides a systematic analysis of the impacts of SAVs, derived through the application of ABMs. This allows to depict the suitability of ABMs in modeling SAV services and operations and to shed light on the impacts of SAVs on the urban environment.

The remainder of the paper is organized as follows: The second section addresses the methodological framework used to conduct the review. The third section explores various aspects of agent-based modeling in SAV related studies, and the fourth section offers a discussion of the findings and recommendations for future research. Finally, the conclusions and contribution of the paper are presented.

2 Methodology

A systematic literature review refers to the process of systematically finding and compiling all available information on an effect or topic area [26],it can be briefly described as a research method for identifying and critically appraising relevant research [18] and collecting and analyzing data from that research [75]. The research themes considered in the context of the review, are referred to as primary studies, while the review itself is a secondary study. This method is an acceptable way to synthesize research findings and show evidence at a meta-level [115], and it is therefore sufficient for the purpose of identifying critical concepts and questions on the topic reviewed. A systematic literature review exploits qualitative approaches developed to assess the quality and strength of findings from different types of studies and to compare these findings [37]. In particular, the processes of systematic literature review may differ depending on the scope and objectives of individual studies.

In this paper, a systematic review is used to gain a thorough understanding of ABMs and their use in the context of SAVs. various trusted databases such as Google Scholar, Web of Science, ScienceDirect, SPRINGER LINK, TRID, IEEE Xplore, Taylor and Francis, SAGE Publishing, etc. are exploited for finding relevant peer-reviewed articles (journals or conference proceedings) since 2013, using keywords that appear in the title, abstract, and body of the articles. Three categories of search terms were used, either separately or in various combinations: (1) agent, agent-based modeling, and agent-based simulation; (2) autonomous vehicles (AVs), autonomous taxi, autonomous mobility on demand (AMoD), shared autonomous vehicles (SAVs), and shared autonomous electric vehicles (SAEVs); (3) impact(s), implication, and effect. These terms were adapted to the specific structure and requirements of each database. Duplicate and irrelevant papers were ignored, and references within identified papers were carefully reviewed. Although some of the articles found were not peer-reviewed (gray literature), such as some pioneering scientific reports, they are still important for broad understanding interests in this field. Τo ensure the high quality of the review, the articles meet the following criteria: (1) they should be written in English; (2) they should use agent-based modeling (ABM) or agent-based simulation as analysis tools; and (3) they should include research on shared autonomous vehicles (SAVs). The final review pool consists of 98 scientific papers.

3 Agent based modeling in SAVs research

For the purposes of this review, three main aspects of ABM applications in SAVs are identified: conceptual, methodological, and impacts. The conceptual aspect refers to specific research questions on SAVs that ABM applications attempt to answer, and especially those related to the design, planning, and evaluation of SAVs services. The methodological aspect includes the framework, attributes, and use of ABMs in SAV related studies; the review revealed two broad categories for ABM applications in the context of SAVs: (a) studies focusing on the planning/design of SAVs systems and services, and (b) work analyzing the potential impact of SAVs on transport supply, mobility, the environment, and so on. Figure 1 illustrates the categorization of impacts related to ABM application in SAVs:

Fig. 1
figure 1

Investigated SAV impacts by ABMs

3.1 ABMs in the planning and design of SAV systems

Given the potential of SAVs in future transportation systems, the first category covers all aspects of SAV service development using ABMs, including planning, their application as first/ last mile solutions, taxi and ridesharing services, pricing, charging infrastructures and user behavior analysis.

3.1.1 Designing SAVs services

Several studies have addressed the design and planning of SAV services using ABMs, with focus on fleet sizing, costing, vehicle location and parking decisions. In this context, Fagnant and Kockelman [32] estimated fleet size requirements and assessed vehicle location strategies for a SAV service in Austin, TX. Their subsequent work delved into the effects of low SAV market penetration within the same city, simulating a fleet’s operational dynamics Fagnant et al. [33]. They later expanded their research by considering dynamic ridesharing in SAV operations [34]. Building on these studies, Chen et al. [21] explored the management of a fleet of SAVs using a regional ABM, for assessing different vehicle ranges and charging infrastructures.

A study by Bösch et al. [16] extended the ABM platform MATSim (see Horni et al. [52] for further details on MATSim) to estimate the required fleet of SAVs needed for a SAV service in Zurich, Switzerland. Another ABM multiscale platform, SimMobility (see Adnan et al., [1] for further details on SimMobility) was used by Marczuk et al. [90] to investigate interplay of land-use, transportation, and communications, in the context of planning an Autonomous Mobility on Demand (AMoD) service in Singapore. Heilig et al., [47] used a microscopic travel demand model to simulate mode choice behavior and calculate fleet size needs for an AMoD service when private vehicles are not available, using the Stuttgart region as a case study. In the same direction, Dia & Javanshour [30] also proposed an AMoD service for Melbourne, Australia, and used the ABM platform Commuter (see Liyanage and Dia [80] for further details on Commuter) for assessing fleet size and parking requirements. Extending previous studies, Loeb & Kockelman [82] evaluated the fleet performance and costs of a SAV service in Austin, TX, identifying key profitability and customer satisfaction factors. Zhou et al. [140] introduced a collaborative model integrating park-and-ride facilities, public transportation, and SAVs in Nagoya, Japan, leveraging ABM insights. Concurrently, Ben-Dor et al. [12] evaluated the feasibility of ridesharing with SAVs in Tel-Aviv, Israel, using MATSim. Finally, Wang et al., [127] employed the ABM platform AnyLogic (see more in Borshchev [15], to examine the impact on travel and energy consumption by strategically formulating SAV platoons.

3.1.2 First/last mile transportation

Some studies have explored the role of SAVs in serving First/Last mile trips; In this context, Scheltes & De Almeida Coreira [109] utilized an ABM to design an Automated Last-Mile Transport system (ALMT), in Delft, Netherlands for connecting the city’s train station and the university campus. Shen et al., [111] tackled the first/last mile problem to and from Mass Rapid Transit stations in Singapore, by integrating a SAVs service and public transport. Using an ABM framework, they assessed the potential replacement of low-demand bus routes with such a service. In Sejong, Korea, Kim et al., [67, 68] examined the viability of an autonomous minibus service using the MATSim simulation framework. Concurrently, Gurumurthy et al., [42] investigated the potential of SAVs as a collective distribution mechanism in Austin, TX, highlighting the versatility of SAVs in enhancing urban mobility and addressing specific transportation challenges.

3.1.3 Planning Autonomous Taxi (AT) services

Autonomous taxi services, a specific subset of Shared Autonomous Vehicles (SAVs), have attracted considerable attention in academic research. Bischoff & Maciejewski [14] were the first to simulate a citywide replacement of private vehicles by autonomous taxi (AT) fleets in Berlin, Germany, using the MATSim simulation platform. Concurrently, Hörl et al. [50] exploited MATSim for simulating autonomous taxis in an integrated population and network-based transportation environment, which considered dynamic demand. Llorca et al. [81] used the same platform to simulate a fleet of autonomous taxis for partial substitution of transport demand in Munich, Germany. ABMs were also exploited by Martinez & Viegas [91] for evaluating different autonomous taxi services, in Lisbon, Portugal. Merlin [94] and Lu et al., [87] explored the potential of automated taxis in Ann Arbor, MI, in the context of either complementing public transport or the replacement of private vehicles. Other ABM applications planning for autonomous taxi services were presented by Hörl [48], Jäger et al., [57], Kim et al., [67, 68], Alisoltani et al., [5], and Chouaki & Puchinger [22].

3.1.4 Dynamic ride sharing

SAVs can play a transformative role in the ride-sharing ecosystem, fundamentally reshaping how ride-sharing services operate and are consumed. Exploiting ABMs, Zhang et al., [138] simulated the performance and estimate the likely benefits of a SAV system with dynamic ridesharing, and Lokhandwala & Cai [84] investigated the benefits of dynamic ridesharing in AT services in New York City, ΝΥ compared to traditional taxi services. Later, Wang et al. [124] applied AnyLogic to simulate dynamic ridesharing systems with both station-to-station and door-to-door services. Hörl et al., [51] exploited MATSim and its dynamic vehicle routing extension to simulate different operational strategies for controlling an automated mobility-on-demand system with sequential vehicle sharing. In a series of studies, Gurumurthy et al., [41] used MATSim to simulate travel behavior in Austin, TX, in the presence of private and shared AVs, with dynamic ridesharing and advanced road pricing policies. Subsequently, Gurumurthy et al., [40] used POLARIS (see further in Auld et al., [7] to assess dynamic ridesharing choices of AVs with geofencing in the Chicago region, IL, and Gurumurthy & Kockelman [39] studied the effects of pick-up and drop-off points on dynamic ridesharing rates in Bloomington, IL.

3.1.5 Costs and pricing of SAV services

Service costs and pricing of SAV services are critical for assessing their potential introduction. Using Austin, TX as a testbed and ABMs, Chen et al., [21] measured the impact of fare structures on the market potential of SAVs, Liu et al., [79] explored the relationship between the level of SAV fares and their impact on modal split of private trips, and Simoni et al., [113] examined the impact of different congestion pricing and tolling strategies. Bösch et al., [17] scrutinized policy combinations for SAV services in Zug, Switzerland, and identified pricing of public and private motorized transport as a suitable one. The impact of utility-based dynamic pricing for Autonomous Transportation Network Companies, using an ABM for Greater London, UK, under monopoly and competitive conditions, was investigated by Karamanis et al., [65].

Wen et al., [126] explored various pricing and hailing strategies for SAVs in a European city, while Nahmias-Biran et al., [102] combined the demand simulator of SimMobility with a meso-micro supply model to analyze service costs for both individual and shared Autonomous Mobility on Demand (AMoD) journeys in Tel-Aviv, Israel. Bucchiarone et al., [19] examined the introduction of autonomous shuttles in Trento, Italy, evaluating their service costs through an Agent-Based Model (ABM). Mo et al., [98] investigated the competitive dynamics between public transportation and SAVs in Tampines, Singapore, focusing on revenue generation and passenger costs. Ben-Dor et al., [13] used the MATSim platform to study the effects of various congestion and parking pricing strategies, as well as different SAV fleet compositions on SAV services in Jerusalem, Israel. Finally, Stevens et al., [119] analyzed the financial sustainability of AMoD systems in Rotterdam, Netherlands, considering strategies for vehicle relocation, ridesharing, and charging through the AnyLogic ABM platform.

3.1.6 Charging infrastructures

Since SAVs are expected to be electric (EV) or hybrid electric vehicles (HEV), some researchers have investigated the performance of SAVs with respect to the available charging infrastructure. Vosooghi et al. [122] related the performance of SAVs to their charging infrastructure in Rouen Normandy, France and found that performance is significantly improved by the placement of fast chargers and battery replacement infrastructures. Charging infrastructure for SAVs was also addressed by Zhang and Chen [134], who proposed a smart charging framework and argued that EVs with larger batteries respond better to charging opportunities with low electricity costs and have greater potential to reduce total energy-related costs. Ahadi et al. [2] argued that charging station locations for SAV depend mainly on the spatial distribution of installation costs and charging demand, with optimal locations in both central areas where demand is high and suburbs where installation costs are lower. Dean et al. [29] coupled repositioning and charging strategies in Austin, TX, resulting in 39% lower average wait times, 28% more daily trips, and 1.6% fewer empty trips. Furthermore, ridesharing is expected to downsize the fleet and reduce the number of charging stations required to keep the fleet operational. Compared to traditional ride-hailing service, in the case of a city model resembling Austin, TX, ridesharing decreases the fleet size and the number of charging stations from 57,279 and 1562 to 25,368 and 1058, respectively [35]. Last, Wang et al. [127] demonstrated that formulating platoons could reduce the existing system-wide energy consumption up to 9.6%. However, Loeb and Kockelman [82], who conducted a dynamic ridesharing model, found that the gasoline hybrid electric (HEV) fleet outperformed the EV fleet while being more profitable, offering response times of 4.5 min compared to 5.5 min of the electric fleet.

3.1.7 User behavior

Modeling user behavior in the presence of SAVs is critical for planning and designing such systems. In this context, Auld et al. [8] used the POLARIS ABM platform to test different levels of penetration of SAVs in the Chicago metropolitan area, IL. Hao and Yamamoto [44] investigated travel behavior in Nagoya, Japan, considering travelers’ intention to use SAVs and their perceptions about ownership and sharing of their private vehicles. In their paper, Kamel et al. [63] investigated user preferences towards choosing SAVs, using an ABM in MATSim. Factors considered included age, gender, and income, which affected preferences for SAVs. Lokhandwala and Cai [86] developed an ABM in which different types of driver preferences for SAV were assumed. A study by Wang et al. [128] developed an ABM for simulating platooning formation and the interactions between SAVs and real-time travel requests,the objective was to capture the real-time behavior of SAVs as trip makers and then to evaluate the performance of an AMoD system with coordinated platooning formation. From a different perspective, Al Maghraoui et al. [4] examined travelers’ willingness to use a SAV service depending on their current transportation mode in Paris, France. Nahmias-Biran et al. [101] used the ABM platform SimMobility in Singapore to analyze AMoD policies with respect to income and accessibility. Finally, Zhou et al. [139] investigated the relationship between AMoD services, accessibility levels and relocation decisions of urban residents; again, SimMobility was used for that purpose.

3.2 SAV impacts

The identified studies on SAVs and ABMs have yielded remarkable and at times contradictory results, which provide valuable insights into the impact of integrating SAV services in urban road networks. It should be emphasized that the modeling techniques used, and especially the simplifications or assumptions applied, play an important role in the expected results [116]. The impacts are grouped with respect to traffic, modal shift, land uses, parking, the environment, energy consumption, and operational and service aspects.

3.2.1 Traffic

Several papers have examined impacts to traffic at the network level, resulting in interesting, yet contradicting findings. According to Fagnant and Kockelman [32], a system of SAVs could save users ten times as many cars as they would need for private-vehicle travel, but that would induce about 11% more trips. Along the same lines, Fagnant et al. [33] argued that SAVs are expected to generate 8% more VKT in a low SAV penetration case, due to unoccupied trips or relocation issues. A study by Auld et al. [8] reported that SAVs could increase total VKT by about 4%, if an 80% increase in SAV service capacity is achieved. According to Javanshour et al. [58], an AMoD system with 10% penetration has the potential to radically reduce the existing private vehicle fleet by 84%, while maintaining the same travel demand. However, this would result in an increase of 29% to 77% in VKT depending upon the type of SAV service.

In line with previous studies, Dia and Javanshour [30] found that reducing the number of vehicles travelling in the case of an AMoD system, increased total VKT by 29% in case of zero waiting time and by 10% when the waiting time is up to 5 min, using Melbourne, Australia; these findings were later verified by Javanshour et al. [58]. Another study conducted by Oh et al. [105] concluded that an unrestricted implementation of AMoD could lead to a significant increase in network congestion and VKT in Singapore. Harper et al. [46] demonstrated that SAVs in downtown Seattle, WA, averaged an 5.6–6.4 km/day (3.5–4.0 mi/day) additional distance of travel, and that at a high penetration rate (50–100% AV), AVs would travel an additional 9.0–13.5 km/day (5.6–8.4 mi/day). Finally, a study by the ITF [55] for Lisbon, Portugal as a case study, found that different autonomous taxi services would yield an increase of up to 89% in VKT. Similarly, Hörl et al. [50] found that the use of autonomous taxis for individuals in Zurich, Switzerland would increase VKT up to 60%. Other studies have demonstrated that the use of SAVs will lead to reduced vehicle trips and VKT. In this context, Heilig et al. [47] argued that in the case of Stuttgart, Germany, if 85% of private vehicle trips were undertaken by SAVs this would result in a 45% reduction in vehicle trips and a 20% reduction in vehicle kilometers. Lokhandwalaa and Cai [84] found that SAV based ridesharing in New York, NY, increased occupancy from 1.2 to 3, and reduced overall mileage up to 55%. Finally, Yan et al. [130] reported that the introduction of SAV ridesharing services could decrease VKT by 17%.

In terms of travel times, Bischoff & Maciejewski [14] reported that an increase of 17% in total network travel times due to idling would be expected because of SAVs, but higher congestion in certain points would not necessarily be expected. Llorca et al. [81] again found that total network travel times for both autonomous vehicle and conventional trips would increase when SAV fleets are introduced, but also argued that peak hour congestion would be reduced. According to Bösch et al. [17], if a small percentage of the population would use SAV, SAV systems can only reduce travel times at the cost of significant modal shifts. On the contrary, Hamadneh & Esztergár-Kiss [43] observed a reduction in both travel time and travel distance. Venkatraman and Levin [121] found encouraging results in reducing the total travel time of people for different SAV fleet sizes and demand levels.

3.3 Modal split

In addition to travel behavior and kilometers or miles traveled, the literature reports that SAVs will certainly affect the modal split. According to Chen and Kockelman [20], the potential share of SAVs in Austin, TX, would likely to range from 14 to 39%, when competing with conventional cars and public transportation. Liu et al. [79] showed that travelers who travel longer distances prefer SAVs over private human-driven vehicles because of avoiding driving burdens. ITF [55] indicated that in Lisbon, Portugal, SAVs would reduce the use of private vehicles, by 23%-65% and Kamel et al. [63] found that in Paris, France, the share of SAV trips would range from 3.8% to 5.3%. The findings of Ishibashi & Akiyama [54] for Tokyo, Japan showed that about 14%-32% of the population would switch to SAVs, and that those who traveled 2.0–8.0 km by train or bicycle would likely switch to SAVs. On the contrary, a study by Cyganski et al. [24] for Braunschweig, Germany, showed only minor changes in the modal split because of SAVs, primarily due to the short distances involved. Similarly, Nahmias-Biran et al. [102] found that the mode shift from active transportation and public transport to AMoD in Tel-Aviv, Israel, was insignificant, due to the travel costs of AMoD services. Regarding taxis, Liu et al. [77] concluded that SAVs could attract more users than conventional taxis due to lower costs. Finally, Ben-Dor et al. [13] reported that if parking or congestion pricing schemes would be applied, SAV users would come equally from PT and private vehicles.

3.3.1 Interaction with land uses

With respect to SAVs impact on land uses, Kim et al. [69] showed that the full integration of SAVs into the urban street environment was expected to lead to dispersed spatial structures; the authors simulated the spatial impacts of SAVs and assumed that the introduction of SAVs would change the location choices of travelers in addition to the attractiveness of regions. On the contrary, Zhou et al. [141] demonstrated that the implementation of AMoD would not lead to outward migration nor intensify the disparity between home and work locations.

3.3.2 Parking demand

Due to their operational model, SAVs will change parking demand; this potential change has been explored in a few studies. Zhang et al. [137] estimated the impact of SAVs on demand for parking demand and reported that SAV systems would be able to eliminate up to 90% of parking demand for customers using SAVs, for a low SAV penetration rate of 2%. According to Dia and Javanshour [30], demand for parking would decrease from 58 to 83%, depending upon travelers waiting time for SAV service. The study of Harper et al. [46] showed in the case of Seattle, WA, parking revenues would radically decline if SAV penetration increased, and that parking services would eventually become economically not viable.

3.3.3 Environment

There are several studies looking at the environmental footprint of SAVs. Zhang, et al. [138] first reported that SAVs can be environment friendly for cities in the long run. In the same direction, Martinez and Viegas [91] emphasized that if private vehicle were replaced by shared ones, this would significantly reduce CO2 emissions. According to Liu, et al. [78], found that different SAV operating strategies could reduce emissions by up 19%. Lokhandwala and Cai [85] showed that a SAV taxi fleet could reduce taxi CO2 emissions in New York City, NY by up to 861 tons/day. Lokhandwala and Cai [84] claimed that dynamic ridesharing of SAVs could decrease carbon emissions by up to 866 tons/day in New York City, NY. Yao et al. [132] found that exhaust emissions from the SAVs fleet in Hangzhou, China, would be 12.3% lower than those from the human-run fleet. According to Oh et al. [106], the application of AMoD in Singapore was found to lead to a reduction in vehicular emissions, particularly NOx and PM, by 4.3–5.7% and 5.6–8.2%, respectively. In contrast, Lu et al. [87] found that the environmental impact of SAVs, when these replace conventional vehicles, does not show significant improvement, because of the emission intensity of the local power grid. When it comes to traffic noise, Zwick et al. [142] demonstrated that replacing all car trips with an autonomous ride-pooling system based on bus stops leads to a dramatic reduction in noise in residential areas in Munich, Germany.

3.3.4 Energy consumption

As SAVs are expected to largely adopt electric propulsion and technologies (Kovačić et al., 2022), energy consumption and charging are important and have been addressed by different studies. According to Chen et al. [21] each SAV with a range of 80 miles would replace 3.7 private vehicles and each SAV with a range of 200 miles replace 5.5 private vehicles, under Level II (240-V AC) charging. Sheppard et al. [112] found that all mobility in USA could be served by 12.5 million SAVs, with an energy demand of 1142 GWh/day (8.5% of the national electricity demand in 2017) and a peak charging load of 76.7 GW (11% of the U.S. electricity peak). In a study by Oh et al. [106], the implementation of AMoD services in Singapore was estimated to result in a rise in energy consumption by 16.94–24.33%.

3.3.5 Operations and services

The last category refers to the operational and service-related impacts of SAVs, including costs and revenues, waiting times, vehicle fleet size, and other trip parameters. It should be underlined that many papers have laid their interest in this specific area. Beginning with costs and revenues, Fagnant and Kockelman [34] reported that dynamic ridesharing reduces total service times and travel costs for SAVs users, even after accounting for extra passenger pick-ups, drop-offs and non-direct routings. The authors also showed that a private fleet operator paying $70,000 per new SAV could earn a 19% annual (long-term) return on investment. When offering SAV services at $1.00 per mile of a non-shared trip. Among other studies, Hörl et al. [50] found that even under conservative pricing a large share of travelers would be attracted to SAVs, and Liu et al. [79] indicated that higher SAV fare rates allow for a larger private vehicle replacement (ranging from 5.6 to 7.7 private vehicles per SAV). In a subsequent study, Liu et al. [78] argued that the non-detour and detour sharing SAV strategies can reduce operational costs by 16% and 24% respectively. Farhan and Chen [35] focused on vehicle occupancy and demonstrated that allowing multiple occupants improves service rate as well as system-wide benefits from $1.34 M to $1.52 M. Yao et al. [131] found the revenue of a SAV system in Hangzhou, China to be approximately four times as the daily system operation costs. Finally, a study by Zhang and Guhathakurta [135] for Atlanta, GA, also indicated that a SAV service would yield reduced commute costs.

Regarding passenger waiting time, Shen and Lopes [110] proposed an algorithm for a SAV system, which could reduce average passenger waiting time by up to 29.82%. As noted by Chen and Kockelman [20], pricing strategies that attempt to match available SAVs supply with expected travel demand can reduce average waiting times by 19 to 23%. Liu et al. [78] claimed that SAV detour and non-detour sharing strategies can reduce waiting times by 62% and 82%, respectively. In the same direction, Hyland and Mahmassani [53] found that SAV dynamic strategies involving drop-off SAVs on the route in the assignment problem reduce traveler waiting times. Findings by Luo et al. [88] indicated that average passenger waiting times fell within an acceptable range in different SAV operations in Gunma, Japan. Wang et al. [128] reported that an AMoD system with vehicle platooning formation significantly affects the average waiting time of users. According to Pulhès and Berrada [108], for different levels of SAV fleet size and penetration, the maximum average waiting time is 15 min, and during peak periods this value can be as high as more than 17 min. Also, De Souza et al. [27, 28], who focused on the impact of a repositioning method for SAVs, also observed an improvement in SAVs waiting times.

Concerning SAVs fleet size itself, Wen et al. [126] noted that SAVs fleet size is the outcome of a tradeoff between service levels and operating costs. Also, according to Javanshour et al. [58], there is a strong quadratic relationship between the SAVs fleet size and VKT when demand is kept constant.

When it comes to the impact of SAVs on the total vehicle fleet in a region, SAVs have the potential to dramatically reduce the number of vehicles needed to meet current travel demand [55]. Looking at the results of various studies, Spieser et al. [117] reported that the fleet size of SAVs needed to serve the entire population of a city is up to 66% smaller than the fleet of private cars. Dia and Javanshour [30] claimed that SAVs in Melbourne, Australia, could lead to a large reduction in the total number of vehicles needed to meet travel demand (reduction of 43% to 88%). In the same direction, Lu et al. [87] pointed out that the needed SAV fleet to meet daily commuter demand with waiting times of less than 3 min is only 20% of the conventional single occupant car fleet. Wang et al. [123] found that during rush hour, about 240–250 SAV vehicles are needed to meet demand, while during off-peak hours, only about 30 vehicles should be in operation. Liu et al. [78] found that non-detour and detour vehicle sharing strategies can reduce total vehicle fleet size by 19% and 27%, respectively. Another study by Lokhandwalaa and Cai [84], highlighted that switching from traditional taxis to shared autonomous taxis could potentially reduce fleet size by 59% while maintaining service levels and not significantly increasing travelers waiting time. In the same line, Llorca et al. [81] showed that it is possible to replace three conventional cars with one autonomous taxi while meeting the demand for trips with reasonable waiting times. In addition, it is found that SAVs could replace 5 to 9 private vehicles in Austin, TX, while maintaining adequate service levels [21, 33]. Bischoff and Maciejewski [14] displayed that a fleet of 100,000 autonomous taxis is sufficient to replace the passenger car fleet in Berlin, Germany on a typical weekday with a high quality of service for customers.

4 Discussion, policy, and future research recommendations

From the papers reviewed, critical insights were gleaned regarding the use of ABMs to explore the impact of SAVs on future urban environments. One notable observation from the analysis of the selected papers is the uneven geographic distribution of case studies examining SAVs and their effects on urban environments. It appears that specific cities or regions are more actively engaged in investigating the potential impacts of SAVs (Table 1).

Table 1 Case studies per city worldwide

Notably, Singapore leads with 10 studies on ABMs for SAVs, followed by Austin (TX) with 8, Sioux Falls (SD) with 6, New York City (NY) with 5, Munich with 3, Melbourne with 3, Bloomington (IL) with 3, Zurich with 3 and Chicago (IL) with 3. These numbers indicate varying levels of research activity in different regions, possibly reflecting the interest, funding, or academic and industry focus in these areas on the topics indicated.

ABMs were also found to be extensively used for analyzing and assessing impacts of SAV services; these impacts are summarized in Fig. 2. Overall, most publications (regardless of the assumptions and the setup of the ABMs employed) conclude that both vehicle-kilometers travelled, and vehicle hours will increase, implying a negative impact in the performance of transportation networks, because of SAV operations. In addition, SAVs are expected to affect the modal split in their favor, reshape the spatial structure of cities, creating conditions for more compact urban agglomerations (due to their intrinsic property of being shared), and contribute to a noticeable reduction in parking demand, supporting the transformation of public spaces. As for environmental impacts, several papers argue that SAVs will support the reduction of emissions and noise pollution. Finally, impacts related to SAV operations include the improvement of traveler waiting times, and the replacement of conventional vehicles.

Fig. 2
figure 2

Shared automated mobility ecosystem through ABM simulation studies

4.1 Policy recommendations

Identified SAV impacts can serve as the foundation for documenting policy recommendations that ensure an appropriate transition to an automated age, which still puts humans at the center [107]. First and foremost, there is a consensus among studies indicating an increase in VKT and VHT due to the introduction of SAVs. This means that urban environments are dominated by motorized vehicles for most of the day, affecting sociality and interaction between people in public spaces [25]. Therefore, it is suggested to prioritize constraints regarding the use of SAVs in urban areas, e.g. by excluding them from certain central areas. Most importantly, urban development should focus on proximity solutions, such as x-minutes cities [83], which can reduce the distances and time between home and facilities and thus create suitable conditions for active and non-motorized travel [70]. This future endeavor could mitigate the negative impacts of more VKT, and time spent traveling.

In terms of modal split, the literature argues that SAVs will gain travelers over other modes, implying that motorization will again prevail in urban areas. Therefore, SAVs should act as a substitute for private vehicles, and not as a competitor to active transportation and public transport [61]. Therefore, planning will need to consider fruitful synergies of SAVs with conventional public transport and active modes, such as a common street classification systems and coordination of services.

As for land uses, SAVs were found to be much better suited to compact urban areas than private AVs, which are expected to lead to dispersed urban patterns [38]. Therefore, future cities should encourage mixed land uses and densification to combat urban sprawl trends and facilitate the movement of SAVs compared to AVs. Another issue that should be of concern are the inequalities that can arise from the geographical distribution of SAV services [95]. To this end, SAV deployment schemes (standard or on-demand routes) should sufficiently serve the urban fabric (taking into account strategic visions for urban development and potential exclusion areas), ensuring solid accessibility for all who live and work in a city.

Moving on to parking demand, as significant decrease is anticipated, focus should be given on transforming these liberated spaces into human-oriented places [36]. The reallocated space should be reserved for pedestrians, cyclists and people with reduced mobility to encourage active mobility and reclaim urban space for vulnerable road users. As for policies related to the environmental impact of SAVs, apart from the potential benefits for the urban environment of replacing private cars with SAVs, synergies with active transportation modes are also a promising approach.

4.2 Future research

Research on planning and analyzing SAV services and the exploitation of ABMs for that purpose has been extensive in the recent years [11, 49, 59, 64, 74, 89, 100, 125, 129, 136]. However, there are research areas on SAV operations and services, for which ABMs can be used for investigating, assessing, and validating this upcoming mobility landscape (Fig. 3). First, SAVs cannot fully function in today’s road network and therefore need suitable road infrastructure. This implies that new road functional categories that facilitate the circulation of SAVs should be introduced [120], and evaluated with suitable tools such as ABMs. Second, it is widely acknowledged that in the complex and dynamic urban structures of the 21st century, conventional public transportation may not be enough. Therefore, public transport should deploy automated mobility solutions [114], which can benefit urban transport systems, and become a driving force towards multimodality [3, 23, 99]. Third, SAVs can be a solution to the transportation needs of often captive travelers, such as elders, children, and people with disabilities (see also Nahmias-Biran et al., [102]). Therefore, future studies should concentrate on designing, implementing, and evaluating multimodal transportation networks and systems focusing on targeted traveler groups, with Shared Autonomous Vehicles (SAVs) playing a crucial role. Obviously, given the complexity of such systems, ABMs can be invaluable tools for their planning, analysis and evaluation.

Fig. 3
figure 3

Directions for future research

Finally, the adoption of SAVs as a transportation mode will depend on public acceptance and user preferences. Research should focus on individuals’ willingness to share rides with others and factors that influence this decision. For instance, Lavieri and Bhat [71] found that people are less sensitive to the presence of strangers when commuting to work than for leisure activities. In terms of acceptance, Nikitas et al. [104] also demonstrated that people across the globe perceive SAVs as a crucial employment disruptor. Consequently, it is a challenge for future studies to consider public acceptance parameters in the design and evaluation of SAVs services, and their subsequent introduction in ABMs.

5 Conclusions

This paper reviewed the literature on the use of ABMs on planning, analyzing, and assessing the impacts of SAV operations and services. The first part of the review identified different ABM applications in the context of SAV services and operations, suggesting the wide acceptance and suitability of these tools for tackling SAV problems. The second part of the review investigated the impact of SAVs in the network performance mode choice, land uses, the environment and parking demand. Most of the studies reviewed suggested, that while network performance will worsen and demand will shift to SAV usage, there will be environmental, operational and land use related benefits because of SAV services. Based on the review outcomes, policy suggestions for the introduction of SAVs and future research recommendations included the need to consider SAV services in conjunction with land use measures, and the integration of SAV in multimodal transportation systems. Overall, this paper aspires to serve as a valuable basis for future research and help policy makers, stakeholders, and local communities into considering appropriate, efficient and sustainable ways of introducing SAVs in future cities.