JSDEWES: A Systematic Review of the Impact of Autonomous Vehicles on Transportation Networks

Empirical studies on the integration of AVs play a pivotal role in the area of research as societies transition towards increasingly using automated transportation systems. Understanding the real-world impacts and challenges of AVs serves as a building block for infrastructure preparation and policy regulation across various industries. Many studies examine the possibilities and difficulties of improving AV performance in existing transportation systems. These studies suggest ways to enhance skills including, but not limited to, collision avoidance and lane-keeping. Deep learning along with the integration of cutting-edge technology into AV systems significantly improve traffic flow and collision avoidance. Empirical studies in this domain aim to spotlight into the practical implications of AV integration, addressing a wide range of topics such as safety, traffic efficiency, societal acceptability, regulatory frameworks, and infrastructural requirements. These studies employ rigorous data collection methods and analytical approaches to contribute valuable evidence-based knowledge to inform decisionmaking processes and shape the future of transportation in the era of autonomy. AV platoons can significantly enhance consumption of energy [31] by co-optimizing gear positions and speed [32]. Furthermore, it saves up to 16% on energy and can reduce travel time by 5% [32]. This is evident in reduction of energy consumption by the framework created for ecodriving that uses SpaT data, which in free flow scenarios, determines the energy sufficient speed profile upstream and downstream [33]. The development of AVs from theoretical concepts to practical applications showcases its potential to enhance traffic flow and safety [34]. Many researchers agree on the ability of AVs to contribute to energy consumption reduction, which makes its utilization advantageous even in the environmental sector. Energy consumption can be reduced through improved driving strategies, such as limiting needless acceleration and braking, hence eco-driving approaches can improve the energy economy of AVs. By accounting for traffic, road conditions, and vehicle state, the suggested dynamic eco-driving control framework enables AVs to function more effectively in real-time. AVs may adjust to changing driving situations and enhance fuel economy by combining this control framework with sophisticated algorithms [31]. For instance, an eco-driving framework uses Signal Phasing and Timing (SPaT) data to allow cars to change their speed beforehand to reduce energy usage. By communicating signal timings to cars, the framework enables them to adjust their speed profiles and minimize stop-and-go behaviors, which lowers fuel consumption [33]. This allows for significant energy savings while preserving safe, smooth driving behaviors. Furthermore, in CAVs, co-optimizing gear positions and speed ensures that a vehicle runs as efficiently as possible based on real-time traffic predictions, which helps cut down on energy consumption. Driving can be made more efficient by minimizing fuel usage by modifying the vehicle's speed and gear in accordance with anticipated traffic circumstances [32]. Moreover, the technology used in AVs can further be developed to assure a safer travel for the AVs and the surrounding vehicles. Although the Reinforcement Learning (RL) used in AVs operates smoothly, the Hierarchical Program-triggered Reinforcement Learning (HPRL) was proposed [35] and it divides tasks into simpler ones that are each run by a set of programs and RL agents. Additionally, a costeffective solution would be prepping AVs with low-power hardware computers and dedicate edge servers for heavy assignments [36]. Through this approach, it can be guaranteed that the probability of hardware failure is mitigated; thus, a safer ride for AVs and their neighboring drivers. The utilization of stitching technology to increase the detection of obscured objects was observed through combining insights of surrounding Autonomous Driving Systems (ADS) [37]. The average detection rate of the proposed system increased by 45.4% relative to the single ADS [37], promoting higher safety and efficiency. Furthermore, considering a coordination system, where AVs and CAVs are linked to a computer framework based on cloud, can reduce fuel consumption and travel time, in addition to great enhancements in average velocity [38]. The high penetration rate of level 2 automation CAVs also tends to reduce energy usage per vehicle [39]. As observed, researchers examine benefits of AVs beyond those affecting traffic flow, such as fuel or energy consumption. It is critical that the integration of AVs is studied thoroughly as it may cause revolutionary changes. Through schemes like Action Governor (AG), which is a control system that improves system performance, AVs can be integrated in traffic systems in a safer manner [40]. Table 1 provides a summary of the empirical studies’ advantages and disadvantages.

The possible effects of AVs on travel time and safety are better understood through simulation research. Stakeholders can predict how traffic signals, pedestrians, and other elements like energy use and environmental concerns will interact by modeling different scenarios. Numerous studies provide thorough illustrations of how AVs affect various user groups. According to the simulation results, AVs affect three different groups of people: part-time workers, senior retirees, and long-distance drivers with high incomes. The study's strength is demonstrated by the comprehensive analysis it offers on how AVs impact these various demographic groups. A significant drawback is that the results might not apply to all populations, making sure the advantages last is the difficult part [41]. Extensive insights into the behavior and implications of AVs are provided by simulation studies. For example, microsimulation is used to simulate the behavior of individual vehicles. Its strength is that it offers in-depth insights into certain vehicle behaviors. Its breadth is constrained, and it misses network-wide impacts, which is a limitation. One of the biggest challenges is making sure that real-world circumstances are accurately represented. In contrast, a much larger road network is simulated by macroscopic simulation. Its ability to capture more extensive traffic patterns and network impacts demonstrates its potency. Its inability to provide specifics on the behavior of individual vehicles is a limitation. It is still difficult to strike a balance between scope and detail for precise forecasts. Agent-based simulation examines how drivers behave in conjunction with pedestrians. Its merit is that it offers a thorough understanding of the interactions between different agents. Its computational complexity and intensity, however, is a drawback. It's difficult to incorporate a variety of behaviors into a coherent paradigm. Simulations often yield good results for AVs, with shorter trip times because of better vehicle performance and faster average speeds. As AVs are expected to closely comply with safety requirements, collision rates decline. Because of their optimal driving practices, AVs also result in reduced energy consumption and carbon emissions. Use of parking spaces is the subject of another important study [42]. As per the study's conclusion, there could be a 50% increase in journey time for autonomous vehicles;

however, it is possible to reuse 15% to 32% of the parking area. This reinforces the study's strength by highlighting an important secondary benefit of AVs in urban planning. A drawback is that the conclusion of longer journey times is at odds with the majority of other investigations. This is an issue of optimizing urban space, while maintaining commute time efficiency. A different research study provides evidence that AVs can greatly increase intersection efficiency [43]. The study demonstrates the potential to enhance time-to-collision results and traffic efficiency at crossings by over 18%. This highlights the study's strength and shows a definite effect in high-traffic regions. A drawback is that these upgrades can necessitate considerable infrastructural modifications. The implementation of these improvements in pre-existing urban areas is a problem. A human-like speed planning technique is suggested to reduce travel time on predetermined routes [44]. This highlights the strength of the study and improves travel time efficiency by imitating human driving behaviors. Its downside is that it depends on predetermined routes, which reduces flexibility. Integrating this approach with dynamic real-world traffic conditions is a challenging task. The effective deployment of AVs depends on safety regulations and procedures. A framework to handle problems with mixed traffic situations and human-machine interaction is proposed [45]. The study examines the shortcomings and difficulties in creating shared safety standards for autonomous vehicles. This highlights the study's strength and provides a thorough approach to safety standards. A flaw in the framework would be that it requires a lot of validation and modification. Reaching a worldwide agreement on safety regulations is still quite difficult. Furthermore, it is anticipated that as AVs usage increases, fewer parking spots will be required, freeing up space in cities for other purposes. The assumptions entered into the simulator, such as technical details and interactions, form the basis of the simulation results. In order to ensure accuracy, simulation results must be validated using real world data. For AVs to be widely used, it is imperative that laws, rules, and technical standards are continuously improved. Stakeholder cooperation can result in the creation of thorough regulatory frameworks that address these problems and encourage the safe and effective use of AVs. Realizing the full potential of AVs also requires developing a global database and encouraging public adoption. Another crucial aspect is how AVs and pedestrians interact. The necessity of AVs having knowledge of pedestrian crossing behavior is considered in some studies [46], particularly in situations when a vehicle's camera may be obscured. This study shows the benefits of edge computing and real-time data processing in enhancing AVs safety. But in places with poor connectivity, the dependence on cutting-edge computing infrastructure presents a problem. In order to improve data transmission and processing speed, future research may investigate merging edge computing with technologies such as 5G networks. Zhang et al. [47] also examined the coordination between AVs and pedestrians. The project focuses on enhancing pedestrian and CAVs cooperation to increase traffic efficiency and safety at industrial sites. The findings may not be applicable everywhere, especially in urban and suburban settings, but their strength rests in addressing safety in complicated environments. Adapting these techniques to various contexts is a difficulty. Simulation analysis aids in forecasting and comprehending the effects of implementing autonomous vehicles in diverse contexts. These insights aid in achieving optimal traffic flow and safety, which in turn lead to suggestions for essential actions by stakeholders and policymakers. The stage can be set for a time in the future, when AVs are integral to the development of safer, more effective, and environmentally friendly transportation systems, by tackling the related difficulties and concentrating on ongoing advancements.

The effectiveness of integrating AVs has been assessed in accordance with the penetration rate of AVs. Variable factors were considered, including but not limited to travel and delay times, road capacity, and traffic speed. AVs can have great impacts on traffic flow if integrated in the right manner, as it enhances overall traffic behavior and increases safety near bus bays [48]. For urban networks, an average reduction of 31% in delay time and 17% in travel time was reported for increased AV penetration rates [49]. A penetration rate surpassing 40% has been noted crucial for a prominent network enhancement [50], [51]. However, for wider road segments, specifically those compromising three or four lanes, an AV penetration rate exceeding 30% exhibited notable traffic flow improvements. This indicates that wider roads derived greater benefits from the integration of AVs, as evident by the clearer impact on the mitigation of average delay time experienced on those road segments [49]. Besides wide roads, congested roads also experience substantial benefits from the presence of AVs [52], where travel time exhibits a reduction of 7.1% [53]. Travel time also reduces by 59%, when CAV are controlled at intersections by a system of CAVSMulti-Agent Deep Reinforcement Learning (MADRL) alongside an advanced Reinforced Autonomous Management system (adv.RAIM) [54]. The Optimal Routing and Signal Timing (ORST), which is a CAV control strategy, reduces travel time by 49% [55]. Travel time tends to reduce in the presence of AVs in high penetration rates [56] and even at low penetration rates it exhibits a reduction by 20% and 4% [57]. It also undergoes notable reduction when a hierarchal control model called COOR-PLT is utilized [58]. Different AV scenarios with variable levels of penetration and automation rates exhibit substantial disparities in delay time mitigation, ranging from 3.50% for basic automation and low AV penetration to 28.53% for a full automation [59]. Through research, it has been proven that AVs have the ability to significantly reduce Time Headway (THW) and increase traffic capacity; thus, alleviating traffic congestion [60]. Congestion can also be reduced through a framework that utilizes collision-avoidance approach called the multi-agent reinforcement learning [61]. Congestion can be mitigated through the use of Roadside Units (RSU) [62], or even entirely eliminated at high AV penetration rates [63]. At the autobahn segment, a 100% AV penetration rate improved the average density significantly by 8.09% during peak hours. In addition, there was an improvement in the average travel speed by 8.48% improvement on autobahn and 7.86% improvement on weaving segments in the same scenario; as a consequence, the average travel time improved by 9% when compared to the scenario without AVs [64]. Furthermore, the network performance of Daejeon City showed that when AVs operate conservatively, passenger’s safety and comfort are promoted [65]; however, these conservative driving behaviors result in trade-offs, as evidenced by a decrease in road capacity of approximately 35%, when AVs comprised 100% of its road vehicle mix [66]. However, road capacity increases by 35% to 59% as per [67], when the AV penetration rate ranges between 25% to 35%. The AV simulation results are summarized in Table 2.

Despite the fact that AVs contribute greatly to improving traffic factors, they give rise to certain concerns; such as potential loss of situational awareness, over dependence on automation, and system failure. This indicates the significant distinction in the behavior of drivers, when driving and trailing a human-driven vehicle versus an AV. This alteration in driver behavior is shown in a decrease in both the mean and variance of THW observed, when human drivers follow an AV [68]. When driving near AVs with short THW, human drivers tend to reduce their own THW and spend more time below their critical THW. Not only that, but driving alongside highly automated vehicles can also diminish the human driver's situational awareness and may increase the risk of drowsiness, especially in cases of light traffic [65]. Nevertheless, it is important to note that in most cases AVs tend to stay at an adequate distance from other vehicles and road users [69]. A study carried out by Kanazawa University [70], which focused on the behavioral algorithm of AVs, demonstrates that with an increase of 10% to 45% in the proportion of AVs within the traffic mix, there is an escalation in the delay observed between OriginDestination (OD) intervals. Nevertheless, as the penetration rate increases from 45% to 50%, the delay gradually diminishes. Furthermore, with the proportion of AVs ranging from 50% to 100%, the delay remains consistent. These analytical findings highlight the influence of integrating autonomous vehicles on vehicle’s behavior and provides insight on the importance of identifying suitable distribution scenarios, and implementation areas within societal frameworks [70]. The impact of AVs on varying age groups was assessed through simulation; three work types were considered: drivers who travel long distances and gain high income, elderly drivers that are retired, and drivers who are part timers [71]. Results revealed that AV penetration impacts each group differently depending on the demand of AVs on the road system and Value of Travel Time (VOT) [72]. It is also important to note the preferred AV transportation modes since it affects human behavior and comfort. In another study [73], travelers tend to incline more towards individual-ride AVs and less towards shared AVs. Moreover, in the case of AVs being utilized as a transportation mode, 64% of the drivers are going to utilize AVs whereas only 15% will use public transportation modes [71]. It is noteworthy to mention that travel patterns and congestion levels are affected by the human’s in-vehicle activities [74].

Achieving the required safety levels for AVs involves eliminating human errors, which are the primary cause of accidents. Deep learning algorithms train AVs to avoid such errors by adhering to safety regulations like maximum allowed speed and safe distance. However, technology does not come free of risk. Expected risks include software malfunctions, sensor quality issues, and the critical risk of AVs being unable to predict the behavior of human-driven vehicles. These risks are integral to the broader challenge of capability integration. Fortunately, new risk management strategies are emerging to enhance and mitigate potential risks. These strategies include redundancy for critical systems (e.g., steering and braking systems) to ensure safety in case of a component failure. Advanced algorithms can analyze risks in real-time and provide suitable and safe solutions. Additionally, Vehicle-to-everything communication system (V2X) technology provides realtime information, enabling AVs to interact with risks as they occur. One effective strategy is to simulate expected risky scenarios, allowing AVs to be trained in advance before real-life deployment [75]. The outspread implementation of AVs holds the promise of reducing traffic crashes and enhancing road safety. Safety is further enhanced by pedestrian detection which is accounted for with the use of accurate positioning systems with sensors [76]. AVs contribute remarkably in enhancing safety, particularly at high penetration rates [77], [78], as they operate with shorter headways in order to optimize road capacity and mitigate delays. At signalized intersections, AVs lead to a noteworthy reduction in conflicts ranging from 20% to 65% with AV penetration rates between 50% and 100%. Similarly, at roundabouts, conflicts are decreased by 29% to 64% with a 100% AV penetration rate [78]. These percentages are demonstrated in Figure 2. This is also evident in the findings of another study, which focuses on a specific freeway crash hotspot in Wuhan, China. Results reveal that at penetration rates below 50%, no significant improvement in safety measures is exhibited such as conflicts, acceleration, and velocity difference. However, as penetration rates exceed 70%, traffic flow at the freeway hotspot demonstrates fewer conflicts, reduced acceleration, and decreased velocity difference, particularly when CAVs are confined to their lanes [79].

Despite its wide uses and great benefits, the safe implementation of AVs is critical in promoting efficient utilization. Nevertheless, road accidents are inevitable and have to be thoroughly considered and assessed. One study concluded that when Time To Collision indicator (TTC) is less than 4s, AVs are more likely to collide with other vehicles [80]. However, when compared to human driven vehicles, the rate of crashes and fatalities was decreased when AVs were utilized [81]. According to Parseh and Asplund [82], accident analysis is crucial to propose decision-making strategies that are useful in studying the implications of these collisions. Three decision making strategies have been proposed, including: advanced Safety Surrogate Measures (SSM) [82], [83], where severity levels are examined; unconsidered heterogeneity that results from collision restructuring must be included; collision configurations are to be recorded for feedback and study [82]. Risk analysis was also studied, where an advanced Network-Level Safety Metrics (NSM) set is proposed to evaluate the safety of traffic flow. It was observed that the use of NSM regulates the probability of collisions [84]. Safety can also be promoted through Model Predictive Control (MPC) planner that creates route tracking, collision-free movements [85], and simulates car-following behavior [86]. A Soft Actor-Critic (SAC)-based driving policy was proposed [87], which predicts intervals and yields safer vehicle driving evident by the collision reduction of 15%. Not only does MPC planner improve safety, but it also enhances driver’s comfort through reducing acceleration [85]. Comfort is also promoted through Deep Deterministic Policy Gradient with External Knowledge (EK-DDPG), which is a suspension control strategy [58]. A significant regulatory challenge is the lack of unified testing protocols for AVs, alongside data security and privacy issues. The absence of standardized protocols means AVs cannot operate seamlessly across different countries due to varying standards. Infrastructure policies also need to be updated [88]; all road-building manuals must be reviewed and updated to include the required infrastructure for smooth autonomous operations. Decision-makers and stakeholders must work collaboratively to produce effective regulatory frameworks to address these challenges [89]. This includes defining liability, insurance, infrastructure requirements, information security, and privacy protocols to ensure the safe deployment of AVs across different regions. By addressing these issues, the way can be paved for the successful integration of AVs into the current transportation systems. The literature includes various studies that address the gaps and challenges in establishing safety standards for AVs. One study [59], investigates the primary gaps and challenges in creating shared safety standards for AVs. It proposes a comprehensive framework to facilitate future research on safety issues related to mixed traffic environments, interactions between AVs and pedestrians or cyclists, and the transition of control from machine to human drivers. This framework provides a structured approach to address these safety challenges, highlighting its strength in offering a thorough overview of current issues. However, the proposed framework may require extensive validation and adaptation to different regulatory environments. Future research could focus on developing specific protocols and guidelines for mixed traffic scenarios and improving human-machine interaction during control transitions. Another study [90], discusses the importance of providing AVs with information about pedestrian crossing behavior, particularly when a vehicle's camera may be obstructed. The study emphasizes the role of edge computing in enhancing safety-related reactions. The strength of the research lies in its emphasis on real-time data processing and the potential of edge computing to improve AV safety. However, reliance on edge computing infrastructure may pose challenges in areas with limited connectivity. Future work could explore the integration of edge computing with other technologies, such as 5G networks, to improve data transmission and processing speed, thereby enhancing the overall safety and reliability of AV systems. Addressing safety concerns for AVs requires not only eliminating human errors but also implementing robust risk management strategies and developing unified standards and protocols worldwide. Public acceptance is also crucial, as it influences the successful deployment of AVs. By focusing on these areas, the safety and reliability of AV technology can be enhanced, paving the way for its widespread adoption. The main advantages and disadvantages of the AV simulation results related to traffic safety are presented in Table 3.

Detailed analysis revealed that permitting AVs optional access to dedicated lanes exhibited improvements in capacity that yields reduced congestion and diminished variance in fundamental traffic characteristics [92], [93]. Moreover, AV-exclusive lanes allow an increase in AV speeds of over 15 km/h [93]. The efficacy of reserved AV lanes is dependent on the MPR and characteristics of the roadway [94]. Moreover, AV reserved lanes exhibited an advantageous potential at penetration rates surpassing 50% for two-lane highways and 30% for four-lane highways [95]. However, another study evaluated the effects relative to penetration rates and it was noted that AV dedicated lanes should be introduced at 20% to 30% penetration rates in order to optimize road capacity [96]. Moreover, a more sensitive capacity increase effect was observed for lower MPR [94]. The study of models demonstrated that dedicated lanes may not yield significant advantages at low CAV penetration rates [97]; nevertheless, their introduction at higher CAV MPRs, led to a notable reduction of 53% to 58% in traffic conflicts, and an increase in lane capacity [98]. The use of CAV lanes by human driven vehicles, when the presence of CAVs on these lanes is low, enhances travel time and speed [95]. Additionally, traffic crashes that were estimated from traffic conflicts decreased with the presence of CAVs, by up to 48% [98], [99]. Through simulation, it was observed that optimal safety benefits for CAVs in dedicated lanes result from an MPR combination of 40% human-driven, 40% 1^st generation AVs, and 20% 2^nd generation AVs. These findings could serve as a foundation for developing regulatory frameworks for local authorities and practitioners; in addition to the crucial insights provided into the safety implications of dedicated lanes for CAVs [98]. Initially, especially at low MPRs, turning a regular lane into a dedicated lane for CAVs may negatively impact total traffic flow. This detrimental effect usually lasts until about 50% of people adopt CAV. However, the most noticeable improvement in network speed occurs when the dedicated lane is implemented as part of an obligatory usage policy, particularly when the MPR is between 50% and 60%. The advantages of this approach tend to decrease as CAVs drive more aggressively, but they are further enhanced in situations with heavy traffic demand. Furthermore, the driving mode and the degree of CAV penetration have a significant impact on how well dedicated lane deployment works. When CAVs are driven in safe or conservative modes and the MPR stays below 80%, the mandatory-use policy continuously performs better than the optional-use strategy. The optional-use approach becomes increasingly beneficial beyond this point. This transition point moves to a 90% MPR when driving in a neutral manner. Notably, the mandatory-use strategy continues to produce better results at all CAV adoption levels, even in situations where CAVs display aggressive driving behavior [100].

A narrow (9 ft) lane designated for AVs influences driver behavior was assessed for safety, focusing on drivers in the adjacent lane to the right. With the use of a tailored driving simulator that replicates conditions of the Interstate 15 smart corridor in San Diego, participants were divided into two groups: one group drove next to a simulated 9 ft narrow AV lane, while the control group drove next to a standard 12 ft AV lane. The findings exhibited no significant differences in speed among drivers, but notable distinctions in lane positioning. It was observed that drivers adjacent to the 12 ft lane exhibited better lane centering compared to those next to the 9 ft lane. Furthermore, analysis suggested that female drivers tend to maintain greater distance from the 9 ft lane and showed poorer lane centering performance, yet outperformed male drivers when traffic was present in the right lane [101]. Table 4 presents the summary of the reviewed studies in this section.