Cost Analysis of Vehicle-Road Cooperative Intelligence Solutions for High-Level Autonomous Driving: A Beijing Case Study

. Te development of the vehicle-road cooperative intelligence can efectively resolve the current technical impediment and cost quandary associated with high-level autonomous driving. Nevertheless, the intelligent infrastructure entails initial deployment costs and ongoing energy consumption and maintenance costs, necessitating a comprehensive and quantitative analysis of the costs of intelligent infrastructure and the corresponding changes in comprehensive costs. Te cost evaluation model for the cooperative intelligent system is designed in this paper, considering the corresponding intelligent infrastructure layout scheme for diferent road types within the technical framework. Te intelligent confguration and corresponding cost transfer from roadside to vehicle side under the synergy efect is also analyzed. Using Beijing as a case study, the results indicate that the deployment of intelligent infrastructure will efectively reduce acquisition and usage costs of high-level intelligent vehicles and achieve a greater “reuse” efect by serving more intelligent connected vehicles (ICVs). Compared to the vehicle intelligence, collaborative intelligence will reduce cumulative total costs by more than ¥200 billion from 2023 to 2050, even with the inclusion of intelligent infrastructure’s costs.


Introduction
Autonomous driving has become a signifcant factor infuencing the development of the automotive industry.On the basis of fundamental autonomous driving capabilities, intelligent vehicles can replace humans in driving tasks.It not only provides passengers with a safer and more comfortable driving experience but also increases travel efciency [1][2][3][4], saves energy, and reduces emissions [5][6][7] and trafc accident rates [8][9][10].Chinese consumers have the highest level of acceptance of self-driving technology in the world [11].In China, the penetration rate of L2 autonomous passenger vehicles with combined assisted driving functions has reached 34.5% in 2022 [12].Intelligent vehicles represent the leading strategic position of future automobiles in terms of product form and key technologies.Te Chinese government is actively promoting the development of intelligent vehicles.Te research presented in this paper is set against this background.However, autonomous driving is still in its early stage of development.Te primary autonomous driving is gradually maturing and becoming commercialized, while intermediate and advanced autonomous driving remains dominated by trials and regional demonstrations.To address the technical challenges faced by intermediate and advanced autonomous driving, as well as the cost barrier to large-scale commercial implementation, ICVs powered by the latest information and communication technologies have been recognized globally as the future direction of automotive development [13,14].Intelligent infrastructure can ofer more extensive sensing data for vehicles, broadening the range and capability of vehicle sensors, improving safety, and resolving technical issues encountered in vehicle intelligence, such as sensing blind spots, beyond-thehorizon challenges, extreme weather, and an array of "perceptual long-tail" problems.Tis can also decrease the performance demands of vehicle-side sensing equipment.By providing a global path planning and decision optimization for ICVs, it can tackle challenges in mixed trafc scenarios, such as autonomous driving games, signifcantly reducing the need for vehicle-side computing power.Lowering the costs for vehicle intellectualization will contribute in increasing the adoption rate of intermediate and advanced intelligent vehicles.Currently, a growing number of countries and companies are emphasizing the route of vehicleroad cooperative intelligence [15].Te intelligent transportation infrastructure will be pivotal for ICVs' evolution.To address the current industrial constraints, the schematic design and deployment of intelligent transportation infrastructure should correspond with the needs of ICVs.Simultaneously, the deployment and operation of intelligent transportation infrastructure necessitate substantial and ongoing expenditures.A quantitative analysis of the cost of intelligent transportation infrastructure and the corresponding changes of comprehensive costs is necessary for national or local decisions to upgrade intelligent transportation infrastructure.
Currently, most related researches are focusing on the functions and technologies involved in intelligent transportation systems at the micro level [16][17][18][19], but few scholars have quantifed and analyzed the cost of intelligent transportation infrastructure and the resulting changes in the total cost to society.Chang proposed the roadside sensing confguration method, the vehicle fusion planning method, and the control method from the perspective of an intelligent networked cloud control system and simulated and verifed using the actual vehicle platform [17].Wan et al. reviewed research on queue control-related communication architectures, communication protocols, trafc models, and control methods, with application scenarios involving trafc fow optimization, dynamic queuing, and queue control [19].Other scholars have explored the deployment methods corresponding to the characteristics of intelligent roadside devices, such as communication environments, roadside units, and sensors [20][21][22][23][24][25][26].Liu et al. analyzed the demand for 5G base-station RSUs and outlined their technological architecture and fundamental functions, based on the development trend of the 5G and cellular vehicle-to-everything (C-V2X) network [21].Zhang investigated the optimization problem of network layout under the vehicle-linked sensor network architecture model and solved the optimal network layout scheme with the lowest deployment cost [22].Fu et al. examined the efcient solution problem of spatial arrangement of cameras for area coverage and proposed a probability-based binary particle swarm optimization technique to optimize both the number of sensors and the spatial arrangement scheme [24].Zhan et al. investigated optimal sensor placement for multitype sensor assignment on highways [25].Liu et al. examined the cost of various upgrading paths and deployment schemes for intelligent transportation infrastructure in two scenarios: open city road and closed highway [27].Terefore, researches on the cost of intelligent transportation infrastructure and the corresponding changes of comprehensive social costs can, therefore, close the current research gap in industry and academia and assist governments and policy makers in making decisions regarding the intelligent upgrading of transportation infrastructure.Te construction of intelligent transportation infrastructure is frequently city-based.Cities, especially frst-tier cities, have a large car population and road usage intensity; the deployment of intelligent infrastructure will generate greater social benefts and amortize its costs through greater "reuse" of large-scale feets.With 685 million motor vehicles, Beijing ranks frst among Chinese cities [28], and it is also the most congested city in China [29].Intelligent transformation of transportation is required to address the city's safety, congestion, environmental, and energy problems.
To fll the current research gap, this paper evaluates the total cost of transportation infrastructure to achieve an intelligent upgrade.Te cost of feet intelligence and total social cost are compared and analyzed under two scenarios: collaborative intelligence and vehicle intelligence.First, the architecture of the vehicle-road cooperative intelligent system is established as the theoretical basis.Ten, the cost evaluation model for the vehicle-road cooperative intelligent system is designed, which includes the submodels of roadside and vehicle side.Based on the characteristics of various road types, the corresponding scheme of intelligent transportation infrastructure is designed.Te logic and principles of intelligent confguration and cost transfer from the vehicle side to the roadside under the synergy efect are qualitatively and quantitatively described, respectively.When combined with characteristic parameters of feet and road in a specifc city, the model can output the comprehensive costs of vehicle side and roadside.Tis study takes Beijing as an example.Tese costs include the deployment and operational costs at roadside, as well as the acquisition and usage costs of intelligent confgurations on the vehicle side.Te corresponding results can provide some cost references and directional recommendations for city managers to help them make decisions on the deployment of intelligent infrastructure.Te cost evaluation model of vehicle-road cooperative intelligent system designed in this study is also applicable to other cities.

Cost Evaluation Model for Vehicle-Road
Collaborative Intelligent System Te cloud platform, roadside infrastructure, ICVs, communication network, and resource platform are the fve core components of the collaborative intelligent system.Figure 1 depicts the architecture of the key connecting pieces between each component.In this study, intelligent infrastructure is characterized as a combination of the cloud platform, roadside infrastructure, communication network, and resource platform.Te communication network connects various nodes of the system.Te wide interconnection and high-performance transmission can be realized through the application of uniform standardization mechanisms.ICVs, roadside infrastructure, and digital resource platform are connected to the cloud platform.Te cloud platform uniformly utilizes the perception results of ICVs and roadside infrastructure as well as data from resource platforms.Trough real-time hierarchical perception fusion on the cloud of each level, it provides real-time data required for the operation of a wide range of collaborative applications.Te overall performance of vehicle movement and trafc operation can be optimized by constructing multiobjective and multitask planning technology for collaborative applications.It can also optimize the allocation of system computing resources to guarantee the safety and performance of collaborative applications serving vehicle and trafc optimization.Te ICVs can also directly connect to the roadside infrastructure and other ICVs, which contributes to sensing, planning, and decision-making at vehicle side.Te subsequent cost evaluation methodology will be theoretically based on this system architecture.
2.1.1.Cloud Platform.Edge, regional, and central clouds make up the cloud platform.Te regional and central clouds typically serve the city or province and mainly run quasireal-time collaborative applications, such as overall trafc guidance at the regional/city level.Te edge cloud typically serves the street or district and primarily runs real-time collaborative applications, such as rapid multiple perception fusion.

Roadside
Infrastructure.It consists of diferent kinds of roadside sensors, mobile edge computers (MECs), roadside units (RSUs), intelligent signal lights, and other trafc-control systems.

Communication Network.
A communication network connects various nodes of trafc participants, roadside infrastructure, and cloud platforms by combining wired and wireless communication technologies on the basis of a standardized communication mechanism.In order to realize real-time data collection and transmission as well as instruction update and transmission (such as real-time acquisition of HD map information to assist vehicle positioning), the reasonably deployed 5G base stations will provide low latency and large bandwidth communication capability.At the same time, the direct wireless communication between OBU and RSU can function as an efcient addition to 5G communication, resolving the issue of a blocked or unstable 5G signal.

ICVs.
To improve vehicle intelligence and realize optimized driving performance, ICVs connect to cloud platforms, roadside infrastructure, and other ICVs, share vehicle-side data, and use the output of intelligent infrastructure for assisted driving or autonomous driving.

Model Framework.
In this section, the cost evaluation model of the vehicle-road cooperative intelligent system is built, as shown in Figure 2. Te characteristic parameters of various road types, the population, and penetration of each type of intelligent vehicle, the cost and energy consumption of intelligent confgurations, and intensity of usage are the inputs.With the full deployment of the intelligent infrastructure, the intelligent confguration will transfer from the vehicle side to the roadside due to the synergy efect.In the model, an optimized vehicle-road cooperative system scheme is designed, which includes the intelligent schemes of vehicle side and roadside.Te fnal output is the deployment and operational costs of the intelligent transportation infrastructure and the acquisition and usage costs of feet intellectualization.Te specifc cost evaluation submodels of intelligent transportation infrastructure and feet intellectualization are described in Sections 2.2.1 and 2.2.2, respectively.

Te Cost Evaluation Submodel of Intelligent
Infrastructure.Te cost evaluation submodel for intelligent infrastructure has been designed, as shown in Figure 3. Te model considers urban expressways, urban main roads, urban secondary roads, motorways, class-1 highways, class-2 highways, and class-3 highways, a total of seven diferent road types.Additionally, urban large intersections, urban small intersections, suburban large intersections, and suburban small intersections were also identifed.Diferent road types and intersection types possess distinct characteristics and demands for roadside intelligent confgurations.For instance, highways are closed roads with relatively simple road scenarios, in contrast to urban main roads or secondary roads, which are open roads that present more complex scenarios.Corresponding roadside intelligence schemes have been developed for each road type and intersection type, taking into account the design service capacity of the roads as well as the complexity of road scenarios.Tese schemes encompass specifc sensing, communication, and computing schemes, as detailed in Tables 1 and 2. Te roadside intelligent confgurations involved are 5G macro stations, 5G micro stations, RSUs, vision sensors, millimeter-wave radar, LiDAR, edge cloud servers, central cloud servers, intelligent signal machines, timing servers, and auxiliary equipment such as brackets and power distribution equipment.
(1) Data.In Table 3, we present the cost, power, and performance parameters for roadside intelligent confgurations.Our group has established the cost database of these facilities [27,30].Currently available or upcoming transportation infrastructure with satisfying performance is selected.Detailed functional parameters and associated costs can be sourced from manufacturers' websites, product manuals, or research reports.
In developing the deployment logic for intelligent devices within the cost evaluation submodel of intelligent transportation infrastructure, factors such as average trafc fow and the complexity of road scenes were taken into consideration.We also reviewed several related studies [20][21][22][23][24][25][26][27] to determine the deployment density of roadside intelligent confgurations for each road type and each intersection type in the city, as displayed in Tables 1 and 2, separately.Detailed explanations of the corresponding intelligent infrastructure solutions can be found in the Appendix.
(2) Communication Infrastructure.Te construction level of the current communication infrastructure varies between urban and suburban roads.In Beijing, urban roads have  largely achieved 5G signal coverage.To enhance the 5G network's coverage and accessibility, 5G micro base stations need to be paired in urban expressways, urban main roads, and urban secondary roads.On the other hand, motorways and highways, which serve as transportation arteries connecting the city's core regions, are often situated in areas with fewer economic activities and lower population densities.For these roads, it is essential to deploy both 5G macro and micro base stations to guarantee the 5G network's coverage and accessibility.Te intersection deployment schemes adhere to this same design rationale.
(3) Roadside Perception.Te solution primarily considers the average trafc fow and the complexity of scenes across various road and intersection types.Urban main and secondary roads have complex environments due to heavy trafc and a mix of participants, especially during peak hours.Given the sensing range limitations of roadside sensors, it is crucial to integrate a denser array of cameras, millimeter-wave radars, and Lidars to minimize blind spots.On the other hand, motorways and urban expressways are more controlled environments with simpler scenarios and singular trafc participants.Tis allows for a reduced deployment density of roadside sensing devices.Te perception scheme at intersections largely depends on their size, with larger intersections equipped with more sensing equipment to ensure complete coverage.
(4) Roadside Computing Platform.Te approach for roadside computing is chiefy infuenced by the average trafc volume.Urban expressways, main roads, and motorways experience higher trafc volumes and, therefore, demand more robust roadside computation.In these areas, the edge computing unit is tasked with serving a larger number of ICVs at each time step.Te deployment logic for roadside computing at diferent intersections adheres to this same design rationale.
(5) Te Deployment Cost of Intelligent Infrastructure.Te deployment cost of roadside intelligent solutions corresponding to each road type is calculated, using the cost of roadside intelligent confguration, as shown in the following equation: where Cost r,deploy is the deployment cost of intelligent infrastructure for per mile road type r, r � 1, 2, 3, ...11 represents road types and intersections types mentioned above, cost i is the cost of roadside intelligent confguration i, i � 1, 2, 3, ...10 represents roadside intelligent confguration introduced before, and Density r,i is the deployment density of roadside intelligent confguration i for per mile road type r.
(6) Te Operational Cost of Intelligent Infrastructure.Te operational cost of roadside intelligent scheme corresponding to each road type is calculated, which consist of energy consumption cost and maintenance cost, as shown in the following equation: where Cost r,operating is the operational cost of intelligent transportation infrastructure for one kilometer of road type r; ecr i , β i , and maintain i are the power, usage characteristics,    Journal of Advanced Transportation and maintenance costs of roadside intelligent confguration i, respectively.Te usage characteristics β i of all roadside intelligent confgurations i take the value of 1, which means all facilities are in an open status throughout the day, to satisfy the security requirements of high-level autonomous driving.In the future, as technology advances, the equipment can adjust power in real-time based on changes in trafc fow to minimize energy consumption.And the usage characteristics β i can be further optimized in the follow-up study.Considering the service life of facilities is around 10-15 years [30], maintain i takes the value of 0.1 of the deployment cost cost i .Te total cost of the roadside intelligence solution matching the road type r consists of the deployment and the operational cost as shown in equation (3).It serves as the basis for the subsequent calculation of the city's total cost to implement the intelligent upgrades of its transportation infrastructure.

Journal of Advanced Transportation
Cost r � Cost r,deploy + Cost r,operating , ( where Cost r is the total cost of intelligent infrastructure for one kilometer of road type r. (7) Te Acquisition and Renewal Cost of HD Map.HD maps serve as a crucial infrastructure for deploying intelligent vehicles, aiding in vehicle localization and path planning.Te methods for collecting HD map data fall into two categories: professional equipment collection and crowdsourcing.Professional equipment collection involves using mapping vehicles ftted with specialized tools, including LIDAR, panoramic cameras, and high-precision inertial navigation systems.While this method ofers superior map accuracy and reliability, it tends to be more costly and less efcient.On the other hand, crowdsourcing leverages afordable sensors in the existing intelligent vehicle feet.
Tese vehicles gather a vast amount of road data during their daily routes and incorporate it into HD maps.Tis method boasts real-time data updates, a wealth of data sources, and cost-efectiveness.However, its accuracy and reliability might not meet the standards necessary for advanced autonomous driving.Currently, HD map construction primarily uses a blend of professional equipment acquisition and a crowdsourcing renewal approach.In scenarios involving vehicle intelligence, HD maps can only be refreshed through "crowdsourcing collection" carried out by the intelligent vehicle feet.
Leveraging vehicle-road cooperative perception, the crowdsourcing renewal process for HD maps achieves both high accuracy and reliability.Tis is achieved by pairing it with fxed-point observations made by roadside perception equipment, allowing for updates within minutes.Tis approach not only shrinks the size of the required feet but also addresses prevalent challenges in the HD map industry, such as maintaining map freshness and high costs.By referring to relevant reports and literature, combined with expert interviews, the integration of intelligent infrastructure will cut the renewal costs of HD maps down to one-seventh in vehicle intelligence scenario [31][32][33][34].Table 4 presents the unit costs associated with HD map and their corresponding data sources.
Te costs of HD map are calculated, which consist of acquisition cost and renewal cost, as shown in equations ( 4)- (6).Te acquisition cost for an HD map is infuenced by its per-unit price and the length of roads, whether in a city or broader country context.Additionally, the renewal cost for an HD map is contingent upon its renewal frequency.Journal of Advanced Transportation Owing to the absence of specifc statistical data for diferent types of intersections in Beijing, we estimate the ratio of urban large intersections, urban small intersections, suburban large intersections, and suburban small intersections to be 10 : 7 :14 : 81.Tis estimation is based on the quantitative distribution of diverse road types.By integrating the characteristic data of Beijing's roads, the overall costs of intelligent transportation infrastructure can be determined using the cost-evaluation submodel designed for intelligent transportation infrastructure.

Te Cost Evaluation Submodel of Fleet
Intellectualization.In this section, the cost evaluation submodel of feet intellectualization is constructed, as shown in Figure 4.By identifying the vehicle-side intelligent scheme under the scenarios of vehicle intelligence and collaborative intelligence, the corresponding acquisition cost and usage cost are calculated.Factoring in the population and adoption rate of each type of intelligent vehicle in Beijing, the total cost of feet intellectualization under these two scenarios is derived.
(1) Te Substitution of Vehicle Perception under the Synergy Efect [36,37].High-dimensional perception from the roadside can provide a more encompassing view of the road, surpassing the perspective of individual vehicles, thus aiding advanced autonomous driving.It can also better satisfy the perception requirements for vehicle in terms of accuracy, time delay, and reliability.Furthermore, such highdimensional perception from the roadside can take the place of redundant vehicle perception and serve as a backup to basic perception.Consequently, this results in a signifcant reduction in both acquisition and operational costs associated with the vehicle-side perception scheme.
(2) Te Substitution of Vehicle Computing under the Synergy Efect [38,39].Roadside computing devices ofer easy deployment and scheduling capabilities.Leveraging the abundant computing resources available roadside allows for efective sharing of computing power among vehicles, ensuring equitable distribution and optimal balancing of computational power across the system.On the other hand, in-vehicle computing devices have their limitations.Tey cannot be scaled up when faced with increased computational needs and cannot be downsized or reallocated when idle.Presently, many vehicles come pre-equipped with a large computing platform to accommodate potential future functionalities and software updates.Tis often leads to an unnecessary wastage of resources and consequently increased costs for the user.
(3) Data.Table 6 details the power and cost associated with intelligence confgurations on the vehicle side.It outlines the primary, intermediate, and advanced intelligence schemes under the vehicle intelligence scenario as well as the advanced intelligence scheme for the collaborative intelligence scenario.
Te current intelligence schemes of vehicles demonstrate greater diferentiation due to the varying technical routes and capability levels of various companies.With reference to the primary, intermediate, and advanced intelligence scheme commonly adopted in the industry, as well as several published works [40,41], typical vehicle intelligence schemes are presented, as shown in Table 6.
In the vehicle intelligence scenario, a higher level of intelligence in AVs necessitates the inclusion of more sensors to enhance sensing capability, additional computing units to increase computing power, and the use of dependable, low-latency actuators such as steering and braking.In the collaborative intelligence scenario, ICVs can interact with the intelligent transportation infrastructure in real-time via high-performance communication modules.Tis interaction allows them to access more comprehensive environmental information and reduces the demand for extensive perception and computing confgurations within the vehicle itself.Simultaneously, roadside fusion perception and positioning information are correlated with the dynamic and static features of the vehicle.Tis correlation enables the acquisition of real-time and high-precision location information for vehicles, reducing the need for automatic positioning equipment on the vehicle side.On roads with fully deployed intelligent transportation infrastructure, ICVs can achieve advanced autonomous driving [34,42], as previously described.Furthermore, due to their basic perception and computing confgurations, ICVs can still achieve primary autonomous driving, even on roads without intelligent infrastructure coverage [34].
Te expansion from intelligent vehicle schemes to feet expenses at the city level depends on the characteristic data of feet in Beijing, as illustrated in Table 7. Te average use intensity of the feet is infuenced and constrained by many factors, including road mileage, road capacity, trafc restriction policies, convenience of other transportation modes, and epidemics.Changes in average utilization intensity in the future are not considered in our study.Te average annual mileage and driving time are based on data in 2019 to exclude the impact of the epidemic.

(4) Te Vehicle Sales and Market Penetration Data in Beijing.
Te quantity of intelligent vehicles in the future are calculated using the sales forecast data in Beijing and the survival law of vehicle, as shown in equation (7).Te forecasted auto market sales in Beijing are primarily based on predictions by industry experts.Te forecasted penetration rates of intelligent vehicles in Beijing are referenced from the "Intelligent Connected Vehicle Technology Roadmap 2.0" [45].
where VS v,y are the vehicle stocks of v type vehicles in the year y; v � 0, 1, 2, 3, 4 represent traditional vehicles, primary AVs, intermediate AVs, and advanced AVs and ICVs; l is the lifespan of the vehicle; Sales j are the vehicle sales of Beijing in the year j; PR v,y is the penetration rate of v type of vehicle in the vehicle sales in the year y; and SR y−j is the survival rate of the vehicle in the (y − j) th year (%).
In order to compare the two scenarios under the same baseline, the forecasted penetration rates of intelligent vehicles in Beijing, whether it is under the vehicle intelligence scenario or the collaborative intelligence scenario, are based on the "Intelligent Connected Vehicle Technology Roadmap 2.0."It is assumed that users in both scenarios have the same willingness to purchase intelligent vehicles and will use them with the same intensity.
(5) Te Cumulative Acquisition Cost of Fleet Intellectualization in Beijing.Te annual acquisition costs of feet intellectualization are calculated using equation (8), and the cumulative acquisition cost in Beijing from 2023 to 2050 is determined using equation (9).Tese calculations take into account the anticipated cost reductions associated with various intelligent confgurations in the future.
Cost fleet,acquisition � where α v is the intelligent scheme vector of the v type vehicle, β s,k is the cost reduction ratio of intelligent confguration s in the year k, and cost v,j is the acquisition cost of intelligent confgurations for the v type vehicle in the year j.
Intelligent confgurations are at various stages of technological development and scale in diferent periods, resulting in diferent cost reduction ratios and unit costs.Te average market price of these confgurations in 2023 is used as the benchmark.Table 8 displays the cost reduction ratios for various intelligent confgurations across diferent time periods.VS v,y × ecr v + maintain v , (10) where ecr v is the annual energy consumption cost of the v type vehicle and maintain v is the annual maintenance cost of the v type vehicle.

Scenario Design.
In our study, Beijing is chosen as the case study, and the characteristic parameters of motor vehicle and various road types in Beijing are collected.Two scenarios, vehicle intelligence and collaborative intelligence, are selected to assess the comprehensive cost of implementing advanced autonomous driving under each scenario.

Annual Cost under the Vehicle Intelligence Scenario 2023-2050.
In the vehicle intelligence scenario, intelligent vehicles rely solely on their own capabilities for sensing, decision-making, and execution, without support from intelligent transportation infrastructure.Figure 5 illustrates the forecasted annual incremental costs of feet intellectualization in Beijing from 2023 to 2050.

Acquisition Cost.
As depicted in Figure 5(a), the "climbing period" of incremental acquisition costs in feet intellectualization from 2025 to 2035 corresponds to the growth of the penetration rate of intelligent vehicles in Beijing.It is projected that in 2035, the annual incremental acquisition cost of feet intellectualization will reach ¥17.35 billion.Among these, intermediate AVs will account for 56.24% of the total acquisition cost, reaching its peak at ¥21.64 billion in 2045.Te increase in the sales of intelligent vehicles will be more than ofset by the reduced costs of intelligence due to advancements in key components and mass production.By 2050, it is anticipated that the annual incremental acquisition cost of feet intellectualization will decrease to ¥20.45 billion.Advanced AVs will become the market mainstream, constituting 93.22% of the total acquisition cost.12 Journal of Advanced Transportation and additional computing power on the vehicle side will lead to higher energy consumption and maintenance costs.

Usage Cost. As depicted in
Consequently, there will be a signifcant increase in the incremental usage costs of feet intellectualization.It is projected that, in 2035 and 2050, the annual usage cost of feet intellectualization will amount to ¥6.65 billion and ¥18.58 billion, respectively.Interestingly, it is worth noting that the usage cost of feet intellectualization in 2050 is roughly equivalent to the acquisition cost in the same year.Overall, the increasing trend in comprehensive costs is evident under the vehicle intelligence scenario, as illustrated in Figure 5(c).It is projected that the total cost of feet intellectualization will be ¥24.00 billion and ¥30.03 billion, respectively, with the usage cost of feet intellectualization gradually representing a larger share of the total cost.Beijing is expected to pay a cumulative cost of ¥703.00 billion from 2023 to 2050 for feet intellectualization.

Annual Cost under the Collaborative Intelligence Scenario 2023-2050.
In the collaborative intelligence scenario, it is assumed that Beijing will complete the deployment of intelligent infrastructure from 2023 to 2024, thereby meeting the technical requirements for a vehicle-road collaborative intelligence system.Te intelligent infrastructure will be put into use from 2025 onwards.ICVs, which synergize with intelligent transportation infrastructure, are expected to dominate the sales of intelligent vehicles due to their superior cost/performance ratio.
Under this scenario, society is responsible for covering the energy consumption and maintenance costs of intelligent transportation infrastructure, in addition to the acquisition and usage costs of feet intellectualization. Figure 6 illustrates the annual incremental costs under the collaborative intelligence scenario from 2023 to 2050.

Acquisition Cost.
As shown in Figure 6(a), the "climbing period" of incremental acquisition costs from 2023 to 2031 is also associated with the growth of the penetration rate of intelligent vehicles in Beijing, albeit at a much lower rate than the vehicle intelligence scenario.It is anticipated that the annual incremental acquisition cost of feet intellectualization will reach its peak at ¥10.85 billion in 2032, which represents approximately 64.64% of the cost incurred under the vehicle intelligence scenario for the same duration.Subsequently, the declining cost of feet intellectualization can be attributed to technological advancements in key components and mass production.By 2050, it is expected that the annual incremental acquisition cost of feet intellectualization will decrease to ¥8.55 billion, accounting for approximately 41.79% of the cost in the vehicle intelligence scenario over the same period.In the collaborative intelligence scenario, the deployment of intelligent transportation infrastructure in Beijing is scheduled to be completed between 2023 and 2024.However, due to the low penetration rate of ICVs in the early stages, society will continue to bear the costs of energy consumption and maintenance.During this period, the cost-efectiveness of the intelligent transportation infrastructure is relatively low.Starting in 2028, the collaborative intelligent system will have a lower total annual cost than the vehicle intelligence scenario, as shown in Figure 6(c).Te cost-efectiveness of collaborative intelligence will increase as the penetration of ICVs continues to rise.A detailed comparison of the cumulative costs under the two scenarios and the cost breakdown for intelligent infrastructure is provided in Section 3.3.14 Journal of Advanced Transportation when compared to the vehicle intelligence scenario.Tese savings are primarily shouldered by users.Regarding the cost of HD maps, the feet scale for "crowdsourced maintenance" can be efciently reduced through the vehicle-road cooperative awareness maintenance strategy.Tis strategy is expected to lower the overall renewal cost of HD maps by approximately ¥20.08 billion.Overall, the collaborative intelligence scheme is anticipated to reduce the total social cost by ¥243.18 billion from 2023 to 2050, factoring in the deployment, energy consumption, and maintenance expenses of intelligent transportation infrastructure in Beijing.Te deployment and energy consumption and maintenance costs of intelligent infrastructure across various road types in Beijing are shown in Figure 8. Te total cost of deploying intelligent transportation infrastructure in Beijing amounts to ¥17.89 billion.Following the deployment, it becomes essential to regularly maintain and update the roadside sensing, communication, computing, and other intelligent equipment to ensure their proper operation.Tese operational activities also entail energy consumption costs.It is noteworthy that the cumulative cost of energy consumption and maintenance, which constitutes a substantial portion of the overall cost, is projected to reach ¥78.52 billion from 2025 to 2050. Figure 8 also provides a breakdown of the components contributing to the accumulated operational cost of intelligent infrastructure.Te energy consumption cost associated with communication equipment is expected to account for approximately 68.2% of the total roadside energy consumption cost (¥32.00billion) between 2025 and 2050.Tis is primarily due to the high energy consumption of 5G base stations deployed for road transportation.By around 2050, the energy consumption costs for roadside perception and roadside computing will have accumulated to ¥4.63 billion and ¥5.55 billion, respectively.

Te Comparison of Cumulative Costs in
Furthermore, the deployment and operation of intelligent transportation infrastructure will introduce more industrial participants, benefting a broader spectrum of stakeholders.Te 5G communication network expands the scope of application scenarios.With the increasing penetration rate of ICVs, communication carriers stand to gain higher revenues from their users for communication services.Building upon the foundation of vehicle-road cooperative perception, HD map service providers can ofer more accurate and reliable updates to HD maps at a reduced cost.Concurrently, communication carriers and HD map service providers will actively engage in the operation of roadside intelligent infrastructure, including 5G base stations and cloud servers.Tey will share the responsibilities of maintaining and covering the energy consumption costs of roadside equipment in use.

Sensitivity Analysis.
As evident from the results depicted in Figure 7, both the acquisition cost of feet intellectualization and the deployment cost of intelligent infrastructure constitute a signifcant portion of the cumulative cost.Consequently, we proceed to analyze the sensitivity of costs associated with various vehicle-side and roadside facilities.
Figure 9 illustrates the efects of ± 20% changes in the costs of various facilities, compared to reference values, on the cumulative costs in Beijing from 2023 to 2050 under the vehicle intelligence scenario.A 20% change in the cost of various vehicle-side facilities would result in impacts ranging from 0.02% to 1.27% on the cumulative cost.Among these facilities, Lidar, high-precision localization, and the central computation platform have the most signifcant impacts on the overall result.Tis is primarily because advanced autonomous driving in the vehicle intelligence scenario necessitates vehicle-side Lidar, localization, and computation with higher performance, leading to higher acquisition and maintenance costs.Journal of Advanced Transportation ranging from 0.01% to 0.84% on the cumulative cost.Notably, the central computation platform and in-vehicle communication module have the most signifcant impact on the overall result.Similarly, a 20% change in the cost of various roadside facilities would result in impacts ranging from 0.0017% to 0.89% on the cumulative cost.Here, the 5G macro site and RSU have the most substantial impact on the overall result.Journal of Advanced Transportation

Conclusions and Policy Suggestions
In this study, we constructed a cost evaluation model for the vehicle-road cooperative intelligent system based on its architecture and optimized scheme design to ensure technical feasibility.We chose Beijing as our case study to assess the upgrade cost of intelligent transportation infrastructure and to compare and analyze the costs of feet intellectualization and the overall cost under two scenarios: collaborative intelligence and vehicle intelligence.
Our fndings indicate that the deployment of intelligent transportation infrastructure in Beijing serves multiple purposes.It not only alleviates the current bottlenecks in the development of advanced autonomous driving technology but also signifcantly mitigates the increase in overall costs associated with the development of advanced intelligent vehicles.Tis, in turn, reduces the acquisition and usage costs for users.Lower costs for intelligent vehicles increase people's willingness to invest in them.Additionally, enhanced intelligent functions further boost people's purchase willingness.Ultimately, the willingness of individuals to acquire such vehicles determines the market penetration rate of intelligent vehicles.Furthermore, the collaborative intelligence solution, by transferring the intelligent confguration from the vehicle side to the roadside, achieves a greater reuse efect, benefting a larger number of ICVs.Tis approach results in greater advantages in terms of total social expenses.Te deployment cost of intelligent transportation infrastructure in Beijing, totaling ¥17.89 billion, accounts for approximately 3.8% of the cumulative cost from prehensive social cost by ¥243.18 billion, demonstrating a high degree of cost-efectiveness.To put this in perspective, the cost of deploying intelligent transportation infrastructure estimated in this study is only about 2.4% of the total investment in transportation infrastructure projects in Beijing for the period 2016-2020, which amounted to approximately ¥750.5 billion [46].Furthermore, the operation of intelligent infrastructure can be shared by communication carriers, HD map service providers, and other industry participants, making it a realistic possibility.
Te deployment of intelligent transportation infrastructure can not only support autonomous driving but also contribute to intelligent transportation, urban management, and other aspects of the development of smart cities. Greater benefts to society will be achieved in contrast to the vehicle intelligence scenario, in terms of trafc safety, efciency, energy conservation, and environmental protection.As for road safety, vehicle intelligence can prevent 60% of trafc accidents, while V2X can cut trafc accidents by 81%, according to the U.S. Department of Transportation's review of 6 million vehicle accidents [47].In term of trafc efciency, telematics technology can increase road efciency by 10%, lowering the costs associated with congestion, including time costs, carbon emission costs, and environmental management costs [48].As a result, collaborative intelligence is a better solution from both the perspective of comprehensive social costs and public benefts.
Te deployment of intelligent transportation infrastructure ofers benefts that extend beyond supporting autonomous driving.It contributes signifcantly to various aspects of smart city development, including intelligent transportation and urban management.Te collaborative intelligent system provides greater societal benefts in terms of trafc safety, efciency, energy conservation, and environmental protection.In the context of road safety, vehicle intelligence can prevent 60% of trafc accidents, while V2X communication can reduce trafc accidents by 81%, as reported by the U.S. Department of Transportation's analysis of 6 million vehicle accidents [47].In terms of trafc efciency, telematics technology has the potential to increase road efciency by 10%, thereby reducing the costs associated with congestion.Tese costs encompass various factors such as time costs, carbon emission costs, and environmental management costs [48].Ultimately, collaborative intelligence emerges as the superior solution from both the perspective of comprehensive social costs and the broader public benefts it ofers.

Government Initiatives.
Aligning with national top-level planning, the government should prioritize the development of standardized systems, accelerate the deployment of C-V2X network environments, promote demonstrations of vehicle-road coordination applications, and prepare for the standardization and large-scale deployment of intelligent transportation infrastructure.Government funding for road intelligence construction should be a shared efort with enterprises, ultimately reducing the total cost of ownership for users.Tis support is critical for the industry to address technical challenges and facilitate commercialization.

Industry-Level Engagement.
Industry participants should each leverage their strengths in the feld of vehicleroad collaboration based on their core competencies.Internet technology companies, with their extensive experience in large data and collaborative software algorithms for autonomous driving, can play a vital role.Upstream intelligent component frms should drive product implementation through technological innovation, developing new sensing, location, and computing devices tailored to both vehicles and roadside infrastructure.Communication carriers should collaborate with the government to construct urban network environments with advanced networking technology, expanding the application scenarios and business models of 5G-V2X.Meanwhile, vehicle manufacturers should actively pursue multiparty cooperation to integrate technology, build vehicle networking systems, and ofer the market ICVs with cost-efective solutions and superior functional performance.
In summary, it is imperative for all stakeholders to collaborate and share resources, leveraging their existing technical capabilities.Active participation in the construction of the industry ecosystem is essential, along with the acceleration of a comprehensive industrial system.Trough these eforts, we can ultimately achieve advanced automation that delivers substantial social benefts while maintaining a lower overall cost to society.
Tis study has several limitations.While collaborative intelligence has emerged as the preferred option for advancing autonomous driving, it has also raised a set of intriguing questions that warrant further investigation.In our study, both the vehicle intelligence and collaborative intelligence scenarios were based on the forecasted penetration rates of autonomous vehicles outlined in the "Intelligent Connected Vehicle Technology Roadmap 2.0."However, we did not account for the potential increase in penetration rates under the collaborative intelligence scenario, attributed to the enhanced value of vehicles stemming from lower costs associated with high-level autonomous driving functions.It is crucial to recognize that people's willingness to purchase intelligent vehicles is infuenced by a multitude of factors, including additional costs, socio-economic conditions, technological maturity, consumer perceptions of utility, and more.Ultimately, the market penetration rate of intelligent vehicles hinges on these considerations.Even without considering the variations in features and performance ofered by intelligent vehicles, establishing clear and articulable correlations between the acquisition and usage costs of vehicle-side intelligent confgurations and people's willingness to adopt them remains challenging.In future research, we intend to explore methods to quantify the correlation between vehicle intellectualization costs and the penetration rates of intelligent vehicles.Furthermore, this study exclusively delved into the comprehensive costs of implementing collaborative intelligence in Beijing to support advanced Journal of Advanced Transportation autonomous driving.However, as previously mentioned, collaborative intelligent solutions have the potential to generate substantial public benefts in areas such as safety, trafc management, and environmental impact.Quantifying these benefts will be a valuable aspect of our future research eforts.

Appendix
Based on the scene characteristics of diferent road types, considering the coverage requirements of high-level intelligent driving on roadside sensing, communication and computing, combining the capabilities, and performance of diferent intelligent devices, the corresponding intelligent infrastructure solutions are determined, which are explained in detail in this section.Tey are also verifed through the related literature and industrial practice.

A. The Deployment Scheme of Intelligent Devices for Motorway
Motorway, as a typical closed road, generally consists of 6-8 lanes in both directions and the width will reach 40 meters or more considering two emergency lanes in both directions.Te opposite lanes are often divided by a greenbelt as the separation zone.Te design service capacity of motorway is relatively high.But the scene is relatively simple, with single trafc participant.
It is necessary to deploy the pole frame and other auxiliary equipment at the greenbelt, as well as the center of the highway, carrying two sets of intelligent devices for both directions to obtain the best observation view and the largest efective coverage area.Tis scheme for motorway can reduce the deployment cost and solve the problem of sensing blind areas caused by the greenbelt blockage, as shown in Figure 11.
Te scene characteristics of urban expressway are similar to those of motorway.Combining the corresponding design service capacity and scene complexity factors, the deployment scheme and density of sensing, communication, and computing devices on urban expressway are determined accordingly.

B. The Deployment Scheme of Intelligent Devices for Urban Main Road
Urban main road, as a typical open road, generally consists of 6-10 lanes in both directions.Te width will reach more than 40 meters, considering two nonmotorized vehicle lanes in both directions.Te trafc scene is complex, including motor vehicles, nonmotorized vehicles, pedestrians, and other trafc participants.Meanwhile, the urban main road is with heavy trafc fow and serious obscuration between various trafc participants.
It is demanded to deploy intelligent devices alternately on both sides of the road, through multiview perception to solve blind areas problem caused by obscuration.More intensive edge computing server will be paired to better cope with the complex scenes of urban main road.
Simultaneously, more roadside units and pole frame and other auxiliary equipment are also required in the roadside intelligence schemes of urban main road, as shown in Figure 12.
Te scene characteristics of urban secondary road, class-1 highway, class-2 highway, and class-3 highway, are similar to those of urban main road.Combining the corresponding design service capacity and scene complexity factors, the deployment scheme and density of sensing, communication, and computing devices on urban secondary road, class-1 highway, class-2 highway, and class-3 highway are determined accordingly.

C. The Deployment Scheme of Intelligent Devices for Various Intersection Types
Te roadside intelligent equipment at urban large intersections is arranged in each of the four corners and needs to be matched with larger scale sensing, communication, and computing equipment to achieve intersection coverage.On the basis of ensuring regional coverage, the equipment at urban small intersections is arranged diagonally to reduce the cost and difculty of deployment, which is shown in Figure 13.

Figure 1 :
Figure 1: Architecture of vehicle-road collaborative intelligent system.

Figure 2 :
Figure 2: Cost evaluation model of vehicle-road cooperative intelligent system.
reconstruction of existing traffic control equipment (traffic signal, traffic sign, etc.)

Figure 3 :
Figure 3: Cost evaluation submodel of intelligent transportation infrastructure.
Figure 5(b), particularly after 2025, as the population of intermediate AVs and advanced AVs increases, the inclusion of more sensing devices

Figure 6 (
b), higher incremental usage costs are expected to occur from 2023 to 2045, aligning with the increasing number of ICVs.Beyond 2045, the annual usage cost of feet intellectualization stabilizes at approximately ¥7.30 billion.Tis represents only about 40% of the usage cost under the vehicle intelligence scenario during the same period.

Figure 6 :
Figure 6: Annual incremental cost of collaborative intelligence system.(a) Acquisition cost-feet intellectualization.(b) Usage cost-feet intellectualization.(c) Total cost of collaborative intelligent system.

Figure 10 :Figure 11 :
Figure 10: Uncertainty analysis for ± 20% changes from reference values for cost of various facilities on cumulative cost from 2023 to 2050 under collaborative intelligence scenario.

Figure 12 :Figure 13 :
Figure 12: Te roadside intelligence schemes of urban main road.

Table 1 :
Roadside intelligence schemes of various road types.
Cost map � Cost map,acquisition + Cost map,renewal , Cost map,renewal � UnitCost map,renewal * road mileage * Renewal Cycle.(6) (8) Te Characteristic Data of Various Roads in Beijing.Te characteristic data of various roads in Beijing is crucial for transitioning from individual road intelligent deployment schemes to the comprehensive cost of intelligent transportation infrastructure at the city level.Several factors infuence the size of the road network.Beyond city

Table 2 :
Roadside intelligence schemes of various intersection types.

Table 3 :
Performance parameters and costs of roadside intelligent confgurations.

Table 4 :
Te acquisition and renewal cost of HD map.

Table 5 :
Te characteristic data of various road types in Beijing.

Table 7 :
Te characteristic data of feet in Beijing.

Table 8 :
Te cost reduction ratio of various intelligent confgurations.