Benefits and Costs of Autonomous Trucks and Cars

Autonomous vehicles are currently developed, and are expected to be introduced gradually. Society needs a basis for decisions regarding market interventions. This study identifies, quantifies and values the benefits and costs of autonomous trucks and cars considering generalized costs, external effects and social marginal cost pricing to consumers with Swedish data. The results show that the greatest benefits are saved driver costs for trucks and decreased travel time costs for car drivers. In the example calculations, capital costs may increase by 22 percent for cars and 36 percent for trucks for benefits to exceed costs in 2025. Subsidies are not needed since the producers and consumers get the major benefits and pay the costs.


Introduction
The vehicle industry is developing new technologies for trucks and cars to be autonomous, also referred to as automated or self-driving vehicles that are expected to be introduced gradually and gain market shares. When commercial success is up to the manufacturers, the benefits and costs to society as a whole are crucial not only for the technology's success but also for the implementation of the right policies and public investments. Questions include whether policies should be proactive, promote development with rapid adjustment of regulations, and give subsidies if the technology has net benefits to society, or whether institutions should evolve gradually in response to innovation within the industry.
We study the relative importance of the different identified effects when autonomous vehicles (AVs) are introduced, in particular the dominance of saved Journal of Transportation Technologies cisions whether the public sector should consider producing on, financing or regulating the market.
The generalized cost (GC) of transportation, used as a key concept in transport economics, is a central term for the identification, quantification and valuation of effects for autonomous trucks and cars. Generalized cost refers to all monetary and non-monetary costs for the transport service user. The benefits of AVs consist of lower generalized costs for the vehicle kilometres (vkm) driven by autonomous trucks and cars instead of MDVs, and of more transportation.
Some of the latter will be newly generated, and some will be transferred from other modes of transport. The extent of the effects depends on elasticities. We will use the term social generalized cost (SGC) for generalized cost when all external effects are internalised in the price to the consumer. Figure 1 illustrates the volume of road transportation with AVs as a function of social generalized cost, and the resulting net benefits if SGC is lower with AVs compared to MDVs.
Area A represents the net benefits of lower social marginal costs for existing trips that are transferred from MDVs to AVs. Area B represents the net benefits of new trips on the road.
The benefits of the change in social generalized cost for trips transferred from MDVs on the road to AVs is calculated as If social marginal cost pricing prevails, i.e. all used resources are considered when deciding to drive another vkm; new vehicle kilometres give a net social benefit to society, as the additional value is higher than the cost. In reality, however, such optimal pricing may not be the case. If the studied (primary) market for driving with trucks and cars or interrelated (secondary) markets do not have social marginal cost pricing and the consumers do not accurately consider all benefits and costs to society of the new vehicle kilometres, adjustment posts need to be added (Boardman et al. [4]).
The net benefits of the change in generalized cost for new vkm on the road is calculated as For passenger traffic by private car, the producer and the consumer of the vehicle kilometres is the same. It is then sufficient to study whether negative and positive external effects are internalized by Pigouvian taxes and subsidies (Pigou [5]), so that the decisions are made based on the social marginal cost. Thus for cars, if needed, an adjustment post is added as For freight transport by truck, producers and consumers may be different actors. Therefore, the competitive condition and pricing strategies must also be studied. On the one hand, it may be that freight service providers pay a price lower than the social marginal cost of using infrastructure if negative external effects are not internalized by Pigouvian taxes. On the other hand, limited competition, or a cost structure of decreasing average costs, may lead to consumers still paying a price that is higher than the social marginal cost. Thus for trucks, if needed an adjustment post is added as Some of the new vehicle kilometres by truck or car will be transferred from substitute markets such as transport by train. If social marginal cost pricing to the consumer does not prevail in these markets, adjustments must be made in the calculation.
A post is added as If the price of the mode of transport is lower or higher than the social marginal cost, a benefit of for example saved negative external effects or lost positive external effects must be added. Only if full competition prevails, public sector regulation with Pigouvian taxes or subsidies is enough to achieve social marginal cost pricing for consumers. If the market is a private monopoly or an oligopoly, or a public company with economies of scale and full cost pricing, or private timate the gains in safety, less congestion, fuel efficiency and parking benefits.
Legal and liability issues, security and privacy concerning vehicle-related data and vehicle costs are regarded as constraints to achieving the benefits. Gruel and Stanford [11] identify long-term potential impacts of AVs and conclude that there are positive and negative outcomes, and that it is unclear whether they will be a societal net benefit or harm. In all scenarios safety and mobility will increase, better use of travel time and lower fuel consumption are among the benefits but more travelling increases costs. Litman [12] summarizes the benefits as reduced fuel consumption, driver stress and paid drivers' costs, mobility for non-drivers, safety, road and parking capacity and support for sharing vehicles.
The costs pointed out are for the vehicles and infrastructure, risks, privacy, increased travelling, social equity and reduced employment.
Milakis et al. [13] carry out a literature review and expect road capacity, fuel efficiency, emissions and accident risks to give positive effects. They find that automated vehicles can lead to additional travel demand, and that the impact of potential land use changes, safety, economy, public health and social equity remain unclear. Wadud [14] concludes that the owners of commercial vehicles are most likely to be early adopters because they can save on driver costs and their ratio of benefits to costs is high. For regular cars, the largest gains are among people with the highest income, because they travel more and place a high value on time.
Bansal and Kockelman [20] find that 25 percent will be autonomous by 2045, with the assumption of an annual five percent fall in price at constant willingness to pay compared to 2015. This share will increase to 87 percent on the alternative assumption of a 10 percent annual rate of decline in prices and a 10 percent annual rise in willingness to pay. Litman [12] projects that in the 2030s, 10 -20 percent of the vehicle fleet will be autonomous with a moderate price premium, while in the 2050s, 40 -60 percent will be AVs, which by then will have become a standard feature on most cars sold. Milakis et al. [13] expect the penetration to range between 1 and 11 percent for 2030, and between 7 and 61 percent in 2050. The Swedish National Road and Transport Research Institute

Type Vehicles
For freight transport by truck we use three vehicle types: one for long-distance transportation with 6 -7 axles and a capacity of 40 tonnes (HGV40), one for dis-

Social Generalized Cost with Manually Driven Trucks and Cars
The Swedish Transport Administration [31] recommends unit values for the costs of travel, such as travel time, traffic safety and environmental impacts of transportation, based on research and valuation studies. These values are used on a regular basis in Sweden. In this study time-related costs have been transferred into costs per vehicle kilometre, with the aforementioned average speed for the different vehicle types. The Swedish Transport Administration [31] recommends an increase over time of values such as travel times and environmental impacts because the relative willingness to pay for these is expected to increase with economic growth. The price of fuel is also expected a rising trend due to increased scarcity. 1 Table 2 shows the unit costs for trucks. The opportunity cost for fuel costs is the product price including VAT, and additionally, the resulting external effects are accounted for in SGC. The variable capital and maintenance costs, the costs for the driver and the value of the goods that is bound during transportation are the remaining parts of SGC.
Concerning passenger transportation, the Swedish Transport Administration [31] recommends values of travel time for car and train that differ between national and regional travelling and between work and private travel. Other unit costs for passenger transport by car are fuel costs, capital costs and maintenance costs, and external costs in the form of accidents, emissions and noise. We use a weighted average of 0.074 litres of fuel per vehicle kilometre, based on the Swedish National Road and Transport Research Institute [33]. Table 3 shows unit costs for passenger traffic by car.

Social Generalized Cost with Autonomous Trucks and Cars
To calculate how the social generalized cost changes with autonomous trucks and cars, we must make some assumptions about how factors change, based on previous research presented in Section 2.
For long-distance trucks, autonomous driving will reduce fuel consumption because platooning can be established on the road, leading to less wind resistance for the vehicles inside the platoon. In addition, fuel can be saved by smoother driving in general. We estimate that fuel consumption can be reduced by 10 percent for long-distance truck, but not for the other types of trucks, as they typically cannot form platoons. Moreover, we conclude that AVs will not be introduced until they are at least as safe as MDVs, and we make a modest assumption that accidents will be reduced by 10% in the example calculation. With AVs, the entire driver cost will be saved, as there is no need for a driver to be on board, or if the driver is on board, the person can perform other tasks, thus freeing someone else. 1 We     what extent reduced emissions will benefit the consumer depends on the internalization of the external effects in the fuel taxes. The only effect that will be external is most of the change in accidents in 2040 because it changes more than fuel consumption. If accidents would change to zero with AVs the effects not internalized would still only account for less than one third of the effects for private travel and one tenths for work travel.

Capital Costs
It is difficult to predict how much different capital costs will be for AVs com-

Uncertain Effects of AVs
This section will discuss effects of autonomous vehicles that are not quantified or valued in this study. These effects may add to both benefits and costs, and reliable information about their future impact is still lacking.

Vehicles and Infrastructure
When autonomous vehicles begin to operate in regular traffic, additional costs may arise, related to the trucks and cars. Some may be of transitional character, such as development costs for the new technology, whereas others may be more persistent, such as higher production costs for more advanced vehicles and maintenance. In the longer term, however, production of vehicles may become less expensive, because they can be designed without any devices for driving, and become more productive when the space for the driver of trucks is transferred into more room for cargo. Capital costs for the vehicles may eventually decrease because of learning by doing and economies of scale in the manufacturing industry. The costs of developing the necessary technology for the vehicles and any higher production costs for the vehicles will be included in the price the consumer pays for the autonomous vehicles. As mentioned earlier, we study how much capital costs can increase in order not to completely counteract the quantified and valued net benefits.
In the short term, the digital infrastructure can require more investment, and, as long as the traffic is mixed between AVs and MDVs, special lanes may be needed. Development costs for the new technologies may also be significant. In the longer term, on the other hand, investments in infrastructure may decrease.
Private cars can use road and parking capacity more efficiently, requiring less space. Fewer or narrower streets and lanes may come as a result, as well as reduced demand for parking spaces in areas with high land value. However, vehicles parking in other areas could induce more traffic. Zakharenko [23] found that with the introduction of AVs, commuters' cost per kilometre would fall and their welfare would increase, as would travel distance and city size. Land rents would therefore go up in central parts and decrease in more remote areas. The magnitude of the effects on infrastructure and land use is uncertain, as is the question of whether the net effects will add to benefits or costs.

Other Benefits and Costs
Older people, people with disabilities and those without driving licenses can benefit from being able to ride in an autonomous car. Their improved potential mobility is discussed by Harper et al. [21]. Based on travel patterns from 2009 and the assumptions made, they estimate an upper limit for traffic to increase by 14 percent. People without a driving license could increase traffic by a maximum of 9 percent and disabled people by up to 2.6 percent according to their study.
AVs may also promote car sharing, both between existing car owners and between new groups. Fagnant and Kockelman [16] present many different scena-P. Andersson, P. Ivehammar Journal of Transportation Technologies rios and find that for the participating persons each AV could replace up to ten regular cars, adding up to ten percent more vehicle kilometres. The future effects of mobility for new groups and car sharing remain uncertain and have not been possible to quantify or value in this study. Bellem et al. [28] studied comfort in automated vehicles and recommended autonomous driving styles that can be perceived as comfortable by participants in a simulation, regardless of their personality or own driving style. Milakis et al. [29] used experts to assess the accessibility of AVs. They find benefits highly uncertain, that AVs can spur more dense urban centres, and that it is unlikely that the benefits of better accessibility will be evenly spread among social groups.
The effects on people's perceived safety and concerns for privacy when large amounts of data need to be processed for the required communication are not quantified and valued in this study. People may presume higher risks related to vehicles without a driver and to reduced privacy, at least during an early stage.
On the other hand, driver-related mistakes are eliminated. Bansal and Kockelman [20] estimated the willingness to pay among a sample of people in Austin to be 7253 USD for a level 4 AV, and that fewer accidents was the perceived most important benefit and equipment failure the biggest risk. König and Neumayr [24] also identified mixed attitudes towards AVs and that safety and reliability would increase over time. Kyriakidis et al. [17] studied 5000 people in 40 countries and 69 percent stated that AVs would reach 50 percent of the market before the year 2050. Risks were related to software, legal issues and safety. It is likely that the new technology will not be introduced on a mass scale until safety is at least as high as with today's vehicles. In this study, this is reflected in lower costs for accidents. Any effects on perceived safety that differs from the assumed change in safety and on the cost for privacy are not quantified in this study.
Problems with congestion can be both enhanced and mitigated. This is determined by how the quantity of travelling, the traffic flow and the number of vehicles change with AVs. Improved mobility may stimulate more traffic, as well as empty vehicles driving to their parking spaces or picking up people sharing cars, which may counteract the positive effects. Transport Analysis [18] has analysed the effects of shorter distances between vehicles on highways and at crossings with traffic signals in Sweden. On highways, mobility will only improve markedly when AVs have significantly outnumbered MDVs. Then, the space between vehicles can be reduced to 0.1 second. At crossings, significant gains will occur with SDV shares of above 80 percent. Zaiqiang and Ting [27] estimated a model with mixed traffic and concluded that with a market share of less than 39 percent, AVs will influence traffic flow negatively and not until the market share exceeds 68 percent will they significantly improve traffic capacity on the road. The shares of AVs used in this study do not reach the levels where the large effects supposedly occur, which is why changed congestion is not quantified.

Example Calculations
In this section, values for SGC presented in Section 4 are used to make an example calculation of the magnitude of the effects of automated vehicles in 2025 and 2040. An initial example calculation of benefits and costs of self-driving vehicles on the road was made by the authors for a Swedish Government Inquiry [36].
The present example is a revised and extended calculation. Increased traffic that will follow as a result of the lower generalized cost associated with AVs (Q 2 -Q 1 in Formula 2) is estimated with elasticities. Table 6 shows the estimated number of vehicle kilometres for the vehicle types and the share of AVs that is assumed for each, excluding the resulting newly generated traffic. This is used to calculate the value of Q 1 in Figure 1. The shares of AVs are based on the literature review. For MGV24 some traffic is assumed to take place in designated closed industrial areas even in the reference alternative.

Traffic Volumes and Shares of AVs
Transport Analysis [39] estimates that in 2015, 65,854 million vehicle kilometres were driven by passenger cars on Swedish roads. The forecast increase in passenger traffic by car is based on the Swedish Transport Administration [40].
As for trucks, this does not include the increase in traffic that will follow as a result of the lower generalized cost of AVs. Table 7 shows the estimated number of vehicle kilometres for an average car and the share of AVs, excluding the newly generated traffic. Source: The Swedish Transport Administration [38], and own estimates based on Transport Analysis [30]. Journal of Transportation Technologies

Elasticities
Litman [41] [42]) and studies referred to in Litman [41] and SIKA [43]. We use a price elasticity of −0.6. Based on a review of cross price elasticities for train due to changes in the prices for driving by car (Dickinson and Wretstrand [44]), a cross price elasticity of 0.25 for passenger traffic by train is used. Combined with the shares of passenger traffic by train and car, 5 percent of the new traffic by car is assumed to be transferred from train. Based on the unit costs for freight trains (The Swedish Transport Administra-Journal of Transportation Technologies tion [31]) and an assumption of the same average speed for freight trains as for trucks, the social marginal costs for freight transportation by train are estimated.

Differences between Price and Social Marginal Cost
Based on the annual report from Green Cargo [45], the largest provider of freight transportation by train in Sweden, an average price for consumers of 2.3 times higher than the marginal cost for the provider and 1.8 times higher than the social marginal cost is estimated. According to the Swedish Transport Agency [46], the profit of freight train businesses is insignificant or even negative. Thus, as the price exceeds marginal cost without resulting in significant profits, the explanation is high fixed costs. Consequently, in the present study, a price for transport by train that is 80 percent over the social marginal cost is used and an adjustment post is added. Freight transportation by truck is considered a competitive business, based on the Swedish Transport Agency [47].
Overhead costs in the industry are 10 -20 percent of total costs. Thus, we use a price in this sector that is 20 percent over the social marginal cost. Additional posts are added in the example calculation for freight accordingly, as shown by Formulas 4 and 5.
Based on the annual report from the main state-owned company for passenger traffic, SJ [48], the traffic volume (Transport Analysis [49]), the average speed, the average number of passengers per train (The Swedish Transport Administration [31]) and external costs (The Swedish National Road and Transport Research Institute [33]), the price for passenger traffic by train is estimated to be 2.5 times the social marginal cost. For traffic shifted from train to car, an adjustment post as in Formula 5, with a cost of 0.51 SEK per vehicle kilometre is added.

Results
In this section, example calculations of the net benefits of AVs are carried out and considered in relation to how much the capital costs (defined earlier) can increase without making AVs unprofitable for society. The analysis includes freight transport by truck and passenger transport by car, and considers effects on the substitute markets passenger transport and freight by train. Transportation in the substitute markets air and sea is not considered. They account for 2.8 and 0.6 percent of passenger kilometres respectively in Sweden (Transport Analysis [39] [50]). When calculating new traffic for passenger transport by car it is, as mentioned, assumed that all external effects are internalized to the GC for the consumer. This means that the savings in GC that the traveller in an AV will perceive are somewhat higher than the savings for society in 2025 but somewhat lower in 2040. The savings from reduced fuel when the price includes all external effects will include noise that will remain unchanged, but the savings in traffic safety in 2040 will be greater than the part included in the fuel price (saved fuel P. Andersson, P. Ivehammar Journal of Transportation Technologies is 10% while increased traffic safety is 30%). The difference is less than five per-   If all external effects are internalized in GC for the consumer, the effects of decreased GC that the traveller in an AV will experience are somewhat larger than the savings for society in 2025 but somewhat lower in 2040. The difference is less than five percent between what the consumers perceive and the total effects for society. For trucks, the difference is less than three percent.

Sensitivity Analysis
A sensitivity analysis means that quantities and values are varied to see the effect on the overall result. One factor at a time is varied in the example calculation for AVs. Table 10(a) shows the results for 2025 and Table 10(b) for 2040. The first row shows how much capital costs can increase to reach the point where society's net benefits are equal to the costs in the main calculation presented above, and then the percentage with some alternatives. It can be noted that transport volume and the share of AVs have no impact on whether benefits exceed costs of AVs in the example calculations, since benefits and costs will increase by the same proportion because all effects in the example calculations are per vehicle kilometre. Thus, forecasts of growth in traffic or shares of AVs will not determine whether AVs' benefits exceed costs to society, only how big the gain (or loss) is. For this to alter, one benefit or cost must change more than the others do. As the sensitivity analysis shows, altering the quantities or values related to saving driver costs has the biggest impact on the results.
It can be noted that transport volume and the share of AVs have no impact on whether benefits exceed costs of AVs in the example calculations, since benefits and costs will increase by the same proportion because all effects in the example calculations are per vehicle kilometre. Thus, forecasts of growth in traffic or shares of AVs will not determine whether AVs' benefits exceed costs to society, only how big the gain (or loss) is. For this to alter, one benefit or cost must change more than the others do. As the sensitivity analysis shows, altering the quantities or values related to saving driver costs has the biggest impact on the results. Table 11 is a summary of the shares of the net benefits for autonomous trucks and cars in the example calculations, without capital costs.

Conclusions
The most important factor for benefits to exceed costs for AVs for freight transport by truck is the share of the number of vehicle kilometres by AVs driven without a driver. The sensitivity analysis underlines that only a minor variation is crucial for the result for autonomous trucks. For autonomous cars, the most decisive effect is the extent to which the drivers' value of travel time changes. Journal of Transportation Technologies  Reduction in fuel consumption because of convoys and smoother driving and the resulting decrease in emissions on the roads are benefits. However, a shift to fossil-free fuels in the future may reduce these benefits. In the example, emission reductions represent 25 -28 percent of the benefit of reduced fuel consumption for cars and 40 -42 percent for trucks. As is demonstrated by the example calculations, these factors contribute to a relatively small share of the total benefits. Better safety is another effect. As the effect of freeing drivers from their duties is so dominant, alternative scenarios about fuel consumption or safety have small effects on the result. The acceptance level for autonomous vehicles is another crucial factor. We have used a modest decrease in accidents by 10 percent for 2025, but it is possible that the autonomous vehicles must be a lot safer than MDVs before they will be accepted on the roads.
As autonomous vehicles have a lower generalized cost, the total traffic will increase. More traffic is a net benefit, as long as the price to the consumer is equal to the social marginal cost. This contributes 2.4 -4.1 percent of the total benefits, including the lost value of rail traffic.
The real gain is achieved if trucks and cars can indeed be driver-free and resources saved for other uses. For drivers of cars, this shift is unproblematic. For trucks, it means a period of structural change in the labour market. As autonomous trucks will be replacing manually driven trucks during a period that will probably last for several decades, the labour market will have time to adjust and no permanent unemployment can be expected.
No assumptions are made about how much an AV will cost. The example calculations show how much higher capital costs (including technology development for vehicles, higher capital costs due to more expensive production costs and changed maintenance costs) for AVs can be compared to MDVs without costs being larger than the benefits. In the example calculations, the costs can increase more than marginally without making AVs unprofitable to society, as presented in Table 10. The margin for increases in capital costs per vehicle kilometre for trucks is relatively higher.
The results show that nearly all benefits will come to consumers and producers of the goods. Thus, an important policy conclusion is that there is no reason for giving subsidies to the industry producing AVs or to the consumers buying Journal of Transportation Technologies them. The market mechanisms will lead to the introduction of AVs without subsidies if it is profitable. The sole external effect in this study is part of a possible future reduction in accidents compared to MDVs, but this effect is relatively small in Sweden. However, it is important to adjust regulations so that the new technology can be implemented if safety is satisfactory.
In addition to the contribution from this study, the value of new groups being able to ride a car, the perceived safety, privacy, changed congestion and change in land use are factors that remain to be analysed.