Impact of automated driving systems on road freight transport and electrified propulsion of heavy vehicles

The technological barriers to automated driving systems (ADS) are being quickly overcome to deploy on–road vehicles that do not require a human driver on–board. ADS have opened up possibilities to improve mobility, productivity, logistics planning, and energy consumption. However, further enhancements in productivity and energy consumption are required to reach CO2–reduction goals, owing to increased demands on transportation. In particular, in the freight sector, incorporation of automation with electrification can meet necessities of sustainable transport. However, the profitability of battery electric heavy vehicles (BEHVs) remains a concern. This study found that ADS led to profitability of BEHVs, which remained profitable for increased travel ranges by a factor of four compared to that of BEHVs driven by humans. Up to 20% reduction in the total cost of ownership of BEHVs equipped with ADS could be achieved by optimizing the electric propulsion system along with the infrastructure for a given transportation task. In that case, the optimized propulsion system might not be similar to that of a BEHV with a human driver. To obtain the results, the total cost of ownership was minimized numerically for 3072 different transportation scenarios that showed the effects of travel distance, road hilliness, average reference speed, and vehicle size on the incorporated electrification and automation, and compared to that of conventional combustion–powered heavy vehicles.


Introduction
In the near future, transportation will experience substantial development in the domain of automated driving systems (ADS), which will revolutionize the way people and freight move on-road, as reported by Wadud et al. (2016) and Flämig (2016). Remarkable advantages in terms of user experience, efficiency, safety, mobility, productivity, energy, environment, and economy have been reported with ADS by Alessandrini et al. (2015), Anderson et al. (2014), Brown et al. (2014), Chan (2017), Harper et al. (2016), Levin and Boyles (2015), Maurer et al. (2016), Wadud (2017), Wadud et al. (2016), Taiebat et al. (2018) and Khan et al. (2019), though significant increases in traffic safety due to highly or fully automated vehicles are not certain as revealed by Kalra and Paddock (2016). However, user objectives and motivations differ for passenger cars and freight transport, as reported by Wadud (2017) and Nowakowski et al. (2015). For passenger cars, the major motivations are user experience and environment, whereas in freight transport, as a subject of this study, the most important driving forces are productivity and profitability. For example, increase and changing lanes belong to level-2. A level-3 DAS (i.e., conditional driving automation) is capable of performing the entire dynamic driving task, but the driver should be ready to intervene upon system request. In levels 1-3 of a DAS, the presence of a human driver is essential in the cabin. In level-4 (i.e., high driving automation) or 5 (i.e., full driving automation), a DAS is capable of performing the entire dynamic driving task, including bringing the vehicle to a minimal risk condition in case of a failure. According to SAE standard (2016) J3016, ADS refer to levels 3-5, where the DAS can perform the entire driving, and a vehicle equipped with level-4 or level-5 DAS is referred to as an ADS-dedicated vehicle (ADS-DV). In levels 4 and 5, the presence of a human driver is not needed in the vehicle; instead, a remote dispatcher verifies the operational readiness of the vehicle and performs the dynamic driving task remotely, whenever necessary. High DAS and full DAS differ in their operational design domains. Operational design domain refers to the conditions under which a given DAS is designed to operate. A high DAS is limited to a specific operational design domain, whereas a full DAS is designed to function on all roads and conditions that are navigable by a human driver. The subject of this study was ADS-DVs, i.e., levels 4 and 5 of DAS.
TCO is usually used for comparative analysis of the competitive technologies, e.g., vehicles with different powertrains, and provides a good means of estimating profitability, according to, e.g., Davis and Figliozzi (2013), Feng and Figliozzi (2013), Lee et al. (2013), Taefi et al. (2015), Wu et al. (2015), Lebeau et al. (2015), Hagman et al. (2016), Taefi et al. (2016a), Taefi et al. (2017), Pelletier et al. (2016), Wadud (2017), Palmer et al. (2018), and Lebeau et al. (2019). TCO measures the life cycle cost, including the operational costs and the depreciation of the purchase price. Operational costs usually include the costs of fuel, maintenance, tax, insurance, and electric energy in the case of battery electric vehicles. In addition, in the case of commercial ADS-DVs, the operational costs associated to remote dispatchers are also included. Purchase price comprises vehicle hardware costs and, in the case of ADS-DVs, the cost of additional sensors and investment on remote dispatchers as well. Wadud (2017) investigated the potential adoption of ADS-DVs on public roads by performing TCO analysis of private and commercial vehicles. The commercial vehicles included taxis and conventional trucks with gross mass of 7.5, 18 and 38 ton. They concluded that the commercial vehicles benefit more from automation, specially small trucks and taxis, because a large share of TCO belongs to the driver cost in small commercial vehicles. They, however, did not discuss electric vehicles.
The literature about TCO of road vehicle competitive powertrain technologies all involve human drivers. As such, Wu et al. (2015), Hagman et al. (2016) and Palmer et al. (2018) are concerned with passenger cars. Vehicle types, operations and purchase decisions are, however, different in road freight transport.
In road freight transport sector, Davis and Figliozzi (2013), used a cost function similar to TCO together with powertrain and logistics constraints. They compared the cost of two different battery electric and one conventional trucks on 243 transportation and driving scenarios with different costumer demands and operating speeds. The battery electric trucks had about 7.5 ton gross mass and 161 km driving range. The authors concluded that high utilization, low speed, frequent stops, tax incentives, and planning time horizon, i.e. vehicle life time, beyond 10 years can help the competitiveness of the electric trucks against conventional counterparts. Feng and Figliozzi (2013) implemented a fleet replacement optimization framework that allowed replacing the conventional trucks with the battery electric trucks. They compared a small electric truck with a conventional truck of the same size on six different scenarios. The driving range of the electric truck was 161 km. They concluded that the electric truck can be cost effective if the annual utilization level is high. Lee et al. (2013) compared TCO of a battery electric and a conventional truck with maximum gross mass of 7.49 ton on two driving cycles. The driving range of the electric truck was 161 km. They concluded that the relative benefits of electric trucks depend on vehicle efficiency associated with the driving cycle, diesel fuel price, battery price and replacement, charging infrastructure, and purchase price. They showed that the electric truck had lower TCO compared to the conventional one, without including subsidies, for a driving cycle with frequent stops and a low average speed. Taefi et al. (2015) analyzed profitability concept of existing urban freight battery electric vehicles by interviewing companies to examine whether and how they operate profitably. Also, they performed a statistical analysis in Europe north sea region to capture the trends of the existing electric urban freight transport. The study identified two current trends of deploying electric vehicles in urban freight transport in north Europe: 1) slow and light electric vehicles, 2) medium heavy electric trucks in last mile logistics. In one of the cases, a concept truck of 12 ton gross mass was profitable that was achieved by considering measures in reduction of purchase and operational costs, and by increasing vehicle utilization. These measures included vehicle customization, subsidies and exemption from city toll, intermediate and quick charging, multi-shift operations, improvement of routing and scheduling, and etc. Taefi et al. (2017) evaluated TCO considering battery health and replacement as a function of mileage at a given average energy consumption and warranted maximum mileage or maximum number of battery charge-recharge cycles. They calculated the costoptimal mileage for three different electric trucks of about 12, 7.5 and 5 ton gross mass and ranges of 200, 160 and 120 km, respectively, and compared their TCO with the conventional counterparts. The authors used a fixed battery resale, i.e. rest, value regardless of the battery state of health as in the other literature. In their case, the electric trucks could not compete with conventional counterparts. They suggested that, in order to reduce TCO of each electric vehicle, the best cost-effective mileage should be calculated and planned, rather than selling the vehicle at a time when the battery end of life is reached. Lebeau et al. (2015) and Lebeau et al. (2019) also evaluated TCO for several different light commercial vehicles. They concluded that these vehicles could compete with conventional counterparts if the vehicle utilization is high, and with the help of governmental subsidies. In addition, they showed that the period of ownership, the residual value and second life of the battery effect TCO of electric light commercial vehicles.
Reviewing the current available literature, the following gaps can be identified.
-TCO analysis of the battery electric ADS-DVs, e.g. ADS-DV BEHVs, is not conducted in the literature, and hence, no comparison is T. Ghandriz, et al. Transportation Research Part C 115 (2020) 102610 made against the BEHVs with human drivers. -Even though the literature suggests that vehicle customization based on the use case helps cost effectiveness of the battery electric trucks, the implications of such a customization on TCO is not studied in the literature. Consequently, there is no TCO minimization for varying vehicle parameters, e.g. battery size. -The literature neglects energy consumption evaluation based on longitudinal vehicle dynamics on roads of different topographies and speeds. Thus, the cost-effectiveness of BEHVs is not investigated for different road hillinesses. -TCO calculation of the electric vehicles in the literature does not include the driver cost, and thus neglects additional driver cost as a result of waiting time for charging during operation of the electric vehicles, compared to the conventional vehicles. Similarly, the trade-offs between the driver cost and charging time, battery degradation, charging power and cost, LU time and cost, and slow driving are neglected in TCO calculations. -TCO analysis of battery electric trucks does not include trucks weighting more than 12 ton, and thus, possible transportation scenarios where BEHVs could be competitive to their conventional counterparts (with or without human drivers) are not identified.
This study tried to fill the above gapes and supports all the previous reports on the factors reducing TCO of the electric vehicles; in addition, we emphasized on the importance of the vehicle-infrastructure simultaneous optimization for a given transportation scenario. Optimizing vehicle-infrastructure is useful for the reason that TCO comparison of different BEHVs by itself does not reflect the profitability, owing to the performance constraint imposed by the range and power of batteries. Nevertheless, this problem can be overcome by sizing the batteries for a given transportation scenario. Moreover, the properties of a driving cycle such as road topography (i.e., hilliness), speed, and distance traveled influence vehicle performance, especially that of BEHVs. Thus, in this study, the performance and TCO of the optimum vehicle propulsion system were evaluated and compared for transportation tasks of different characteristics.
Furthermore, full automation of freight transport involves automation of LU to replace the role of the driver that performs part of this task. Automated guided vehicle systems (AGVS), developed with the purpose of optimizing material flow and reducing personnel, as reported by Flämig (2016), can be used for automated LU. AGVS facilitate 24/7 operation of vehicles, because they navigate automatically by themselves and perform well, owing to the repetitive nature of the operations. AGVS have been widely deployed in the industry and warehouses to reduce the TCO, especially in multishift operations, as reported by Ullrich (2015) and Liu et al. (2004). However, there exists no standard yet for automated truck loading and unloading. Furthermore, the utilization of charging stations at the LU point or consumer locations is motivated by the fact that it saves cost and time compared to the use of publicly scarce charging stations, according to Kopfer and Vornhusen (2017), and that virtually no high-power charging stations are available yet for BEHVs. Moreover, it will be more feasible if charging can be accomplished during the same time as performing LU, as reported by Taefi et al. (2016a) among measures of supporting freight electric vehicles. Thus, this study included the cost of infrastructure in the calculation of the TCO which comprises the cost of LU and charging stations.

Methodology
The influence of different aspects that are related to the transportation task and vehicle propulsion system was studied for ADS and vehicle electrification. More specifically, these aspects included the following.
• Transportation task -Vehicle size -Driving cycle, i.e., the distance between charging stations and LU points, average reference speed, and road hilliness -Infrastructure, i.e., charging power and LU scheme • Vehicle propulsion system -Type of battery -Size of battery -Type of electric motors -Number of electric motors -Size of internal combustion engine (ICE) The different aspects of the transportation task and vehicle propulsion system explained above were examined on four different plans. The plans concern different DAS levels and vehicle sources of power, shown in Table 1.
In this study, the operational design domain of the DAS comprised all defined transportation tasks, including all the trips, roads,

Table 1
Plans concerning the DAS levels and vehicle power source.

CHV BEHV
With driver (level-2 or level-3) Plan-1 Plan-2 Without driver (level-4 or level-5) Plan-3 Plan-4 T. Ghandriz, et al. Transportation Research Part C 115 (2020) 102610 and dynamic driving tasks within them. Furthermore, vehicles were designed to operate exclusively on their assigned transportation tasks during their entire service life. Moreover, the ability of vehicles to operate outside the assigned transportation task might be constrained mainly by the propulsion system, and not by the DAS. Hence, in this study, the ability of vehicles to operate outside the operational design domain as governed by DAS was not emphasized, thereby including both levels 4 and 5 in a single plan. In addition, it should be noted that, in all the plans, vehicles were equipped with a DAS higher than level-1, which meant that they all benefited equally from the eco-driving and fuel/energy efficiency offered by the DAS, as reported by Mersky and Samaras (2016) and Brown et al. (2014).
After the exclusion of the driver on-board in ADS-DVs, they can be considered to be without driver interfaces, such as steering, braking, and acceleration input devices. Moreover, given the purpose of exclusive freight transport, all seats and the cabin can be removed. Fig. 1 depicts the freight heavy ADS-DV combinations of different sizes, as well as the heavy vehicle combinations with human drivers, that were considered in this study. These vehicles are called "rigid truck," "tractor and semitrailer," "Nordic combination," and "A-double" from the smallest to the largest, in that order.

Transportation task
A transportation task is defined by a distribution network comprising nodes, the routes between them, and pick-up/delivery demands. In this study, a distribution network included only two nodes. A vehicle was completely loaded/unloaded at each node while charging (in the case of BEHVs). Different transportation tasks were considered with driving cycles, i.e., different distances between nodes, road hillinesses, and average reference speeds, in order to investigate their influences on the incorporation of ADS and electrification. Average reference speed refers to the speed that a vehicle tries to maintain during the entire trip. Moreover, it was assumed that there always existed goods that needed to be transported within the network, therefore, both ADS-DVs and vehicles with human drivers operated 24 h every day (24/7) on a repetitive basis. The sensitivity of the results to lower utilization levels was also revealed.

Road hilliness
To investigate the influence of road hilliness on vehicle performance and propulsion system, roads with different hillinesses were considered in this study. According to the definition of global transport application (GTA) reported in the papers by Edlund and Fryk (2004) and Pettersson et al. (2018), road hilliness was categorized into four levels: flat, predominantly flat, hilly, and very hilly. In order to not be restricted to a specific geographical area, the roads in the different categories were modeled mathematically in this study, according to the works of Johannesson et al. (2016) and Pettersson et al. (2016).
H m be the selected hill length, = L 50 s m sample road distance, y variance in road slope, y k slope of the road in percentage, and k road grid index. Then, the road topographic profile can be generated using an auto-regressive model: (2) where, N denotes a normal distribution with standard deviation e . Finally, elevation z k is given by Parameter y determines the level of hilliness according to GTA. y values in the ranges (0,1.3], (1.3,2.3], (2.3,3.2], and larger than 3.2 correspond to flat, predominantly flat, hilly, and very hilly roads, respectively. Roads of different lengths and hillinesses were generated using the model mentioned above by choosing y as 0.5, 1.5, 2.5, and 3.5 for the different levels of hillinesses. As an Fig. 1. Different sizes of freight heavy ADS-DV combinations (right) and heavy vehicle combinations with human drivers (left). Vehicles, from the smallest to the largest, are respectively called "rigid truck," "tractor-semitrailer," "Nordic combination," and "A-double.".
T. Ghandriz, et al. Transportation Research Part C 115 (2020) 102610 example, the different road elevations have been depicted in Fig. 2 for a road of length 10 km for travel back and forth between the two ends.

LU schemes
There is a trade-off between LU duration and TCO, because LU duration influences the temporal utilization level of a vehicle (i.e., vehicle-time on-road). Moreover, charging was considered to take place at the same time as LU occurred, thereby influencing the charging power required for providing sufficient energy to reach the next charging station. This study considered four LU schemes that were all executed automatically, mostly with the aid of AGVS, and are as follow.
-on-board waiting; in this case, vehicles wait until LU is performed by automated guided vehicles such as an automated lift-truck, which positions pallets of goods inside the containers using prescribed coordinates (Ullrich (2015)). -straddle carrier (SC); in this case, containers can be lifted and carried by an automated straddle to assigned positions.
-additional semitrailer (AST); in this case, a semitrailer can be connected/disconnected to/from a tractor or dolly by an automated docking/undocking mechanism, while the vehicle is parked in a prescribed position. -on-board lift; in this case, an on-board lift installed on the tractor and/or semitrailer is carried by the vehicle and automatically performs vehicle LU upon reaching the prescribed position.
The investment and operational cost of the LU schemes described above as well as their durations were considered in the evaluation of the TCO, and are provided in Appendix. It should be noted that the cost of emptying or filling an unloaded container was not considered, because it was not part of the transportation task and was constant in all the transportation scenarios.

Vehicle propulsion system
The most significant components of the propulsion system considered in this study include battery type and size, as well as electric motor type and number, in BEHVs; and the size of the ICE in CHVs. Given a source of power (i.e., battery or ICE), the role of human driver and ADS, vehicle size, and driving cycle, an optimum propulsion system was determined by minimizing the annual TCO per unit freight transported. The components of the propulsion system (i.e., the design variables) were selected from given discrete sets. The LU scheme and charging power at each node may also be considered as the design variables of the optimization problem. The optimization problems and constraints have been defined in Appendix.

Transportation mission management system (TMMS)
In real world, a driver may perform tasks other than dynamic driving; for example, assisting in LU or strategic functions such as trip scheduling and routing. These kinds of tasks are not performed by ADS. They are executed either automatically by AGVS or by a personnel/dispatcher as part of a TMMS. Dispatchers are responsible for monitoring the operational readiness of vehicles, as well as performing dynamic driving tasks remotely whenever necessary; for example, returning a vehicle to the depot in case of dynamic driving task performance-relevant system failure. A dispatcher may also perform strategic functions or monitor the functionality of AGVS for LU.
The costs pertaining to a TMMS include the operational costs such as the salary of the dispatcher and the equipment maintenance and investment cost related to, for example, a control tower. This study considered the following equivalent TMMS costs that resulted in the same annual cost. T. Ghandriz, et al. Transportation Research Part C 115 (2020) 102610 -two personnel/dispatchers per fleet vehicle, each working 40 h per week with the same salary as that of a driver on a per hour basis; -one personnel/dispatcher per fleet vehicle who works 40 h per week with the same salary as that of a driver on a per hour basis, and 350000 € on investment; -one personnel/dispatcher for five fleet vehicles who works 40 h per week with the same salary as that of a driver on a per hour basis, and 620000 € on investment.

Cost function
The annual TCO per unit freight transported between two nodes C t was considered as a measure of vehicle performance. The cost function includes the operational costs and the depreciation of the purchase price, and is defined as follows.
where N v denotes the number of fleet vehicles; f tr denotes the annual number of freight units transported in a round-trip between two nodes; c c c c c c , , , , , elec fuel driver maint tax insu , and c tmms indicate the annual costs of electricity, diesel fuel, driver labor, vehicle maintenance, which includes tires, taxes, insurance, and TMMS, respectively; c dep denotes the depreciation or yearly cost of investment and is defined as follows.
where r n R R p , , , , , and p batt tot , denote the interest or discount rate, the economic life span in years, the vehicle-infrastructure resale value, the batteries resale value, and the purchase price of the vehicle-infrastructure excluding the batteries, and the purchase price of the all batteries including the replaced ones, respectively. The purchase price was calculated using the following equation and included the price of the vehicle chassis p chass , cabin price, including driver interfaces p cab , electric motors p em , transmission systems p trans , ICE p ice , ADS p ads , LU components p lu , recharging infrastructure p rech , and investment cost related to the TMMS such as that for a control tower.
Vehicle and battery-degradation models were implemented for calculating the operational and purchase costs in accordance with the works of Ghandriz et al. (2016) and Ghandriz et al. (2017). The maintenance cost of BEHVs was considered as 50% of that of CHVs, as suggested by Feng and Figliozzi (2013) and Lee et al. (2013). The ADS price p ads includes the price of all the sensors and computers needed for object and event detection and response. Moreover, in BEHVs, the resale value of the vehicle-infrastructure might be different from that of last replaced battery depending on battery state of health. In this study, the batteries were replaced if the battery capacity reached 80% of the initial capacity, and their resale value was set to zero. A possible second life application, as reported by Lebeau et al. (2015), was neglected. In addition, for calculation of p batt tot , , an yearly decrease in battery price, due to battery technology development, were considered. Furthermore, additional payload was allowed for BEHVs according to EU directive 2015/719, without considering any direct fiscal incentives. Please refer to Appendix A for further details.
By using such a cost function, optimization problems were defined to find an optimum vehicle-infrastructure design for a given scenario. A detailed definition of the optimization problems has been provided in Appendix A.

Results 1
A scenario comprises a given road with its hilliness and distance between LU nodes or charging stations, and set average speed, vehicle size, powertrain type (i.e. battery electric or combustion-powered), and role of driver (i.e., ADS-DV or human driven). Combining all these parameters yields 3072 different transportation scenarios. For each of the scenarios and for a single vehicle in the fleet with 100% utilization, the vehicle-infrastructure optimization problem was solved and the results analyzed. Fig. 3 reveals the annual TCO per unit freight as a function of the average reference speed of optimum tractor-semitrailers on a flat road with different lengths, as well as for different powertrains and roles of driver, when there is a single vehicle in the fleet and 100% utilization. Ghandriz et al. (2020) have provided the data and results for other vehicle sizes and road hillinesses. In the figure, each dot corresponds to an optimum vehicle-infrastructure and represents a solution of the optimization problem defined in Appendix A. It can be seen as to how the competitiveness of BEHVs against CHVs is affected by the distance between LU/charging nodes; moreover, reductions in TCO can be realized by using ADS in both BEHVs and CHVs. The following conclusions were drawn from Fig. 3.
-An optimized battery electric tractor-semitrailer with an optimized infrastructure can be more profitable than an optimized conventional combustion-powered tractor-semitrailer, if driving distances remain less than about 40 km for a vehicle with a 1 A representative set of results has been provided in this paper. The complete set can be found in the paper by Ghandriz et al. (2020).
T. Ghandriz, et al. Transportation Research Part C 115 (2020) 102610 human driver that is fully loaded on a flat road and with 100% utilization. -The range of driving a profitable battery electric ADS-dedicated tractor-semitrailer increases to about 80 km, which is twice that of a vehicle with a human driver. -The reduction in TCO achieved by electrification is higher in ADS-DV than in the vehicle with a human driver; likewise, the reduction in TCO achieved with ADS is higher in BEHVs than in CHVs. -The optimum average driving speed is different for different powertrains and roles of driver. The optimum average reference speed is between 60 and 80 km/h for a battery electric ADS-DV, whereas it is between 70 and 90 km/h for a BEHV with a human driver. A higher optimum average reference speed is observed for CHVs. -The TCO changes slightly within the range of average reference speeds between 50 and 90 km/h, in BEHVs, whereas the change in the TCO with speed is steep in CHVs, being up to 90 km/h.
It was also observed that the optimum vehicle-infrastructure (i.e., the type and number of electric motors, type and number of battery packs, LU schemes, and recharging power (P ch )) might be different in a battery electric ADS-DV and a battery electric vehicle with a human driver for driving at an optimum average reference speed on the same road. An example has been provided in Table 2. It must be noted that the optimum average reference speed might also be different in the two cases.
Furthermore, the effect of electrification and automation on different vehicle sizes was studied. The results for a scenario with a Table 2 Optimum vehicle-infrastructure of ADS-DV tractor-semitrailer and the one with human driver on a flat road of length 10 km. AST AST 130 10 * Specifications of the electric motors (EMs) and battery packs (BPs), e.g., EM 2 and BP 2 are given in Appendix. † LU i : LU scheme at i th node; P ch i , : recharging power at i th node. T. Ghandriz, et al. Transportation Research Part C 115 (2020) 102610 flat road of length 160 km are presented in Fig. 4. Please refer to the paper by Ghandriz et al. (2020) for the results for other roads. It can be seen that battery electric Nordic combination and A-double equipped with ADS display lower TCOs than those vehicles with human drivers up to the travel range of 160 km. The hilliness of a road was also observed to affect the competitiveness of BEHVs. Fig. 5 reveals the annual TCO per unit freight for a tractor-semitrailer on a 160 km road with different hillinesses. The results for other vehicle sizes and road lengths are provided in the report by Ghandriz et al. (2020). It can be seen that the distance between the cost curves of a vehicle with a human driver increases with an increase in hilliness. Moreover, a battery electric ADS-DV almost affords a lower cost than a CHV on a 160 km flat road, whereas such a vehicle completely loses its competitiveness on a very hilly road of the same length. The reason is that hilly roads require large batteries. As an example, the optimum vehicle-infrastructures of tractor-semitrailers equipped with ADS have been shown in Table 3 for roads of length 160 km with different hillinesses.
The reduction in cost achieved with ADS-DV ranges between 27% and 46% for BEHVs and between 11% and 41% for CHVs for the different scenarios. Fig. 6 illustrates the cost components of vehicles with different sizes for different roles of driver (i.e., ADS-DV and a  human-driven vehicle) and types of propulsion systems (i.e., battery electric and combustion-powered) on a flat road of length 160 km. It can be seen that the cost reduction achieved with ADS-DV is larger for BEHVs than CHVs in all the cases. Moreover, the cost reduction is lower for larger vehicles. The optimum vehicle-infrastructure designs corresponding to the scenarios shown in Fig. 6 can be found in Table 4.

Discussion
It has been demonstrated that the employment of ADS renders BEHVs competitive with CHVs over longer travel ranges compared to that of BEHVs with human drivers. Moreover, the optimum propulsion system setup and infrastructure of an ADS-dedicated BEHV was observed to be different from that of a BEHV with a human driver, in addition to differences in the vehicle hardware such as cabin, driver interfaces, sensors, computers, and actuators. ADS result in BEHVs with lower TCOs, mainly owing to there being no need to heat the cabin in ADS-DVs, as is the case when a human driver is involved, as well as the reduced optimum speed in ADS-DVs, which makes it possible to use smaller batteries. For a given transportation scenario, if a similar propulsion hardware as in a BEHV with a human driver is used in an ADS-dedicated BEHV, then the TCO might increase between 0% and 25%, compared with that of a ADS-dedicated BEHV with a uniquely designed propulsion system, depending on the vehicle size and transportation scenario. However, no change in the optimum propulsion hardware (i.e., the ICE) was observed in CHVs when replacing the human driver by ADS. Moreover, it was revealed that the TCO might increase by up to 35% if a BEHV designed to operate in a travel range of 80 km was instead used in transportation tasks where the travel ranges are only 10 km, regardless of whether it was a ADS-DV or not. It can be concluded that, in order to ensure competitiveness across different scenarios, the propulsion hardware should be adapted to the use case of BEHVs, irrespective of whether a human driver is involved. Such an adaptation, performed by vehicle-infrastructure optimization, explains why BEHVs with human drivers showed lower TCO than CHVs with human drivers, in many transportation scenarios of short road lengths, which was not observed in the literature before.
ADS reduce the TCO of BEHVs in all scenarios, but do not decrease it sufficiently to make them competitive with ADS-dedicated  T. Ghandriz, et al. Transportation Research Part C 115 (2020) 102610 CHVs on long and hilly roads. It has already been reported by Davis and Figliozzi (2013) that having many starts and stops in the driving cycle helps improve the competitiveness of BEHVs, owing to the possibility of energy recuperation. With this reasoning, it can be concluded that the hilliness of a road can exhibit a positive effect on the competitiveness of BEHVs owing to the recuperation of energy downhill. However, this reasoning is not entirely correct for BEHVs. In this study, it was observed that the hilliness of a road displays a negative effect on the competitiveness of BEHVs. On a hilly road with a grade of 11.5%, negotiation of the grade at the speed of 10 km/h requires approximately 2100% more power than what is needed for moving the vehicle on a flat road. Therefore, on hilly roads, the battery size is constrained by the power, and not by the energy needed for completing the trip, consequently, a larger battery is required. Moreover, frequent charging-discharging of batteries increases their degradation and shortens their service life. Consequently, the gain in cost due to energy recuperation is not comparable to the cost of the large battery, which renders BEHVs less competitive than CHVs on hilly roads. Contrary to this observation is the case when a power-optimized battery is used on short roads, or a vehicle is equipped with both power-optimized and energy-optimized batteries. The optimum reference speed was observed to be lower in BEHVs than in CHVs, which is in accordance with the results reported Fig. 6. Contributions of the different cost components to the annual TCO per unit freight (ton) of optimum vehicles with different sizes on a 160 km flat road. The cost structure has been shown for different optimum propulsion systems (BEHV and CHV) and infrastructure (LU and charging power), and for different roles of driver (i.e., ADS-DV and human-driven (HD)). The cost of ADS specific hardware is included in the chassis price.
T. Ghandriz, et al. Transportation Research Part C 115 (2020) 102610 by Taefi et al. (2016b), Lee et al. (2013). Likewise, ADS-dedicated BEHVs reveal lower optimum reference speeds than BEHVs with human drivers. The reason for this is difference in the energy consumption model and trade-offs between the power and battery size required in BEHVs and the absence of driver cost in ADS-DVs. However, the optimum reference speed seemed to converge to a high value (90 km/h) in large vehicles for all the scenarios as the depreciation cost increased; please refer to Table 4. Slow driving can be important for employing ADS-DVs on-road, because there are high safety requirements that need to be met along with technological limits on object and event detection and response. Thus, slow driving in ADS-dedicated BEHVs, with the corresponding optimum propulsion system setup, can yield the dual benefits of increased safety and reduced TCO, as in some transportation scenarios where the optimum speed of BEHVs is about 60 km/h. Moreover, the plots of annual TCO versus optimum reference speed reveal a mild slope for the TCO curve at a reference speed higher than 50 km/h in ADS-dedicated BEHVs. Therefore, if slow driving at 50 km/h is required, the resulting additional expense will be up to 10% of the TCO in BEHVs, whereas it can be up to 23% in CHVs. It can be seen in Fig. 6 that the cost reduction achieved with ADS is lower for larger vehicles. The reason is that the contribution of driver cost to the annual TCO is lower for larger vehicles, whereas the costs of TMMS and ADS remain almost constant for all vehicle sizes. This observation is in accordance with the results of Wadud (2017).
The reduction in cost achieved with ADS is highly dependent on TMMS costs. TMMS cost comprises operational and investment costs. Operational cost refers to the expenditure associated with maintaining the ADS-related equipment and the salaries of dispatchers. Investment cost is related to equipment that include "control tower," etc. The cost reduction achieved by employing ADS-DVs was observed to be between 27% and 46% for BEHVs and between 11% and 41% for CHVs for different scenarios, in the case of the assumptions regarding TMMS costs described in Section 3.3. Instead of analyzing the results for variations in the TMMS costs, the maximum achievable reduction in the TCO has been displayed in Fig. 7 for different scenarios, whereas both the operational and investment costs related to TMMS are considered to be very low owing to the inherent uncertainties in them. The results shown in Fig. 7 can be interpreted as the investment profit margin that is the maximum fraction of the cost of a vehicle with a human driver that can be invested/paid on a TMMS such that ADS-DV remains profitable compared to a vehicle with a human driver. The profit margin arises from the elimination of driver salary (35%-55%), an increased vehicle-time on-road owing to the fact that no T. Ghandriz, et al. Transportation Research Part C 115 (2020) 102610 resting time is needed (0%-15%), an increased payload due to removal of the cabin (1.2%-5%), as well as from the optimized propulsion system (0%-20%), provided that both the vehicles are driven at a rate close to the optimum utilization rate as much as possible. It can be seen that the profit margin of ADS-dedicated BEHVs is larger than that of ADS-dedicated CHVs. Furthermore, larger vehicles reveal lower profit margins when employing ADS. It must be noted that the results obtained may change if a different parametrization is used. The competitiveness of BEHVs is sensitive to the vehicle utilization level, fuel or ICE efficiency, life time, discount rate, and prices of diesel fuel, electric energy, battery, and ADS-specific hardware, i.e., on-board equipment for object and event detection and response. The nominal cost of equipment has been provided in Appendix, whereas the sensitivity of the TCO to these parameters, with lower and upper bounds of 45%, is given in Fig. 8 for a tractor-semitrailer on a 160 km flat road. For the same vehicle size and road, the components of the TCO for different vehicle utilization can be seen in Fig. 9. In these figures, vehicle utilization refers to the maximum fraction of the yearly time when the vehicle is in operation, i.e., when the vehicle is on-road or performs LU or charging; it includes the minimum rest time of the driver, as set by European Commission regulation 561/2006. It can be observed that for 30% utilization ADS-DVs are more expensive than vehicles with human drivers, mostly owing to high TMMS costs. It can also be seen that the price of ADS-specific hardware exhibits a minimal effect on the TCO, because it constitutes only a small fraction of the purchase and operational costs. T. Ghandriz, et al. Transportation Research Part C 115 (2020) 102610 Moreover, it should be noted that TCO of a BEHV also depends on the fuel consumption of a CHV, which operates in a same transportation task as the BEHV, as seen in the upper left plot in Fig. 8. The reason is that, based on the assumptions of this study, the maintenance cost of BEHVs is proportional to that of CHVs which is proportional to their fuel consumption. Please refer to the paper by Ghandriz et al. (2020) for the sensitivity results for other vehicles and roads, together with the changes in the cost components. BEHVs can display a lower TCO if many vehicles can be employed in a fleet, whereby the cost per vehicle can be reduced by sharing the recharging station and LU infrastructure. In the case of employing 10 vehicles in a fleet, the ADS-dedicated battery electric tractor-semitrailer and the vehicle with a human driver can be competitive with their combustion-powered counterparts up to 320 and 160 km travel ranges, respectively, in case of 100% vehicle utilization, in contrast to a fleet comprising a single vehicle (shown in Fig. 3). Another scenario can involve excluding the cost of the charging infrastructure from the TCO, which will lead to more profitable BEHVs, as observed in Figs. 6 and 9, assuming that the size of the battery is not affected by the cost of the charging infrastructure. However, such a scenario cannot be considered to be realistic if no high-power charging stations are publicly available.
In the literature, e.g. Davis and Figliozzi (2013), Lee et al. (2013), Lebeau et al. (2015), Taefi et al. (2017), and Lebeau et al. (2019), usually, the large BEHVs with human drivers are not competitive to CHVs with human drivers. The main reasons are the high purchase cost of BEHVs, and that the driving range of the vehicles, studied in the literature, is usually about 160 km with 40%-60% vehicle utilization. The result presented in this study is in accordance with the literature. In this study, none of the BEHVs with human drivers, operating in a similar travel ranges and utilization as of the vehicles studied in the literature, are competitive to CHVs operating in the same conditions. However, in this study, BEHVs became competitive to CHVs in shorter travel ranges due to performing vehicle-infrastructure optimization, despite considering the cost of the charging infrastructure in TCO of BEHVs. Moreover, as larger the vehicle, the lesser is the difference between the annual TCO per unit freight transported of BEHVs and that of CHVs with human drivers, as can be seen in Figs. 4 and 6. The main reason is that charging infrastructure depreciation cost per unit freight transported reduces considerably in the case of a large BEHV, similar to the case when several small BEHVs are used in a fleet.
Finally, the performance of BEHVs can be further improved by considering additional incentives or the higher taxes imposed on CHVs through regulations. Furthermore, electric road systems and dynamic charging, whereby charging is possible on-road while driving, entirely facilitate the competitiveness of BEHVs, as reported by Alaküla and Márquez-Fernández (2017) and Fyhr et al. (2017). Moreover, further reductions in fuel and energy consumption can be achieved by implementing speed profile optimization on-road considering the topography, and within the maximum and minimum speed limits around the reference speed according to, for example, the works of Johannesson et al. (2015), Hovgard et al. (2018), and Torabi and Wahde (2018). A requirement for maintaining the set optimal speed is that the motion of the vehicle is not abstracted much by the surrounding traffic, such as driving in a dedicated lane, or by an "intelligent traffic management system" that controls the entire traffic, as reported by Milanes et al. (2012).

Conclusion
By implementing mathematical models together with optimum choices of vehicle-infrastructure through TCO minimization, this study identified those transportation scenarios, with or without human drivers, where utilizing BEHVs is more competitive compared to CHVs. Moreover, the study showed that ADS affect other vehicular systems, in particular, the optimum setup of the propulsion system of BEHVs. ADS lead to decreased TCO of between 27% and 46% for BEHVs and between 11% and 41% for CHVs, which render BEHVs profitable over longer travel ranges by a factor of four, compared to that of BEHVs with human drivers. Furthermore, the profitability of ADS-DVs as well as the competitiveness of BEHVs against CHVs has been demonstrated for different transportation scenarios. It was observed that ADS-dedicated BEHVs tend to exhibit lower optimal speeds than vehicles with human drivers. Moreover, the reduction in speed for safety reasons was shown to be less expensive to realize in ADS-dedicated BEHVs than in ADS-dedicated CHVs; in many scenarios, low speeds down to 60 km/h actually reduced the TCO.
Furthermore, owing to the uncertainty in the parametrization, sensitivity tests were carried out. Moreover, the maximum reduction in the TCO that can be achieved by adopting ADS at a very low cost of TMMS was presented for different vehicles and roads. The reduction in the TCO in ADS-DVs was mainly achieved by removing the driver salary (35%-55%), increasing the vehicle-time on-road (0%-15%), increasing the payload by removing the cabin (1.2%-5%), as well as by optimizing the propulsion system (0%-20%).
Consideration of many transportation scenarios with different road types and vehicle sizes resulted in the production of a large volume of data. The data revealed that in order to achieve profitable operation with zero emission, the propulsion hardware should be adapted to the use case of BEHVs, irrespective of whether a human driver is involved. All the produced data and figures have been provided in Ghandriz et al. (2020) giving practitioners valuable information on the feasibility and profitability of an intended freight transport operation involving automation and electrification for a given use case.
However, the results of this study are limited to the repetitiveness of transport operations on known roads. Even though sensitivity analysis of the vehicle utilization can provide a rough estimate of the TCO for non-repetitive assignments, the optimum propulsion hardware cannot be reused, unless it is designed for the worst-case scenario where the vehicle is intended to operate, which requires more study. Furthermore, this study simplified a transportation scenario to include only one vehicle type and two pickup and delivery nodes. Future studies shall involve hardware-infrastructure optimization of a fleet of vehicles of different types that are operating in a transportation network comprising more number of nodes, where the optimum propulsion hardware for each vehicle and optimum location of charging stations will be determined based on minimization of the TCO.

Acknowledgment
This work was supported by the Swedish national research program FFI managed by Swedish Energy Agency.

Appendix A. Optimization problems
The optimum vehicle-infrastructure can be found by minimizing the annual TCO per unit freight transported c t for a road with given length and hilliness, average vehicle reference speed, vehicle size, and plan that defines the driver role. Variable names with (~) on the top denote functions of other variables and input parameters, whereas all other variables denote either the given trajectories or input parameters that depend only on the design variables. The input parameters can be found in Tables B.6,B.7,B.8,B.9,B.10,B.11,B.12. The argument x, representing vehicle position along the road, together with other function arguments have been omitted from the equations for clarity of notations.

A.1. Conventional combustion-powered vehicles
The optimization problem was defined as follows for a CHV.  Ghandriz, et al. Transportation Research Part C 115 (2020)  where, = … a k n , 1 k sc denote the design variables of n sc dimensional space S CV , in which the range of k th dimension is specified by set S CV k , containing discrete choices of the design variable a k . The design variables of a CHV include the type of ICE, LU of the first unit/ semitrailer at the first node of the transportation task, LU of the second semitrailer (if any) at the first node of the transportation task, LU of the first unit/semitrailer at the second node, LU of the second semitrailer (if any) at the second node, the ranges of which are denoted by type LU LU LU , , , ice stTr ndTr stTr 1 ,1 2 ,2 1 ,1 , and LU ndTr 2 ,2 , respectively, as presented in Table B.5. Different types of ICE are described in Table B Ghandriz, et al. Transportation Research Part C 115 (2020) 102610 respectively; c c , tax insu , and c tmms denote the annual cost of taxes, insurance and TMMS, respectively. In Eq. (A.2), p F , f c and N t denote the prices of diesel fuel, and the fuel consumed during a round-trip (described by Eq. (A.10)), and the number of trips a vehicle performs per year, respectively. In Eq. (A.3), p d and t tr denote driver salary and round-trip time, which are described by Eq. (A.12). Eq. (A.4) describes the maintenance cost proportional to the fuel cost that is quantified by a proportionality factor c m . In Eq. (A.5), m m m m , , , gcm chass cab ice , and m obl represent the masses of a fully loaded vehicle (i.e., gross combination mass), vehicle chassis, cabin, ICE, and on-board lift, respectively. Eq. (A.6) gives the depreciation cost of hardware based on interest rate r, service life span n y , vehicle-infrastructure resale value R v (described by Eq. (A.8)), and price of the vehicle-infrastructure p , described by Eq. (A.7),including the chassis price p chass , price of cabin along with driver interfaces p cab , transmission price p trans , ICE price p ice , price of all the vehicle hardware required for ADS p ads , investment cost of LU p lu , and investment cost related to TMMS p tmms . Eq. (A.9) gives the number of trips per year, where T year stands for a time span of one year in the same unit as t tr , and < < u 0 1 represents vehicle utilization. It should be noted that the performance of a vehicle was simulated for an entire year.
The fuel consumed during a round-trip can be calculating using Eq. (A.10), where , , ice pgf f tor and P ice denote the maximum efficiency of ICE, energy per gram of diesel fuel, fuel density, travel time on road (described by Eq. (A.11)), and ICE power (described by Eq. (A.14)), respectively. In Eq. (A.11), x f denotes the overall distance covered in the round-trip and v is the vehicle speed at the distance x on road, described by differential Eq. (A.19), where ( ) denotes time derivative. In Eq. (A.12), t lu i , denotes the LU time and t res i , represents driver resting time. A driver must rest 45 min after every four hours of driving, according to European Commission Regulation 561/2006 (Eq. (A.13)), when considering a transportation task involving two nodes. It must be noted that a driver can rest during LU process. In Eq. (A.14), ctr denotes transmission efficiency and F prop denotes the total propulsion force on the vehicle tires, as described by Eq. (A.16). The ICE power is constrained between the minimum and maximum values, P ICE min , and T ICE min , stand for the transmission efficiency and the maximum and minimum ICE output powers and torques. Eq. (A.26) gives the friction brake force required to reach the reference speed during negative acceleration. It should be noted that, in these equations, the grip limit between the road and the contact patch of the wheel is not considered.
Finally, the constraints (A.27)-(A.29) ensure proper vehicle performance on-road. Constraint (A.27) requires a gradeability higher than a set value. Gradeability is the maximum grade on which a vehicle is capable of maintaining a set forward speed (e.g., 80 km/h). Constraint (A.28) ensures that a vehicle is capable of starting the forward motion on a given grade, which is referred to as startability. Constraint (A.29) guarantees that the acceleration capability of a vehicle is higher than a minimum set value. These constraints were evaluated by programming based on their definitions. Edgar et al. (2002), Sadeghi Kati (2013), Sadeghi Kati et al. (2014, Kharrazi et al. (2015)) provide further descriptions of the performance-based characteristics of heavy vehicles.

A.2. Battery electric vehicles
Similar to the case of a CHV, the optimization problem for a BEHV was defined as follows. T. Ghandriz, et al. Transportation Research Part C 115 (2020)  T. Ghandriz, et al. Transportation Research Part C 115 (2020)   k se denote the design variables of n se dimensional space S EV , in which the range of k th dimension is specified by set S EV k , containing discrete choices of the design variable a k . The design variables of a battery electric vehicle include the type of electric motor, number of electric motors, type of battery pack, number of battery packs, LU of the first unit/semitrailer at the first node of the transportation task, LU of the second trailer/semitrailer (if any) at the first node of the transportation task, LU of the first unit/ semitrailer at the second node, LU of the second trailer/semitrailer (if any) at the second node, and recharging power at the first and second nodes, and the ranges of these are denoted by type N type ,1 , and P ch,2 , respectively. Several battery packs connected in series form the vehicle battery. The elements of the vehicle-related design sets are described in Table B.5. In the cost function (A.30), N v denotes the number of vehicles, f tr denotes the transported freight per year (described by Eq. (A.34)), and c elec and c maint ev , denote the annual electric energy cost and the cost of maintenance (described by Eqs. (A.31)), respectively. In Eq. (A.31), p el and E el denote the price of electric energy and the electric energy consumed during a round-trip (described by Eq. (A.40)). Eq. (A.32) is similar to Eq. (A.3). Eq. (A.33) describes the maintenance cost of a BEHV proportional to the maintenance cost of a CHV of the same size that operates in the same scenario with a proportionality factor of 50%, according to Davis and Figliozzi (2013) The electric energy consumed during a round-trip can be calculated using Eq. (A.40), where t tor and P batt denote the travel time on road (described by Eq. (A.41)) and the power of battery packs (described by Eq. (A.42)), respectively. In Eq. (A.41), x f denotes the overall distance covered during the round-trip and v is the vehicle speed at the distance x on road (described by differential Eq. (A.48)), where ( ) denotes time derivative. In Eq. (A.42), P em denotes the useful power of electric motors (described by Eq. (A.44)), P em loss , denotes the energy loss of the electric motors (described by Eq. (A.59)), and P batt loss , represents the energy loss of the battery packs (described by Eq. (A.56)). The power of the battery packs is constrained between the minimum and maximum values, P batt min . The maximum and minimum powers and torque on wheels, as well as the friction force, are described using Eqs. (A.51)-(A.55). In Eq. (A.56), I batt represents the electric current in battery packs (described by Eq. (A.57)) when the battery nominal voltage V batt is known, R batt denotes the resistance of the battery packs and P cabHeat the average power used in heating the driver cabin (if any) described by equation ((A.58)). The average heating power was assumed to be proportional to the consumed power by a proportionality factor c heat . Eqs. (A.59) and (A.60) describe the power loss in electric motors, where k denotes a constant related to electric motor specifications and is the rotational speed of the electric motor. This model of energy loss corresponds to electric motor operation at the highest efficiency, wherein the gearbox is capable of selecting any gear ratio that is very close to the optimum value. Such a type of transmission can be referred to as continuous variable transmission.

= …
The state of charge (SOC) of the batteries SoC x ( ) must always be within the limits SoC min and SoC max , as specified by constraint Eq. (A.61). The SOC can be calculated using Eq. (A.62), wherein SoC x ( ) i denotes the SOC at the exit of node i x , denotes the distance traveled, and C batt denotes the total capacity of battery packs. The SOC at the exit of node i is given by Eq. Owing to non-convexity and non-smoothness of the constraints, these optimization problems were solved using stochastic optimization methods, in particular particle swarm optimization, as described by Wahde (2008).