Automatic Guided Vehicles Introduction Impacts to Roll-On/ Roll-Off Terminals: Simulation and Cost Model Analysis

Automatic guided vehicles (AGVs) have been successfully applied to cargo terminals to reduce operating costs and improve productivity. However, the focus was on container terminal operations. Ports with roll-on/roll-oﬀ (RORO) terminals still heavily depend on human resources for the loading/unloading processes. Work operations are aﬀected by human errors and safety issues. In particular, terminals where vehicles cannot be stacked pressure workers to handle cargo more rapidly, which induces more errors. In this study, we propose automating RORO terminal operations by using AGVs. We assessed the impact of AGVs on the productivity, cost eﬃciency, and environment. A series of simulation models was developed on the basis of the current loading system at an actual port to test the impact of AGVs. Then, we developed a cost model to analyze the economic beneﬁt of AGVs compared with the current loading system. The environmental beneﬁts were also analyzed. Results revealed that a system using 29 AGVs matched the productivity of the current loading system, and using more AGVs increased the productivity. For a given productivity level, the total operating cost of the AGV system was three times less than that of the current system over a 15-year period. The AGV system also showed great potential for improving the environmental friendliness of terminal operations. This is the ﬁrst study to propose automating RORO terminal operations to improve productivity and sustainability through AGV technology rather than human factors. AGVs are expected to become a good option in the future to address labor shortages and the “untact” era.


Introduction
Seaports are one of the most important elements of a country's economy and play an important role in international commerce and trade [1].
ey not only handle commercial cargo, but also are centers of economic activity [2]. Accordingly, port administrations are under great pressure to increase their productivity and compete with other ports around the world [3]. e increased competition in the port industry has encouraged the development of automated terminals to reduce operating costs and improve productivity, safety, and environmental sustainability [4]. e greatest proponents of automation in the port domain are container terminals [5]. In contrast, roll-on/roll-off (RORO) terminals are still heavily dependent on human resources because drivers have to move vehicles one by one.
Furthermore, vehicles in the terminal cannot be stacked, so workers are under pressure to complete tasks quickly in the given time [6]. is had led to human errors and safety issues [7,8]. In addition, the movement of thousands of vehicles inevitably produces a large amount of CO 2 emissions. is can be a problem because maritime sectors have been requested to find green solutions in the face of increasingly stringent environmental regulations [9].
A significant element of automated terminals is AGVs. Automated container terminals using AGVs have seen many benefits such as reduced costs and improved productivity, safety, and environment [5].
e AGV system was not considered in the RORO terminal before as navigating AGVs in car carriers may be more complex than operating an AGV just on a flat surface due to their complex layout with collapsible decks and steep ramp. First-generation AGVs were designed to navigate according to line-following principles using technologies such as embedded guide wires, paint stripes, magnetic tape, and laser guidance. Since then, the advent of new technologies has transformed how AGVs navigate and to what they can be applied. e development of intelligent autonomous vehicles (IAVs) and automated lifting vehicles (ALVs) has alleviated the limitations of previous AGVs that had to follow a fixed track [10,11]. IAVs can pick up/drop off cargo by themselves and can navigate without following any fixed track by using a wireless link with an intelligent virtual real-time simulator. Research has been increasing on the subject of unmanned vehicle navigation. In particular, many studies have focused on the development of unmanned forklifts that can navigate without the line-following principle [12,13].
e French start-up company Stanley Robotics recently developed an AGV capable of parking vehicles at airports [14]. eir AGV uses Global Navigation Satellite System, a camera, and Li-DAR-based simultaneous localization and mapping technologies for pathfinding, which is potentially applicable to any environment. erefore, such AGV system can now be applicable in the RORO terminal to load the cars from the yard to the RORO ship for the automation. e pure car and truck carrier (PCTC) currently occupies most of the RORO ship market. e loading process of a PCTC starts with each driver taking a vehicle onto the ship via the stern ramp [15]. Each driver has to wait 5 s before departing the yard to avoid colliding with the previous vehicle. Once drivers reach the parking lot, they leave the vehicle and head to the shuttle van. After the van collects 8-10 drivers, it drives back to the yard for the drivers to bring the other vehicles one by one. Employing AGVs changes several aspects of the current operation. First, the main resource for the loading process becomes the AGV. In other words, the human intervention in the loading process is reduced, and the number of resources can be increased and limited at any time as desired. Second, AGVs move around the terminal individually. While drivers have to wait for the shuttle van to fill up before they can leave the ship, AGVs can go directly to the yard to pick up the next vehicle. is eliminates the waiting time for each batch of drivers, and AGVs that have begun the loading process do not need to wait 5 s to avoid simultaneous departures. Additionally, the current loading system uses vehicles powered by fossil fuels for shipping cargo and transportation. In contrast, AGVs are powered by electricity, which should reduce CO 2 emissions.
We propose automating RORO terminal operations to improve port productivity and sustainability. In this study, we were focused on demonstrating the potential benefits from introducing them to current RORO terminals. We investigated the effect of using AGVs on the operation of RORO terminals with regard to productivity, cost efficiency, and environmental impact. e contribution of this study lies as follows. First, a detailed simulation model for the actual RORO terminal shipping process is developed. Second, this study proposes automating RORO terminal operation with AGV technology. ird, a detailed cost model is developed to estimate the total operational cost with different parameters for the AGVs loading system and the current loading system (i.e., car and van). Fourth, environmental impact from the operation of the vehicle in the terminal is evaluated. e current loading system has been highly dependent on drivers. erefore, developing the automated loading system using AGVs in the RORO terminal could significantly improve the current operating system and address the labor shortage. e rest of the paper is structured as follows. Section 2 reviews relevant materials and methods on RORO terminal productivity and the development of simulation models using Arena software. In the following section, we investigated the port performance with different numbers of AGVs to identify the optimal number of AGVs, followed by the results of data analysis from an actual terminal. Section 3 presents the simulation results and the cost model analysis on the overall benefits of the AGV loading system and environmental impact. Section 4 gives the conclusions.

Relevant Studies on RORO Terminal Productivity.
In the last few decades, few studies have been carried out on the operation of RORO terminals compared with other terminals so the relevant literature is scarce. However, the main factor for the RORO terminal operation is human resources.
is review focuses on literature relevant to RORO terminal productivity from two different perspectives: worker productivity and simulation studies.

Worker Productivity.
Workers are the main influence on the productivity of RORO terminals [16,17]. However, the human resources for RORO terminal operation have limited capacity for further improvement, which reiterates the necessity of automation. ere are many ways to improve worker productivity, and they are interrelated. Many studies have focused on the link between working hours and productivity [18][19][20]. Most revealed that the working hours are not proportional to the work productivity. Especially for skilled jobs such as stevedores, ensuring an effective working time is important for reducing the onset of fatigue, which can affect productivity. A RORO terminal typically has a 9 h workday with six break times per day. However, many RORO terminals are under competitive pressure with an increasingly global maritime trade [3,21], which has resulted in a lot of overtime. is has inevitably led to human errors during terminal operations [7,8]. More workers can be hired to reduce human errors, but labor overuse usually greatly increases costs and inefficiency [22].
Many studies have also noted concerns over worker safety and health at ports [23,24]. Major health concerns include fatigue, back pain, and headaches due to poor air conditions, the noisy environment, and lack of facilities [25]. Such occupational health and safety issues are a major reason for the reduced productivity at ports [26,27].
Over the past few decades, automation and robotics have been promoted as a solution to the aforementioned issues [28][29][30]. Many industrial sites have improved human factors and productivity through automation. A large volume of literature is available on automation in container terminals, but studies on automating RORO terminals are scarce. To the best of our knowledge, we are the first to suggest automating the loading system as a solution to improving the productivity of RORO terminals, rather than focusing on worker productivity. e following subsection introduces simulation studies to measure the productivity of the RORO terminal.

Simulation Studies.
As seen in the above section, the productivity of the RORO terminal highly depends on the worker's productivity, and it is difficult to be measured. So many researchers have experimented with simulation studies. In this subsection, relevant studies on the RORO terminal simulation are introduced.
Bottlenecks can have a significant impact on the performance of a port. To ensure the smooth flow of cargo, bottlenecks should be identified in advance and optimized to reduce congestion. is not only improves port productivity, but also prevents overuse of resources and overinvestment. Simulation approaches are generally used to optimize terminal operation and determine the optimum levels of investment and resources.
Demirci [31] used the simulation program AweSim to identify operational bottlenecks for Trabzon Port in Turkey.
e simulation model was constructed on the basis of realistic data related to port operation. In the full-capacity situation, loading/discharging vehicles were investigated as a bottleneck, and the number of vehicles was optimized under economic constraints. Adding vehicles improved the port performance and reduced the ship turnaround time by 8 days. e strength of the simulation model at analyzing the port performance was demonstrated.
Keceli et al. [21] used the software Arena to develop a simulation model for RORO terminal operation. ey identified waiting area 3 as a bottleneck and suggested that it should be enlarged. e simulation results demonstrated the usefulness of the simulation model and its potential applicability to other RORO terminals. e authors also discussed the importance of building a simulation model to predict the effects of any planned changes.
Muravev et al. [32] compared two software programs for discrete-event simulation (DES) modelling. ey modelled the operation of a RORO terminal with Arena and AnyLogic independently. eir results showed that the two software programs were practically similar. Small differences in the results were attributed to random numbers created by the different mechanisms of the programs. Experimental results indicated that Arena is suitable for simulation modelling of RORO terminals to optimize the system operation and identify failures in advance.
RORO terminal is the work site heavily reliant on workers' performance during the loading process. As can be seen from the review above, the reviewed simulation studies have addressed the general operating system, but the details of the loading system in the terminal were not studied. Also, past literature did not consider automating the operation in RORO terminal. erefore, in this study, we developed a series of simulation models to test the loading system in the RORO terminal and suggest the automation with AGVs for the first time.

Simulation Model Development
2.2.1. Target Port and Ship. As a case study, we considered port A, which is the largest automobile import-export gateway in South Korea. Figure 1 shows the layout of port A. Glovis Splendor is a PCTC, which is one of the most widely utilized vehicle carriers on major deep-sea trade routes with a carrying capacity of 7353 R/T. R/T is the largest number of standard-sized vehicles that Ro-Ro ships can load motivated from RT43, a 1966 Toyota Corolla [15]. e parking space in the yard was calculated as 6.39 m 2 per unit for 7353 vehicles considering clearance space. e actual parking space can differ depending on the vehicle type. However, we considered the standard vehicle size to calculate the largest number of vehicles that can be loaded on a ship, which is also called the nominal vehicle carrying capacity [33]. e parking space in the yard was calculated for 7352 vehicles and divided into blocks a-d.

Arrival Distribution.
e arrival distribution represents the arrival of entities as a discrete set of points. Because the 7352 vehicles in the yard are at different positions, the probabilities for arrival at the next point are different between the first and last vehicles. However, in most cases, the vehicles are parked in the yard in rectangular blocks, so the minimum and maximum probabilities for the first and last vehicles to arrive at the next point are clearly defined. erefore, we used a triangular distribution to define the transfer of 7352 vehicles from the yard to the vessel.

Vehicle Speed and Loading Strategy.
e vehicle speed was set in compliance with the Hyundai Glovis transportation and handling manual. Table 1 indicates that each section of the port has different speed limits, so different speeds were applied to each section. e AGVs had a maximum speed that was less than the speed limit, so we allowed them to operate at maximum speed in every section. We also reduced the safe following distance for AGVs considering their speed. e loading strategy was considered to reduce the queueing bottleneck; vehicles nearest to the vessel were prioritized for loading first.

Scenario Development.
A simulation model can be used to depict operation concepts that are not used in a real port. We developed scenarios to demonstrate improvements to the loading system. Two scenarios were considered with different path plans for the AGV loading system. In scenario 1, vehicles follow the same path as the current loading system: one-way movement within the external ramp. If vehicles entering from different sides meet at the ramp, the first one to enter the ramp takes precedence. Scenario 2 allows two-way movement, which means that vehicles can Journal of Advanced Transportation 3 bypass each other on the ramp. erefore, three different loading systems were compared: the current loading system, AGV loading system scenario 1, and AGV loading system scenario 2.

Arena Simulation Modelling.
Arena is a DES software that models real-world systems with a stochastic nature. Arena is a good option for simulating terminal operation. Numerous studies have used Arena to model container terminals because it can be used to analyze complex port systems [34,35]. Although few studies have considered RORO terminals, most also used Arena for simulations [21,32]. Arena simplifies the model building and overcomes mathematical limitations [36]. By running simulations at slow speed, the logic of the process can be checked, and discrete points where a large queue occurs can be identified [37]. In this study, we used Arena 14.0 for building, testing, and analyzing the simulation models. Because we were focused on evaluating the terminal performance, we used the module function "tnow" to measure the loading time.

Simulation Assumptions.
We made the following assumptions to reduce unnecessary details and compare the simulation models under the same conditions: (i) e loading/unloading places are identical, and the loading cargos are the same assuming that all 7352 standard size cars are loaded (ii) e surfaces of deck pillars are not considered because the space between pillars is large, and their surfaces are small (iii) e vehicles depart from the yard to the ship at the same time (iv) A specific stowage plan considering the balance of the ship is not detailed

Arena Simulation Models.
Two simulation models were developed and tested in Arena; the current loading system and the AGV loading system scenarios 1 and 2.
Before testing the simulation model of the AGV loading system, we confirmed the validation against the real data from the actual loading system. Because drivers are under pressure to complete the tasks quickly, the actual loading process is carried out at speeds beyond the speed limit; thus, we used our simulation model to calculate the standard loading time as specified in the manual. According to the real data from the Pyeongtaek Port, the average vehicle loads recorded 100 per hour. With an average vehicle loads of 95 per hour, we reported the result of the simulation's confidence interval at the 95 percent (�0.05) confidence level to assure results validity. e result shows that the simulation reproducing current loading system is valid and accurate. It hence can be used to investigate the impact of AGV loading system. e CREATE module is the starting point for entities in a simulation. e DISPOSE module is the entity's ending point. e ASSIGN module allows you to assign values to entities including entity type, variables, and attributes.
e BATCH module can group entities permanently or temporarily, and temporary batched entities must be split later using the SEPARTE module. e HOLD module places an entity in a queue to wait for a signal or a specified condition. Figure 2 shows the simulation model for the current loading system. e vehicles' arrival to the yard was represented by the CREATE module, which is located at the top of the figure as the first module. e DECIDE module divided the vehicles among the yard blocks a-d.
e SEIZE module restricted the number of drivers. Since each driver has to wait for 5 s before departing the yard to avoid colliding with the previous vehicle. e PROCESS module, which is located on the very right side in the figure, was used. e BATCH module simulated the collection of drivers by the shuttle van. After drivers left the van in the SEPARATE module, they started a new loading process with the RELEASE module until all vehicles were loaded. Figure 3 shows the simulation model for the AGV loading system. e 5 s rule was only implemented for the first group of AGVs. e BATCH module was removed because AGVs went back to the yard individually. e SEIZE module was used to control the number of AGVs for the sensitivity analysis.   Journal of Advanced Transportation 2.4. Sensitivity Analysis. As discussed previously, port operation with vehicles involves congestion at certain points. Increasing the number of resources will not continuously improve the productivity. us, sensitivity analyses help with properly assessing the impact of key variables. is is done by varying the value of a particular variable while fixing the value of other variables. Park et al. [15] tested several AGVs using mainly inside car carrier. Increasing the number of AGVs reduced the total loading time, but it also increased the waiting time in the bottleneck. e result deviated slightly from the result in this study as the previous study did not address the AGVs traffic at the yard. In addition, the previous study did not detail the impacts of introduction such as the economic and environmental benefits.
Nguyen and Kim [38] performed a sensitivity analysis on the number of ALVs and buffer capacity at a port. eir results showed that increasing the number of ALVs improved port productivity and reduced the total delay in quality control operations. In contrast, the buffer size had a larger effect with fewer ALV. ey used mathematical models such as heuristic algorithms for optimization and commented that simulation experiments could be used to solve problems in more dynamic environments. In addition, they only addressed the dispatch of ALVs without integrated scheduling.
Pjevčević et al. [39] used Arena to develop a simulation model of the container handling process and investigated the effects of dispatching rules and the number of AGVs. eir results showed that the number of served   Journal of Advanced Transportation containers increased with the number of AGVs. However, increasing the number of AGVs also decreased the active rate of AGVs. is study indicated the importance of appropriately choosing the number of AGVs and dispatching rules. ey did not consider the economic benefit of AGVs because their focus was on the dispatching rule and number of AGVs.
Kavakeb et al. [40] used the software Flexsim CT to develop a simulation model for a container terminal in Europe. ey then performed a sensitivity analysis on the buffer size and number of IAVs. ey developed a cost model to estimate the total cost-benefit of using IAVs.
eir results showed that IAVs can significantly improve port performance compared with trucks despite their slower speed and had a much lower total operating cost. ey also used an advanced vehicle dispatching strategy to improve the port performance further. However, although the IAV used in their study is environmentally friendly, they did not investigate the environmental benefits.
e aforementioned literature demonstrates the usefulness of sensitivity analysis for evaluating the impact of AGVs on terminal operation. In our study, we performed a sensitivity analysis to optimize the number of AGVs, and we investigated the economic and environmental benefits of the optimized AGV loading system compared with those of the current loading system.

Simulation Results.
e results of the simulation models were compared for the current loading system, AGV loading system scenario 1, and AGV loading system scenario 2. For the AGV loading systems, the number of vehicles varied from 10 to the maximum number. Figure 4 shows the simulation results for the current loading system. e longest waiting time was attributed to the batch delivery process, where drivers were collected by the shuttle van. e secondlongest waiting time was attributed to the 5 s rule, which was applied to all vehicles to avoid collision with preceding vehicles. e total loading time was 93,099.70 s, which implies that the actual work time charge can be almost three days. As mentioned in Section 2.3.2, these results are similar to the actual data of port A.

Current Loading System.
Based on the performance of the current loading system, the AGV loading system had a target time of 93,099.70 s to complete the loading process. To investigate the impact of AGVs on port A, a sensitivity analysis was performed, where the number of AGVs varied from 10 to the maximum. e maximum number of AGVs that affected the loading time was 40. Table 2 presents the results of AGV loading system scenario 1. e simulation results are reported the same way as in Figure 4 and we tested the use of 10-40 AGVs. e average waiting times within the external ramp and the total loading times were compared. For scenario 1, the minimum number of AGVs to meet the target performance was 31.
Above the maximum number of 40 AGVs, the productivity remained the same, whereas the waiting time within the external ramp increased.

AGV Loading System Scenario 2: Two-Way Movement.
In the current loading system, vehicles do not pass each other on a ramp, even though the ramp has enough space for two vehicles. is is to avoid collisions between vehicles travelling fast. Moreover, vehicles returning to the yard and entering the ship are less likely to meet because drivers are sent back to the yard by a shuttle van. However, the slower speed of AGVs caused a large queue to form within the external ramp with AGV loading system scenario 1. us, we developed an alternative path plan for AGVs to reduce the congestion. Because advanced sensors allow autonomous vehicles to detect objects much faster and more accurately than human drivers can [41], AGVs should be able to bypass each other on the ramp. In addition, the maximum speed of the AGVs in this study was less than 10 km/h, so collisions are less likely to happen. erefore, in scenario 2, vehicles were allowed to cross within the ramp. is was implemented in the simulation model by doubling the resources used for the external ramp. e results were retrieved in the same way as in scenario 1. By reducing the average waiting time within the ramp, the maximum number of AGVs that could be employed in scenario 2 was increased to 70. e difference between the two scenarios was large when more AGVs are used. In scenario 2, with 29 AGVs, the total loading time was 93107 s reaching the performance of the current loading system revealed in Subsection 3.1.1, and its average waiting time was 1.914 s. As can be seen in Table 2, in scenario 1, the use of 31 reached the performance of the current loading system, and its average time was 11.852 s, which was 9.938 s larger than scenario 2. e maximum use of scenario 1 was 40 AGVs, and its loading time was 87394 s with 54.296 s average waiting time. e maximum use of scenario 2 was 70 AGVs, and its loading time was 44869 s with 55.365 s average waiting time.
is implies that scenario 2 can achieve much higher productivity with lower waiting time. Figure 5 compares the results of scenarios 1 and 2. Scenario 2 increased the overall productivity and reduced the minimum number of AGVs to meet the target performance to 29. e average waiting time at the ramp was much less than that in scenario 1, which indicates less congestion. e results of scenario 2 demonstrate that the congestion caused by the slower speed of AGVs can be solved by applying a suitable path plan. e simulation result from either scenario shows that the use of AGV can reduce the work time charge less than the current loading system when used more than 29 AGVs. In particular, the use of the maximum number of AGVs in scenario 2 showed the possibility to reduce the working hour time to less than one day. Indeed, the impact of adopting AGVs is larger in terms of cost efficacy. erefore, in the following subsection, we compare AGVs and current loading system based the on the total capital and operational cost in a 15-year periods.  Figure 4: e simulation results for the current loading system. e longest waiting time was attributed to the batch delivery process, where drivers were collected by the shuttle van. e second-longest waiting time was attributed to the 5 s rule, which was applied to all vehicles to avoid collision with preceding vehicles. e total loading time was 93,099.70 s. As mentioned in Section 2.3.2, these results are similar to the actual data of port A. Table 2: Results for AGV loading system scenario 1. e simulation results are reported in the same way as in Figure 4 and we tested 10-40 AGVs.    Journal of Advanced Transportation

Optimal Number of AGVs.
To compare the total costs of the current and AGV loading systems, the minimum number of AGVs that matched the productivity of the current loading system was identified. With three gangs of stevedores, the current loading process takes 93,108.30 s. e AGV loading process required a minimum of 29 AGVs following scenario 2. erefore, we considered the optimal number of AGVs to be 29.

Cost Model Comparison.
A cost model was developed to identify the economic benefits of AGVs compared with the current loading system (CLS). A robot generally has a life cycle of 80,000-100,000 h, which is 10+ years. e cost model was used to calculate the total operating costs of port A for a 15-year period with the optimal number of AGVs. e 15-year period was selected to consider the 10year lifetime of AGVs and another 5 years with new AGVs. e first factor was the capital cost of the vehicles. For the AGV loading system, the vehicle capital cost accounted for a significant part of the total cost. At the time of submission, the cost of AGVs was not available. We assumed that the unit cost was approximately €150,000 based on expert speculation. is is a conservative estimate, and actual AGVs are not expected to cost as much. For the current loading system, the main capital costs are for people and the shuttle van. e capital cost for the van was set to €114,000, and the capital costs of the drivers were calculated in terms of wages. To consider vehicle failure and a charge rotation, the minimum number of AGVs was increased 20% to include six additional AGVs as spare. e total energy costs of the current and AGV loading systems differed because they consumed different types of energy. e shuttle van was powered by diesel, whereas the AGVs use electricity. In addition, they consume different amounts of energy because they have different total travel distances. e total energy costs of the shuttle vans and AGVs per loading process were calculated as follows: where Dl van is the liters of diesel consumed per 100 km travelled by the van, El AGV is the electricity consumed per 100 km travelled by the AGVs, p d is the price per litre of diesel, and p kwh is the price per kilowatt-hour of electricity. t van is the travel distance(km) of the van, t AGV is the travel distance(km) of the AGVs, E van is the total energy cost of the van, and E AGV is the total energy cost of the AGVs for each loading process. e next intermediate parameter is the costs for workers' wages. e annual salary can be calculated by multiplying the wages per loading process with the number of loading processes per year. e wages per loading process were calculated as follows: where h is the total working hours per loading process of the 7352 R/T size vessel, p svd is the hourly pay for a stevedore, p AGV is the hourly pay for an AGV operator, W svd is the total wages for stevedores per loading process, and W AGV is the total wages for AGV operators per loading process. e aforementioned intermediate parameters can be used to calculate the annual operating costs of the current and AGV loading systems. We calculated the annual salaries for workers and annual energy costs for vehicles as follows: where n s is the number of loading processes per year, n svd is the number of stevedores, n gang is the number of gangs, T AGV is the total number of AGVs � n AGV + n AGV−spare , s van is the total service cost per van, and s AGV is the total service cost per AGV. e operating costs for the 15-year period were calculated from year 0 and with the inflation rate I: e vehicle capital costs in year 0 were calculated as follows: where p van is the price per shuttle van, p AGV is the price per AGV, Ll r is the land lease rate per square metre, sac st is the surface area per charging station, and c st is the cost of the charging station and parking garage. e capital costs for the AGVs include the costs for the charging facilities. e capital costs of the vehicles for the 15-year period were calculated as follows: Vehicles were assumed to have a lifespan of 10 years. R cls t is the total cash flow of the current loading system in year t, and R AGV t is the total cash flow of the AGV loading system in year t. e total cash flow of year t is the summation of the operating cash flow and vehicle capital cost:

Journal of Advanced Transportation
To extend the cost model analysis to an environmental perspective, the cost of CO 2 emissions was considered. We calculated the CO 2 emissions from the vehicles and converted them to monetary values. e CO 2 emissions from the current loading system are mainly from the operation of vehicles and the van. In contrast, the CO 2 emissions from AGV operation were zero, but the emissions from electricity production needed to be considered. Holmberg and Ali [42] calculated the CO 2 emissions from the internal combustion engine and electric vehicles per kilometre. eir work was used to calculate the total CO 2 emissions from the two loading systems: where co car 2 is the CO 2 emissions from a vehicle per kilometer, co van 2 is the CO 2 emissions from the van per kilometer, co AGV 2 is the CO 2 emissions from an AGV per kilometer given that the electricity generation mix is coal, emiss clp is the total CO 2 emissions produced by the current loading system, emiss AGV is the total CO 2 emissions produced by the AGV loading system, and t car is the travel distance of vehicles per loading process. e estimated CO 2 price has varied greatly as climate change has become an increasing concern [43]. Recent studies on CO 2 abatement have estimated the average CO 2 price to be €40-€70 [44,45]. e intermediate parameters in equations (14) and (15) can be used to calculate the total CO 2 costs per year with the two loading systems: where co 2 pr is the CO 2 price per metric tonne. en, the total CO 2 costs for the next 15 years can be calculated as follows: e discount rate r was included because analysts forecast the CO 2 price to rise in the future [46]. Tables 3  and 4 present the values of the initial and intermediate parameters used in the cost model. e intermediate parameters were calculated using equations (1)-(4) and (14)- (17). Table 5 presents the cash flows for the 15-year period as calculated using equations (5)- (13). Figure 6 compares the cash flows in each year for the current and AGV loading systems. In year 0, the AGV loading system had higher costs than the current loading system because of the initial capital costs of the AGVs. In year 1, the AGV loading system had a much lower operating cost than the current loading system. e main difference was attributed to the costs of wages. Although AGVs need some operators to control the system, they replace a significant part of the human resources required by the current loading system. In year 10, the AGV loading system sees another increase in cost, because new vehicles are purchased. Over the 15-year period, the total cash flows for the current and AGV loading systems were €53,558,339 and €17,021,863, respectively. e total cash flow for the AGV loading system was almost three times less than that for the current loading system despite the initial capital costs for vehicles. Figure 7 demonstrates the environmental impact of using AGVs. To clarify the environmental benefit, the monetary value of CO 2 emissions was not included in the Table 4: Intermediate parameters calculated using equations (1)-(4), (14) and (15    operating costs but was shown separately. Over the 15year period, the total CO 2 emissions from the current and AGV loading systems were 1276 and 453 m 3 /t, respectively. ese values are equivalent to €86,993 and €30,904, respectively.
ese results show that the AGV loading system can be operated at a much lower cost than the current loading process and also has great environmental benefits.

Conclusions
We proposed automating the operation of RORO terminals by using AGVs to reduce human factors and improve performance. e impact of using AGVs was investigated in terms of productivity, cost efficiency, and environmental impact. A series of simulation models were developed, and a sensitivity analysis was performed to optimize the number of AGVs. A cost model was developed to analyse the economic benefits of the AGV loading system with the optimal number of AGVs compared with those of the current loading system. For a 15-year period, the total cost of the AGV loading system was almost three times less than that of the current loading system. Finally, the environmental impact of the AGV loading system was estimated in terms of CO 2 emissions and demonstrated to be significantly less than that of the current loading system. is study has some limitations. First, we only tested one type of vessel (i.e., Glovis Splendor). However, there are many different ships of different sizes, so the dispatch strategy of AGVs and their optimal number may differ. Also, the actual size of the cars to be loaded differs from the standard size vehicle which we tested here, so the total number of vehicles that can be loaded probably differs from the nominal vehicle carrying capacity [15]. To implement AGVs in the real world, the type of vessels may be needed to classify AGV dispatch strategies, and the study must be developed to be applied to more varying vehicles.
erefore, further case studies need to be tested. Second, costs incurred by vessels at the port were not included in the cost model and environmental impact analysis. Indeed, vessels incur much higher costs at ports than at sea, and emissions released by vessels at port are a continuing concern [47,48]. Regardless of the vehicle capital costs and efficiency, the most effective cost reduction strategy may be minimizing the loading time.
ird, more sophisticated vehicle dispatch/schedule strategies can be considered in the future. Developing a more effective schedule to release vehicles to the vessel can reduce the waiting time and improve the port performance. We may investigate these topics in the future and present our results in subsequent papers. To the best of our knowledge, we are the first group to propose automating RORO terminal operation with AGV technology. Our study demonstrated that the potential impact is significant, so AGVs are expected to become a good alternative option in the future for addressing labour shortages and the "untact" era.

Data Availability
e data used to support the findings of this study are included within the article.

Conflicts of Interest
No potential conflicts of interest were reported by the authors.

Authors' Contributions
Conceptualization was done by S.K. and S.P.; methodology was done by S.P.; validation was done by S.P. and J.H.; formal analysis was done by S.P.; data curation was done by S.P. and J.H.; original draft preparation was done by S.P.; reviewing and editing were done by S.Y. and S.K.; visualization was done by S.P., S.Y., and J.H.; supervision was done by S.K.; project administration was done by S.K.; funding acquisition was done by S.K. All authors have read and agreed to the published version of the manuscript.