Traffic Simulation of Automated-Driving Ground Support Equipment at Tokyo International Airport

Abstract

In Japan, the shortage of airport ground-handling personnel has become a serious concern with the growing demand for aviation, necessitating improvements in operational efficiency. Accordingly, the expectations for automating aircraft ground support equipment (GSE) vehicles are growing to achieve labor savings. This study evaluated the changes in GSE traffic flow performance (travel speed, travel time, and number of stops) through traffic simulations under various scenarios of automated-driving GSEs penetrating the entire airport restricted areas. We simulated the traffic flow at Tokyo International Airport using the observation data of each GSE driving through the airport. Simulation results indicated that GSEs experience a reduced travel speed in some vehicle corridors when automated-driving GSEs, considering the safety risks associated with existing automated technology, run at lower speeds to ensure reliable driving performance. Consequently, the total travel time of the GSEs for the entire airport increases. These results confirm that the penetration of automated-driving GSEs can be facilitated by implementing measures, such as developing technology for reliable driving performance or operational rules at intersections to enable these vehicles to run at a speed equivalent to that of manned GSEs and to prevent speed reduction and travel time increase in airport vehicle corridors.

1. Introduction

In recent years, global airport passenger traffic has increased markedly and is projected to reach ~22 billion by 2053 [1]. In Japan, aviation demand continues to increase [2], and further increases are projected because of the robust growth in the number of foreign visitors [3]. This can be attributed to the inbound policy of the Japanese government, which has a target of reaching 60 million foreign visitors by 2030 [4]. However, labor market shortages, high turnover, and frequent employee transfers in many countries [5] make it difficult for ground-handling (GH) companies to hire additional GH personnel. Thus, GH personnel shortages have become an increasingly important issue since the onset of the COVID-19 pandemic. In response to a sharp decline in demand in the aviation industry during the lockdown, the GH personnel left the workforce in Japan. However, demand in the aviation industry has recovered since the pandemic, making it crucial for GH companies to secure the required human resources [6]. Almost 5–10% of all major delays can be attributed to GH, underscoring its critical impact on airport performance [7]. As shortages in GH personnel can affect the quality of airport ground services, airports must optimize their resources and enhance the efficiency of their GH operations to foster sustainable air transportation [8,9]. Consequently, the importance of efficient GH operations has increased. Therefore, expectations are growing with respect to automating aircraft ground support equipment (GSE) vehicles and achieving labor savings through automated driving of these vehicles. Although this trend is particularly prominent at large-scale airports, where numerous personnel are required for GH, medium and small airports are also affected because they are adversely influenced in the air transport system by delays, crews running out of their shifts, and flights being canceled [10].

Considering the similarities among procedures, aircraft configurations, and clustered interfaces, the automation of GH operations is an achievable goal and a promising research area. Automation can help ensure more effective resource management and is a new research direction for GH tasks. Therefore, robotic and automated-driving vehicle technologies have been developed [11]. Automated-driving GSEs have already been implemented at various airports worldwide [12], which include field operational tests (FOTs) of automated-driving vehicles at Gatwick (United Kingdom, UK) for employee transport, at Heathrow (United Kingdom, UK) for luggage dollies, and at Dallas Fort Worth International (United States, US), Christchurch (New Zealand), and Tokyo International (Haneda Airport) (Japan) for passenger transport buses [12]. Ref. [12] reported that Gatwick Airport in the UK tested automated-driving electric vehicles in 2018 and used them to transport employees within the airport. A new method of luggage transport using an automated-driving vehicle called dolly was developed, and the testing started in March 2019 at Heathrow Airport in the UK. The Dallas Airport in the USA is implementing a new program to transport passengers using an automated-driving vehicle called Easy Mile Mobility Assistant (EMMA), which has a capacity of 12 passengers and does not require a driver although a safety employee remains onboard. At Christchurch Airport in New Zealand, FOTs for an automated-driving electric vehicle with 15 passengers were introduced in 2017 [12]. According to the Society of Automotive Engineers, there are six levels of automation ranging from Level 0, where systems can issue warnings, to Level 5, where the driver/passenger only determines the destination and sets off a fully automated-driving vehicle [13]. Most automated-driving vehicles are at Level 3 or 4. Level 3 is a conditional driving automation, whereby the system performs all driving tasks; however, the driver must respond appropriately to system intervention requests when operations become difficult. At Level 4, the system performs all driving tasks in a limited domain and responds when operations become difficult. The semi-automatic aircraft towing and pushback vehicle TaxiBot received official certification for Boeing 737 dispatch towing from the European Aviation Safety Agency and Israel Civil Aviation Authority in 2014 [14]. Remote-controlled aircraft towing and pushback vehicles, such as the Mototok, are operated at several airports in Japan. Some airports are aiming to fully automate their GSE operations. For example, the Royal Schiphol Group at Schiphol Airport (Amsterdam), is endeavoring to fully automate its airside operations by 2050 [15]. Anticipating the spread of automated-driving GSE, the International Air Transport Association (IATA) added a description on automated-driving GSE to the IATA Airport Handling Manual, published every year [16].

In Japan, the penetration of automated-driving GSEs into restricted airport areas is being promoted under the leadership of the government in cooperation with the private sector [17]. Since 2018, FOTs have been conducted for towing tractors or passenger transport buses to enable them to drive automatically within the airport restricted areas of Sendai, Narita, Haneda, Chubu, and Saga airports [18]. Technical verification of Level 3-equivalent automated-driving was achieved through FOTs conducted up to FY2020. Additional FOTs are being developed for Level 4-equivalent automated-driving GSE by FY2025.

In addition to the implementation of FOTs, the development of a common infrastructure and operational rules are being studied for airports that adopt automated-driving GSEs [18]. Common infrastructure requires infrastructure facilities to be shared among GH operators in airport restricted areas to provide support for automated-driving GSEs and ensure safe and smooth driving. This includes signal facilities at intersections, float management systems to manage traffic flow in airport restricted areas, and priority roads for automated-driving GSEs. Operational rules are commonly imposed on GH operators to support automated-driving GSEs that cooperate with surrounding traffic and to secure safe and smooth traffic within airport restricted areas, including the priority/non-priority entry orders at intersections and the criteria for permitting automated-driving GSEs at airports. In December 2024, the first version of the “Common Infrastructure Guidelines,” which define the functions and requirements for various common infrastructures, was formulated [19]. For the operational rules, the “Driving Guidance for Automated-Driving Vehicles in Airport Restricted Areas” formulated in 2020 has been sequentially revised.

Understanding the change in traffic flow caused by automated-driving GSEs within airport vehicle corridors and estimating the effectiveness of common infrastructure and operational rules in advance are crucial for studying and implementing these strategies. However, the penetration of automated-driving GSEs is still being tested both in Japan and overseas, and the data available on previous examples and actual penetration are insufficient to serve as a reference. Obtaining all the necessary data directly by conducting various FOTs at airports during operation is not practical; therefore, conducting a GSE traffic simulation would be useful for evaluating traffic flow changes in penetrating automated-driving GSEs by setting various scenarios in advance without conducting large-scale FOTs. Some previous studies simulated a part of the GSE traffic flow considering the real airport status to optimize specific GH operations, such as airport bus scheduling [9] and several tasks except for baggage transportation [20].

To the best of our knowledge, there are no existing simulation-based studies that model full-airport-wide GSE traffic using real observation data of each GSE traveling, as opposed to specific GSEs, for evaluating the entire airport traffic flow when penetrating an automated-driving GSE. It is possibly because GSE activities occur in restricted areas of airports. As described in previous studies [21,22,23], the airside area is a secured area where tracking GSE vehicles is focused on specific targets, and it is difficult to accumulate a dataset for an entire airport. The traffic rules for GSEs traveling within airport restricted areas are set by each airport administrator, and each GSE travels between terminal buildings, aircraft parking spots, and other facilities such as GSE parking areas in accordance with these traffic rules. Furthermore, in normal condition, GSE dispatch is planned in advance based on scheduled and estimated times of aircraft arrival and departure and then updated in real time, following changes in aircraft schedules. Each GSE driver must respond to the updated dispatch plan in a timely manner. GSE have different traffic routes between dispatch destinations due to congestion or clearances to be respected against aircraft, buildings, and obstacles. Consequently, determining GSE travel patterns within airport restricted areas and estimating GSE origin to destination (OD) trips for traffic simulation input data are difficult. Therefore, using the observed data of each GSE as simulation input seems to be an effective approach for evaluating the traffic flow performance of an entire airport.

Given this background, this study aimed to evaluate the changes in GSE traffic flow performance parameters—such as travel speed, travel time, and number of stops—when automated-driving GSEs penetrate the entire airport restricted area, based on observed OD trip data of GSE vehicles within the airport as input to simulation. Furthermore, the performance at intersections controlled by traffic signals and/or priority rules was evaluated. Similar to many studies about mixed-flow conditions, we considered different scenarios of automated-driving GSE penetration ratio, technical level of driving performance, and operational rules at intersections to estimate changes to the traffic flow performance in terms of operation efficiency and safety. The simulation was carried out by taking as a reference the restricted area of Haneda Airport, the busiest airport in Japan, for which improvement in GH efficiency due to automated-driving GSEs are mostly expected and FOTs are actively conducted.

The remainder of this paper is organized as follows: Section 2 summarizes previous studies. Section 3 describes the model-setting method for the GSE traffic simulation conducted in this study. Section 4 presents the simulation results and discusses them. Finally, Section 5 concludes the paper.

2. Literature Review

Ground-handling operations include all operations performed around a parked aircraft on the apron to handle passengers and cargo, as well as to supply facilities and supplies (i.e., food, fuel, cleaning, water) [7,24]. The various GSEs operating in airport restricted areas interact with each other [20], following flight schedules. An established line of research investigates the adverse impact of disruptive events, such as lacking GH resources, on airport turnaround operations [10]. The automation of GH operations, including GSE driving, was introduced in the scientific literature [25] thanks to the availability of reliable online information on GH services. Airport collaboration decision-making tools such as Eurocontrol support the automation of GH [26] to enhance the efficiency/reliability/punctuality of airport operations. Studies on methods that enabled task automation included online GSE fleet allocation [7] and frameworks for multiagent optimization based on multiagent route planning and task assignment of refueling, catering, baggage handling, water, and lavatory services around the aircraft [11].

Several FOTs exist for penetrating automated-driving GSEs [12,27], such as automated-driving towing tractors and passenger transport buses, primarily at Haneda and Narita airports in Japan [27]. These FOTs provide useful data on automated-driving GSE driving status, the surrounding environment, and number of overrides to help organize automated-driving GSE-driving performance issues that require improvement at airports [27]. Semi-automatic aircraft towing and pushback vehicles have also been studied for aircraft taxiing procedures to reduce the environmental impact [14].

Research on efficient GH management, including monitoring GSE movements and improving GSE scheduling, was conducted from several perspectives prior to considering automation of GH services. Some studies [28,29] focused on improving the management of apron movement control by observing the GH on the apron. Other studies [21,30], monitored the aircraft and GSE vehicles on an apron through AI-based visual detection. These studies utilized deep learning and computer vision methods to detect and analyze the key ground service behaviors of aircraft parked on the apron, using a single fixed camera image [21,30]. Another study [31] guided unmanned vehicle movements by setting barcode markers on the apron surface within a limited test field instead of across the entire airport. In another study [22], aircrafts were detected instead of GSE vehicles, thereby estimating the aircraft movement context using images captured by a running GSE vehicle. However, these monitoring studies were conducted only in a limited area within the apron, or to an extent that could be covered by a specific GSE.

The vehicle routing problem for efficient GSE scheduling determines how to organize and call certain vehicles for GH services at a series of demand points [32]. As the constraints increased, the vehicle routing problem with time windows (VRPTW) and vehicle routing problem with pickups and deliveries for the GH vehicle scheduling was considered, and a more efficient algorithm was developed to obtain the optimal solution [32]. Describing a single goal is difficult for further studies, which led to a multi-objective vehicle routing problem [32]. The optimized objects had the shortest overall travel distance, lowest overall transportation cost, lowest service time cost, and the smallest number of vehicles required [33]. Single- and multi-vehicle scheduling problems (VSPs) have also been studied. The targets for single VSPs include de-icing vehicles [34], shuttle buses, refueling vehicles, and cargo trucks [32]. For one type of VSP, the VRPTW was established by considering the shortest total vehicle mileage, lowest operating cost, or minimum delay as the objective function, wherein constraints were designed as a heuristic [35] or calculated from the exact solution algorithm [36]. For multiple VSPs, Ref. [20] integrated decision-making by considering these entanglements instead of separately scheduling various tasks and GSEs to optimize the entire GH process. This helped to realize the two objectives of minimizing the total turnaround completion time and waiting time before starting an operation for the overall reduction in the corresponding timeframe. This study targeted several GH tasks of unloading and loading baggage, deboarding, boarding, catering, cleaning, fueling, potable water, and toilet services, except for baggage transportation, at two Spanish airports, using each set of flight schedule data, considering the order and interaction between each GH task. Ref. [9] developed a systematic framework for simultaneously optimizing different objectives, such as ground movement problems, runway scheduling, airport bus scheduling, and fuel consumption using an integrated multi-objective approach, which enabled the studying of airport efficiency, environmental assessment, and economic analysis. Targeting airport bus scheduling at a real airport in Qatar, they used a single day real flight schedule, considering the sum of the total aircraft taxiing time and runway delay. Another study [32] improved scheduling responsiveness to information changes by accounting for the uncertainty in flight arrival times. Optimization methods such as fuzzy uncertainty [37,38] and robust optimization [39,40,41] are used to solve uncertain problems when decision-makers cannot trust the model. Ref. [42] reviewed numerous previous studies on aircraft gate assignment and classified them based on the objectives, number of optimizations, or solution method. Owing to the promotion of carbon tax concept, carbon emissions have gradually become the main objective of optimization [43]. One study optimized both greenhouse gas emissions and airport traffic efficiency by examining a mixed-integer linear programming mathematical model to realize optimal placement of electric aircraft towing vehicles and taxiing routes to minimize jet fuel use [44]. Another study on mixed fuel and electric vehicles optimized GSE vehicle placement by minimizing the sum of time and energy and emission costs [43].

However, no previous study has reproduced the full airport-wide traffic flow in vehicle corridors via traffic simulations using real observation data of each driving GSE, to evaluate the changes in traffic flow from the perspective of efficiency or safety for automated-driving GSEs in a real airport.

3. Methodology

This section describes the setup of the model, input data, and scenarios for GSE traffic simulations. Haneda Airport, the busiest in Japan, with an annual passenger volume of 87.41 million and a domestic market share of 26.3% (2019 figures) [45], was the target airport in the simulations. As of June 2019, of the ~7000 GSE vehicles registered at the airport, ~40% were self-propelled and the remaining non-propelled were towed by self-propelled GSEs [46]. Automated driving is expected to improve GH efficiency at Haneda Airport, and FOTs are being actively conducted.

3.1. Simulation Model

The simulations described in this study used ‘AVENUE’ ver.5 (abbreviation of ‘an Advanced and Visual Evaluator for road Networks in Urban arEas’) [47], which is a microscopic traffic simulation model developed by i-Transport Lab. Co., Ltd., Tokyo, Japan [48]. AVENUE reproduces second-by-second vehicle movements using the hybrid block density method (HBDM) [48,49] and dynamically updates the route plan of each vehicle to its destination by considering traffic congestion and road network regulations. The details of the road network configuration, including lane usage in turning movements, vehicle type, streamlines at intersections, and signal phases controlling the streamlines, were also considered [50]. AVENUE was validated by the Japan Society of Traffic Engineers through engineering tests [51], and the results are disclosed via the Traffic Simulation Clearing House [52]. As AVENUE is a commercial product, it is widely used in Japan to enable planners to make more informative and better traffic management decisions for implementing traffic management strategies, such as traffic signal control systems to reduce congestion [53]. Among the many commercial microscopic simulators available, e.g., Vissim [54] or Aimsun [55], AVENUE was selected because one of the authors is a developer and it offers the advantage of allowing modifications at the source code level to express various GSE vehicle movements around aircrafts.

In the fundamental diagram (FD) of traffic flow, HBDM, which is characterized by a few parameters such as capacity and free-flow speed, has a theoretical basis and may offer the advantage of easy calibration [48]. As plain HBDM assumes that all vehicles obey the FD and ignores the heterogeneity of vehicle movement, AVENUE was extended to treat a mixture of multiple vehicle classes with different maximum speeds and accelerations [56]. In this study, we applied this extension to express the slower running speed and duller acceleration of GSE vehicles.

3.2. Input Data for the Simulation Model

3.2.1. Map Data of Haneda Airport

The GSE vehicle corridors in the restricted areas of Haneda Airport were modeled as inputs to AVENUE. The study area covered almost the entire area south of Runway B at Haneda Airport and included the main ODs where GSE traffic is frequent, mainly around all passenger terminal buildings, maintenance area, and cargo areas, as shown in Figure 1. Each OD in this study was defined using the 53 nodes in Figure 1. A list of the 53 nodes is provided in Appendix A Figure A1.

3.2.2. GSE OD Trip Data

The GSE OD trip data of traffic flow were input into the simulation model. We used the moving trip data created by Ref. [46]. Their dataset is based on the measured data of 2234 self-propelled GSE travels in the Haneda Airport restricted area in late November 2019. We selected late November 2019 as the representative period for the simulations. Haneda Airport recorded one of the highest passenger volumes in the world over the last 20 years, except between 2020 and 2022, because of the COVID-19 pandemic [57]. This airport is extremely busy throughout the year, with no significant difference in the number of departures and arrivals per month [45].

Figure 2 shows the time series of the number of moving trips on 21 November 2019; each time point pertains to the start of a moving trip. The number of moving trips changes with time, suggesting that GSE traffic volume increases or decreases in accordance with the peak hours of aircraft arrivals and departures at Haneda Airport. Therefore, the target hours for the simulation were set from 6:00 to 20:59 on 21 November 2019, for a total of 15 h during the peak hours of aircraft arrivals and departures.

For the GSE OD trip, the movement of each GSE vehicle follows the GSE dispatch plan in accordance with the departure and arrival of the aircraft because GH operators allocate the required types and number of GSE vehicles based on the size of the aircraft and whether it is a domestic or international flight. According to the dispatch plan, GSE vehicles travel to designated aircraft parking spots, considering the work time before and after the aircraft enters and leaves the spot. Therefore, the GSE OD trip data for the simulation must reflect real traffic, including total traffic volume at Haneda Airport. Consequently, we used a moving trip after expansion estimation based on the measured traffic volume at a certain point in the airport. The expanded moving trips were randomly extracted from measured discrete moving trips on a trial basis. The total number of moving trips during the target hours was 23,018. The OD table of the moving trips between the 53 nodes in Figure 1 input into the simulations is provided in Appendix A Figure A2.

3.2.3. Vehicle Specifications and Driving Performance Values of GSE

The body size and running speed for the different types of GSE inputs to the simulation model are listed in

Table 1. Specifications and driving performance settings for different types of GSEs.

Symbol	Vehicle Type	Body Size		Running Speed		Speed Limit for Haneda Airport (km/h)
		Length (m)	Width (m)	Maximum Running Speed (km/h)	Maximum Acceleration (m/s2)
WT	Aircraft towing vehicle/Towbarless aircraft towing vehicle	9.1	2.9	30	0.7	30
HL	High-lift loader, etc.	12.1	3.8	14	1.1	15
BL	Belt loader	7.7	2.2	25	1.1	15
CT	Cargo truck/Unit load device transport truck	6.1	2.2	30	1.1	30
FL	Food loader vehicle	11.8	2.5	30	1.1	30
TT	Towing tractor, etc.	11.8	1.4	20	1.1	15
SC	Servicer	5.4	2.1	30	1.1	30
PW	Water supply truck	5.0	2.1	30	1.1	30
LS	Sewage truck/Drainage truck	7.5	2.2	30	1.1	30
BUS	Ramp bus/Large ramp bus/Passenger transport bus, etc.	11.5	2.5	30	1.1	30
MB	Minibus/Shuttle bus	6.3	2.1	30	1.1	30
PS	Passenger step vehicle	9.0	3.4	20	1.1	30
Other than the above		5.0	2.1	30	1.1	30

. These vehicle specifications were based on the results of a survey of GSE manufacturers and previous studies on road traffic [58]. The maximum running speeds were set with reference to the speed limit in the restricted area of Haneda Airport, as indicated on the right-hand side of Table 1. However, the maximum running speed values of high-lift loader (HL) and passenger step (PS) vehicle were set lower than the speed limits in the reference survey of GSE manufacturers because they work inside an aircraft parking spot. The maximum running speed value of towing tractor (TT) was set higher than the speed limit in reference to the GSE driving condition survey conducted in Ref. [46]. All GSE running speeds were set uniformly for the input data in the simulation model for the GSEs to accelerate and reach the “maximum running speed,” as shown in Table 1. As with other microscopic simulators, acceleration is calculated according to a formula that uses “maximum running speed” and “maximum acceleration” as parameters and is determined based on the running speed.

3.3. Scenario Settings

The scenarios for the simulations were set for the types of vehicles to be automated, penetration ratio of automated-driving GSEs, running speed and acceleration of automated-driving GSE as the technological level of automation, and intersection priority entry for the automated-driving GSE as the operational rules of the airport. A list of these scenarios is provided in

Table 2. List of scenarios of automated-driving GSEs.

	Base	Scenario 1	Scenario 2	Scenario 3		Scenario 4
Vehicle type to automate	None	TT, BUS, and MB
Penetration ratio of automated-driving GSEs	0%	10%	50%	100%		100%
Maximum running speed of automated-driving GSEs	-	Slow				Fast
	-	BUS/MB: 20 km/h				BUS/MB: 30 km/h
	-	TT: 15 km/h				TT: 20 km/h
Priority for intersection entry of automated-driving GSEs	-	Without		Scenario 3-1	Scenario 3-2	With
				Without	With

. The vehicle types to be automated included TT, passenger transport bus (BUS), and minibus (MB), which are specifically considered for automation penetration through FOTs in Japan. The penetration ratios of the automated-driving GSEs were set to 10, 50, and 100%, assuming a gradual progression of automation from Scenario 1 to 4. In this study, we originally set the penetration ratio as 10, 50, and 100%: 10% as the near future scenario, considering that upgrading many manned GSE vehicles to automated-driving vehicles for GH companies in Haneda airport would not be easy; 50 and 100% as far future scenarios to clearly evaluate the difference in simulation results between scenarios. In Scenarios 1, 2, and 3, the maximum running speed for automated-driving GSEs were set to slow, i.e., 20 km/h for BUS/MB and 15 km/h for TT, to ensure reliable driving performance, considering the safety risks associated with existing automated technology. Although the maximum accelerations of those automated-driving GSEs were the same as those of the manned GSEs indicated in Table 1, the achieved accelerations were more moderate because of their slower settings of maximum running speeds, to be determined based on the running speed as mentioned in Section 3.2.3. In Scenario 4, they were set to fast, equivalent to those of manned GSEs. For intersection priority entry, the automated-driving GSEs were set to enter an intersection without priority in Scenarios 1, 2, and 3-1 and with priority in Scenarios 3-2 and 4. The “with/without priority“ scenario follows the gap acceptance model to prioritize the entry of a non-signalized intersection. Gap acceptance is the threshold time gap between pausing and entering an intersection. That is, the vehicle approaching from a yielded direction first stops at the entrance of the intersection and estimates the time gap of the prior vehicles, dividing their distances from the intersection by their running speeds. It then judges whether it has been yielded by comparing the smallest time gap with the threshold value. The yielded direction is set either to a minor road at the intersection where major and minor roads cross or to a vehicle seeing another vehicle approach from its left-hand side (in Japan) at the intersection treating all approaches as equal. This is the case even if two automated-driving GSEs approach each other simultaneously. In this study, the “with priority“ scenario was simulated by shortening the gap acceptance to 2.5 s for automated-driving GSEs, which is half of the 5 s set for manned GSEs. This is assuming that the automated-driving GSEs can smoothly enter an intersection, thereby yielding to each other through inter-vehicle communication with other GSEs because automation technology would be sufficiently developed. In contrast, in the “without priority” scenario, the gap acceptance of automated-driving GSEs was set to a larger value of 10 s on a trial basis according to the FOTs in Japan. This is assuming more risk-avoiding judgment of automated-driving GSEs than manned GSEs against oncoming or right–left-turning vehicles because the automated-driving performance was still developing. Note that the sight distance of approaching vehicles here is enhanced up to the adjacent intersection without considering sight obstacles in real life. The automated-driving GSEs were set without an overtaking function in Scenarios 1 and 2, assuming that the automated-driving performance was still developing and set with an overtaking function in Scenarios 3 and 4, assuming that automation technology would be sufficiently developed. GSE vehicles can only overtake other vehicles in two-lane corridors, such as between nodes 8 and 10, nodes 8 and 20, nodes 20 and 24, and nodes 29 and 33.

The computation time for one case (15 h of simulation from 6:00 to 20:59 for the entire study area) was ~5–6 h. We used a computer with an Intel^® Core™ i7-8700K 3.7 GHz six-core 12-thread processor (Intel, Santa Clara, CA, USA) and 64 GB RAM connected to the network with ~23,000 links (directional arcs) and to the GSE vehicles with ~23,000 trips to run AVENUE. Furthermore, we arranged the network not only at vehicle corridors in the simulation model but also at setup zones around the aircraft at each parking spot. This resulted in a longer computation time for a single case, which significantly increased the number of routes for which options were selected.

4. Results and Discussions

4.1. Simulation Results

The method described in Section 3 was used to evaluate the changes in traffic efficiency and safety for GSE operations in an airport restricted area by setting up several scenarios assuming the gradual penetration of automated-driving GSEs. We calculated and compared the average travel speed, total travel time, and total number of stops of the GSEs in the simulation of each scenario, as described in Section 3.3.

4.1.1. Average Travel Speed

Figure 3 shows the average travel speeds for each scenario from 8:00 to 8:59 on 21 November 2019. The average travel speed is defined as the average speed of all GSEs that traveled each segment of the vehicle corridor during the relevant time and is considered an indicator of efficiency of the GSE operation. These diagrams show the average travel speeds obtained by coloring each segment of the vehicle corridor. The blue corridors have higher speeds than the red lower-speed corridors. In Scenarios 3-1 and 3-2, the average travel speed was particularly slow compared with that of the Base case. In these scenarios, the reductions in the GSE speed are more apparent than those in the Base case, especially in segments of the vehicle corridor around the passenger terminal building where GSE traffic is heavy [46]; there is 10–15 km/h of the average travel speed in this area, which indicates that automated-driving GSEs with low speeds can affect the overall speed of the traffic flow, including that of manned GSEs. In Scenario 4, the running speed and acceleration of automated-driving GSEs were set to be equivalent to those of manned GSEs; there is no such reduction in the average travel speed compared with that of the Base case. In Scenarios 1 and 2, the average travel speed gradually decreased with an increase in the penetration ratio of the automated-driving GSE; however, the effect was less significant than that in Scenario 3.

4.1.2. Total Travel Time

Figure 4 shows the total travel time for each scenario. The total travel time is the sum of the travel times of all GSEs in the entire study area during all periods covered by the simulation. The total travel time is considered an indicator of the efficiency of the GSE operations at the airport. Scenarios 1, 2, and 3-1 increase the total travel time by 3, 12, and 22%, respectively, compared with that of the Base case. Thus, the total travel time increased with an increase in the penetration ratio of the automated-driving TT and BUS/MB. In Scenarios 1, 2, and 3, the running speed and acceleration of automated-driving GSEs were set to lower values than those of manned GSEs. An increase in the number of automated-driving GSEs with low speeds is expected to increase the total travel time of the entire airport. In Scenario 4, the running speed and acceleration of automated-driving GSEs were set to be equivalent to those of manned GSEs, and the total travel time was almost the same as that of the Base case although the penetration ratio of automation for TT and BUS/MB was 100%, as in Scenario 3. In addition, in Scenarios 3-2 with the intersection priority entry of automated-driving GSEs, the total travel time was reduced by ~1% compared with that in Scenario 3-1. The operational rule that provides intersection priority entry to automated-driving GSEs can have a slight effect on reducing the total travel time; however, it is not as effective as increasing the running speed and acceleration of automated-driving GSEs.

Furthermore, the total travel time of the manned GSEs was confirmed from the breakdown of the automated/manned GSEs in Figure 4. The total travel time of manned GSEs is considered an indicator of the labor-saving effect of GSE operations at the airport. The total travel time of manned GSEs in Scenarios 3 and 4, where the penetration ratio of automation for TT and BUS/MB is 100%, was reduced by half compared with that of the Base case. The results indicate a reduction in work hours and labor savings for GSE drivers. The total travel time of the manned “Others” in Figure 4 increased by ~11% in Scenario 3 compared with that of the Base case because the automated-driving GSEs run at a lower speed. Conversely, in Scenario 4, the total travel time of the manned “Others” remained almost unchanged compared with that of the Base case, and all the travel times for automated-driving GSEs can help save labor.

4.1.3. Total Number of Stops

Figure 5 shows the total number of stops for each scenario. The total number of stops is the sum of the stops for all GSEs in the entire study area during all periods covered by the simulation. The number of stops represents the number of times each GSE applied brakes to stop when entering an intersection or approaching another GSE traveling ahead and the number of times each GSE slowed down to match the speed of another GSE traveling ahead in a congested area, remaining below 1 km/h for 3 s or more. The total number of stops is considered a surrogate indicator of the safety of the GSE operation at the airport because it can be regarded as an opportunity to brake to stop or slow down to avoid a collision with the leader GSE. The total number of stops in Scenarios 1, 2, and 3-1 were reduced by 2, 4, and 6%, respectively, compared with that of the Base case. As shown above, the total number of stops decreased in Scenarios 1, 2, and 3-1 with an increase in the automation of the TT and BUS/MB. In Scenario 3-2 with intersection priority entry for automated-driving GSEs, the total number of stops was reduced by 7% compared with that of Scenario 3-1 and by 13% compared with that of the Base case. An operational rule that provides intersection priority entry to automated-driving GSEs can have a certain effect on reducing the total number of stops.

In Scenario 4, the total number of stops increased compared with that in Scenario 3-2 and is almost the same as that in the Base case although intersection priority is given to automated-driving GSEs, as in Scenario 3-2. In Scenarios 1, 2, and 3, the running speed and acceleration of automated-driving GSEs were set to lower values than those of manned GSEs, and the number of automated-driving GSEs with slow speed increases the opportunity for dense convoys in the airport. However, in Scenario 4, the running speed and acceleration of automated-driving GSEs were set to be equivalent to those of manned GSEs, and the GSEs are expected to travel in a relatively scattered manner in the airport because they do not reduce running speed and acceleration. The total number of stops is higher in Scenario 4 because scattered traffic has lower instances of possible entries and more stop opportunities at intersections than dense convoys.

4.2. Verification of the Reproducibility of the Simulation

The simulated values of average travel times were compared with the measured values for some specific segments, referred to as “target segments”, to verify the reproducibility of the simulations. The measured values were obtained from moving trips based on observational data from the airport. Each average travel time is the average time of all the GSEs traveling along each target segment. We selected four target segments, A, B, C, and D, as indicated in Figure 6: different segments as a representative with high GSE traffic volumes with lengths of ~1–3 km from Terminals 1, 2, and 3, and the area connecting the terminals in the airport restricted area. Some segments had differences in travel time and length between the outbound and return routes caused by one-way routes or right/left-turn directions at the intersections. Therefore, the reproducibility of each of the four target segments was confirmed for both outbound and return routes. For example, there is a one-way corridor along the Terminal 1 building; therefore, GSEs traveling from south to north along this building must detour around nodes 2, 4, and 6, as depicted in Figure 6.

The gap between the simulated and measured values of the average travel time for each target segment is given by

$$\epsilon =\left|{\displaystyle \frac{t’}{t}}-1\right|\times 100\left(\mathrm{\%}\right),$$

(1)

where $t$, $t’$, and $\epsilon$ represent the measured value of the average travel time for a specific segment, simulated value of the average travel time for a specific segment, and the gap between the simulated and measured values, respectively.

Figure 7 shows the reproducibility of the target segments: (a) Segment A, (b) Segment B, (c) Segment C, and (d) Segment D. Segment A between nodes 9 and 3, illustrated in Figure 6, is a target segment from the Terminal 1 area; the outbound route of Segment A is 1272 m in length from node 9 to nodes 8, 7, 5, and 3. The return route is 1351 m in length from node 3 to nodes 4, 5, 6, 7, 8, and 9. The return route runs from south to north and is longer than the outbound route because the GSEs must detour around points 4 and 6. The simulated values reproduced the trend of the time variability of the measured values for both the outbound and return routes; however, there was a certain gap between the simulated and measured values. The gap varied based on the time of day, from 8.8 to 22.8% (~21–49 s) for the outbound route and from 51.3 to 61.9% (~104–121 s) for the return route. Segment B between nodes 16 and 50 is a target segment from the Terminal 2 area. The outbound route of Segment B is 2129 m in length from node 16 to nodes 15, 14, 12, 11, 49, and 50. The return route is 2129 m in length from node 50 to nodes 49, 11, 12, 14, 15, and 16. The simulated values were consistent with the measured values across many time intervals for both the outbound and return routes. However, a certain gap exists between the simulated and measured values in certain time intervals, and the simulated values were sometimes lower than the measured values. The gap between the simulated and measured values varied based on the time of day, from 0.6 to 16.1% (~2–61 s) for the outbound route, and from 0.6 to 16.8% (~2–80 s) for the return route. Segment C between nodes 30 and 31 is a target segment from the Terminal 3 area. The outbound route of Segment C is 997 m in length, from node 30 to nodes 29, 28, and 31. The return route is 997 m in length, from node 31 to nodes 28, 29, and 30. The simulated values were almost consistent with the measured values during all target hours for both the outbound and return routes. The gap between the simulated and measured values ranged from 0.1 to 8.7% (~0–16 s) for the outbound route and from 0.1 to 8.4% (~0–15 s) for the return route. Segment D between nodes 16 and 30 is a target segment from the area connecting the terminals. The outbound route of Segment D is 3725 m in length, from node 16 to nodes 17, 21, 20, 8, 53, 38, 37, 36, 33, and 30. The return route is also 3725 m in length, from node 30 to nodes 33, 36, 37, 38, 53, 8, 20, 21, 17, and 16. Note that there is a long but narrow tunnel between nodes 38 and 53. The simulated values were consistent with the measured values across many time intervals for the outbound route and reproduced the trend of time variability of the measured values for the return route. However, a certain gap exists between the simulated and measured values. The gap varies based on the time of day, from 0.1 to 6.2% (~0–40 s) for the outbound route and from 12.3 to 19.5% (~71–113 s) for the return route.

For the four target segments, the simulated values reproduced the temporal variability trends of the measured values. The gap between the simulated and measured values varied within a range of 0.1–61.9% (~0–121 s) based on the target segment or time of day. The simulated values almost always exceeded the measured values. This gap is attributed to some real GSE vehicles traveling faster than the maximum running speed indicated in Table 1, which was the upper limit of the GSE running speed set in the simulations. There was a certain degree of gaps, especially for the return routes of Segments A and D. The reason for this gap for Segment A is that the simulated travel speed is reduced by TTs which may slow down at the entrance of aircraft parking spot to enter in and may delay the following vehicles on the single lane. The penetration of TTs is relatively high, ~30–40%, and many of them running the section around nodes 4 and 6 on the returning route may enter the spots to serve aircrafts, causing the travel time to increase. However, in a real situation, TTs may more smoothly enter spots and have less effect on the following vehicle speed; thus, the real travel speed on this route would possibly be higher. The reason for this gap for Segment D may be that the tunnel operation rules between the simulation and real condition are different. In reality, GSE vehicles must stop temporarily or check the alarm display/traffic light at each entrance of the tunnel to secure safety entry such as tunnel clearance, which was likely why GSE vehicles from the outbound route waited longer at that time. However, the simulation cannot precisely reproduce this situation; we decided to stop every GSE vehicle at each entrance for a time interval 120 s to express the expected waiting time to ensure tunnel clearance. The simulated values were temporarily below the measured values for Segment B because a real GSE vehicle may take more time than in the simulation to travel the route between nodes 11 and 49 across the GSE underground passage, as indicated in Figure 6, stopping temporarily and running to keep a large distance from another GSE traveling ahead.

4.3. Discussion on Simulations

In Scenarios 1, 2, and 3, GSEs experienced a reduced travel speed in some vehicle corridors, especially when automated-driving GSEs run at lower speeds to ensure reliable driving performance, considering the safety risks associated with existing automated technology. Consequently, the total travel time of the GSEs for the entire airport also increased. These results indicate that the penetration of automated-driving GSEs can be facilitated if some measures are implemented for automated-driving GSEs to develop technology for ensuring reliable driving performance or by setting operational rules to enable running at the same speed as the manned GSEs to prevent GSE speed reduction and travel time increase in airport vehicle corridors.

In Scenario 3, for automated-driving GSEs running at a slower speed than manned GSEs, there was a slight increase in the total travel time of manned GSEs, as indicated by “Manned Others” in Figure 4. This indicates that not all travel time spent by automated-driving GSEs can achieve a labor-saving effect. In Scenario 4, the total travel time of manned GSEs remained almost unchanged compared with that of the Base case, where automated-driving GSEs can travel at a speed equivalent to that of a manned GSE with intersection priority entry. This indicates that the entire travel time spent by automated-driving GSEs can have a labor-saving effect.

In Scenario 3-2, the total number of stops was reduced where the operational rule prioritized the intersection entry for automated-driving GSEs. This indicates that the operational rules established for automated-driving GSEs at intersections can contribute to improving the safety of traffic flow across the entire airport. Scenario 4 maintained the total number of stops in the Base case. This indicates that airport traffic safety was maintained at the same level as before the penetration of automated-driving GSEs when automated-driving GSEs run as fast as manned GSEs and enter intersections smoothly with priority.

Thus, the results indicate that the change in traffic flow caused by automated-driving GSEs in the entire airport restricted area is almost zero when automated-driving GSEs travel as rapidly and smoothly as manned GSEs, as in Scenario 4. However, the change in traffic flow caused by the overtaking functions for automated-driving GSEs in Scenarios 3 and 4 was not confirmed by the simulations. Although the reduction in travel speed in some vehicle corridors and increase in total travel time in Scenario 3 were confirmed, they could not be confirmed for Scenario 4 compared with the Base case. This is probably attributed to the fact that two-lane corridors where GSE vehicles can overtake are limited in the airport.

5. Conclusions

This study conducted traffic simulations using observation data from each GSE traveling in the airport, to evaluate the changes in GSE traffic flow performance (such as travel speed, travel time, and number of stops) when automated-driving GSEs penetrate the entire airport restricted area. This resulted in GSEs experiencing a reduced travel speed in some vehicle corridors, especially when automated-driving GSEs run at lower speeds to ensure reliable driving performance, considering the safety risks associated with existing automated technology. Consequently, the total travel time of the GSEs for the entire airport increased.

These results confirm that the penetration of automated-driving GSEs can be facilitated by developing technology to enhance driving performance or setting operational rules at intersections to enable automated-driving GSEs to run at speeds equivalent to those of manned GSEs and to prevent speed reduction and travel time increases in airport vehicle corridors. Quantitatively verifying the changes in traffic flow in terms of efficiency and safety across the entire airport when automated-driving GSEs penetrate the entire airport through simulations is effective. Simulations under various scenarios enable a preliminary evaluation of the effects of common infrastructure and the operational rules to be implemented at airports without large-scale FOTs. These findings are expected to benefit aviation authorities, airport managers, automated-driving GSE vehicle operators, and vehicle manufacturers. The simulations can be used to verify the changes in each GSE operating condition at each vehicle corridor and intersection.

Regarding the limitations of this study, in the simulations, the OD trips added for the expanded estimates were discrete OD trips, randomly extracted from the observed GSE OD trip data on a trial basis. Random expansion is required to ensure realistic traffic volume simulations. However, this random method was not considered in terms of statistical reliability or validated in terms of its ability to reproduce airport traffic. In GSE OD trip data, the movement of each GSE vehicle, input into the simulation, must reflect the real traffic at Haneda Airport. Hence, the OD trip data input into the simulation model should be confirmed to ensure that they reflect the GSE dispatch plans and actual service paths of all GSE vehicles. In this study, we used the OD trip data created by the GSE driving condition survey data; however, this was not sufficiently verified to be consistent with actual service paths of all GSE vehicles. When OD trip data are not consistent, supplementary estimation methods should be considered to ensure that OD trip data reflect the actual service paths of all GSE vehicles. Thus, future research should focus on identifying the actual service paths of all the GSE vehicles. In addition, the OD trip data could not distinguish whether each trip of each vehicle type was related to departure or arrival flights because the data were based on the observed raw GSE OD trip data from the GSE driving condition survey. According to our survey of GH operators, there are trips that involve moving between multiple spots because one task starts directly after completing another. In such cases, it would be difficult to completely distinguish whether each trip pertains to the departure or arrival flights.

All the GSE in the simulations accelerated at the same value and reached the same maximum running speed, whereas in reality, departures from these conditions may occur. Further studies are required to determine the running speed and acceleration, considering the variations based on the actual driving conditions of each GSE. Therefore, the simulation results must be evaluated based on a relative comparison between scenarios rather than the simulated values themselves, such as the average travel speed and total travel time. Regarding the intersection priority entry of automated-driving GSEs, we set their gap acceptance to be short in the simulations. In the simulation model, controls such as stopping other manned GSEs when automated-driving GSEs approach intersections are desirable to reflect intersection priority entry more appropriately. Aircraft ground movements were not explicitly covered in the simulations and were only implied using the GSE OD trip data based on the GSE driving condition survey because the measured GSE driving conditions imply aircraft ground movement at the airport. However, aircraft ground movement should be explicitly included in the simulations to evaluate GSE vehicle pauses and route detours caused by aircraft congestion in service lanes where taxiways and vehicle corridors intersect. Therefore, aircraft ground movement data should be added to the input data to explicitly reflect aircraft movements in the simulations. These are the next challenges to overcome in order to improve the simulation method proposed in this study.

The methodology presented in this study can be applied to simulate changes in GSE traffic flow when automated-driving GSE penetrates other airports. It is essential to consider each airport’s specific features, such as the layout of facilities, traffic rules governing GSE vehicles, and the actual driving conditions of GSE vehicles because these features can differ for each airport. However, the methodology is applicable to other airports not only for the AVENUE simulation model but also in the preparation of simulation input data, such as map data, GSE OD trip data based on the observation survey, GSE vehicle specifications and driving performance values, and scenario settings. Additionally, the evaluation method is applicable to GSE traffic in the entire airport for the purposes of efficiency and safety.