1 Introduction

1.1 Background

Freight transport, as a major industry in the European Union (EU), has a huge negative impact on the environment and society. According to forecasts, freight transport activities will increase by 30% by 2030 and more than double by 2050, leading to an increase in negative impacts (ITF 2021). To achieve a sustainable, smart, and digitised transport system of the future, the European Commission has launched a Sustainable and Smart Mobility Strategy (EC 2020). Doubling rail freight by shifting freight to rail and combining other measures to reduce emissions by 90% is expected to make the EU carbon neutral by 2050 (EC 2019). To decarbonize freight transport, the Rail Freight Forward (RFF) initiative has committed to increase the share of rail transport by 30% by 2030 (CER and RFF 2020). To increase its market share, improve services to customers, and boost economic growth, the rail freight sector should transform its operations by accepting the continued adoption of innovative solutions as part of the digitization of rail transport. In 2014, the Shift2Rail Joint UndertakingFootnote 1 was established by the EU to carry out activities related to the transformation process towards a digitalized and automated rail freight system.

According to the RFF initiative, one way to achieve the modal share goal is through autonomous train operation (ATO) (CER and RFF 2020). The main advantages of ATO are regularity, a different operating cost structure, higher capacity, energy savings, and the potential to reduce operating costs. The implementation of ATO technology is differentiated by the degree of automation, train type, technology, and component. In terms of automation level, ATO is divided into automation grade 1 (GOA 1), GOA 2, GOA 3, and GOA 4 (Jansson 2020) (see Table 1).

Table 1 Grade of automatisation
Full size table

All ATO GoAs can apply to passenger or freight trains. In terms of technology, ATO can include communication-based train control (CBTC), European Rail Traffic Management System (ERTMS), automatic train control (ATC), and positive train control (PTC). In terms of components, ATO includes components such as camera, accelerometer, odometer, tachometer, and others (Akshay and Supradip 2019; MarketResearch 2021). The vision S2R JU was to improve the attractiveness, cost structures, efficiency, and sustainability of rail freight's transport based on smart and environmentally friendly rail freight equipment and technologies. Since 2016, S2R has initiated freight automation activities in the form of three demonstration projects under Innovation Program 5 (IP5). In the Freight ATO project, the main objective was to demonstrate the applicability of the technology to rail freight (Kahl et al. 2018). The role of ATO in ensuring more efficient and effective operation of rail systems is highlighted in the vision for the target rail system (ERA 2020) and by the flagship area targets in the Europe's Rail work programme (Europe'sRail 2021). Among other innovations resulting from the Europe's Rail Joint Undertaking and its Master Plan, the Ben@Rail project (Shift2Rail 2021) considers ATO as part of the needs of rail stakeholders and end users.

1.2 Aim and scope of the paper

Digitization has been a top priority for all economic sectors in the EU in recent decades, and railways are no exception. The main objective of the railway sector is to take advantage of all the opportunities offered by digital transformation in order to provide efficient and attractive services to customers (Scordamaglia 2019). Despite the wave of digitization in the railway sector and some long experience with autonomous operation, especially on urban rail lines, the application of ATO on main lines or high-speed lines is still limited. Nevertheless, numerous theoretical studies and field trials are being conducted to test the feasibility and suitability of ATO implementation (Yin et al. 2017). Therefore, the question of whether ATO can be implemented on a long-distance train line or on a nationwide network is increasingly being raised (Emery 2017). Recently, the signalling manufacturer AŽD Praha demonstrated the first autonomous train on the Čížkovice—Most line in the Czech Republic and plans to transfer the development and testing of ATO to other lines (Clinnick 2021). The Finnish autonomous freight train development project is also entering the testing phase, and the autonomous train will be ready for operation on the tracks by 2023 (Papatolios 2021).

Based on the research question in Emery (2017), this study aims to identify and elaborate on the opportunities and challenges for the deployment of ATO in freight rail operations. In recent years, there has been significant effort and progress in implementing the ATO. In 2018, nearly a quarter of the world's metro systems have at least one fully automated line in operation. ATO technology has been applied to many newly established urban transit lines, such as the Paris Métro, London Underground, Beijing Subway, and Tokyo metro (Yin et al. 2017). New automated metro lines are currently introduced in Barcelona, Spain. Similar projects for fully automated subways are being implemented in other countries (Briginshaw 2019). Currently, there are already 64 fully automated metro lines worldwide, most of them in Asia. Asia leads the world with 50% of the km of fully automated metro lines in operation, especially thanks to the opening of five new lines (in Korea, Malaysia, and China) in the last 2 years. Europe remains in second place with 30%, followed by North America and the Middle East with 11% and 8%, respectively. However, in 2023, the number of automated metro kilometres worldwide will triple, with most of the growth occurring in China (UITP 2018). Due to the numerous benefits of ATO, different levels of automation are also being tested on main lines to prove that ATO can provide all the benefits. The benefits of ATO have been proven through trials around the world and the results are summarised in Poulus et al. (2018). The progress of ATO adoption is expected due to the various benefits it brings, such as improvement in safety, reduction in operating costs, potential elimination of human errors with full automation, reduction in energy consumption and emissions generated, better capacity utilisation, improved reliability of rail services, and others (Wang et al. 2016; Singh et al. 2021).

In freight rail, the first steps towards automation were taken by Deutsche Bahn in 1996 with the development of a self-propelled single wagon. Subsequently, a next-generation train concept for freight transport with autonomous single cars was presented by the German Aerospace Center (DLR) (Müller 2020). The first driverless train for freight rail was deployed by Rio Tinto in Australia starting in June 2019. The second project of an autonomous freight train by SNCF was tested in the same year and is expected to be completed in 2022 (Singh et al. 2021). A recent study supporting the digitization of all rail applications evaluated the electronic systems on board a freight wagon (Losada et al. 2022).

While there are studies on the introduction of ATO in passenger rail, with the exception of a few attempts, there is no comprehensive discussion on the introduction of ATO in rail freight. Nevertheless, it may be of interest to all stakeholders to know the extent of the costs and benefits of ATO in rail freight, as well as an overall picture of the potential of certain GoAs. Therefore, the objective of this paper is to analyse the feasibility of deploying ATO in rail freight and to consider the importance and dependency of the determinants before deploying ATO. The main contributions of this paper are: (1) the combination of Delphi and ANP techniques was applied for the first time in rail literature to investigate the determinants of ATO deployment in rail freight operations; (2) the identification of drivers, challenges, and risks of ATO deployment in freight rail operations, as well as railway subsystems that could be affected by ATO; (3) defining determinants in the analysis of ATO deployment decisions for the first time; (4) estimating costs and benefits of ATO in freight rail for the first time based on our knowledge.

This paper is organised as follows: Sect. 2 provides a literature review of ATO research and the application of the techniques used in this paper. Section 3 describes the methodology introduced for our purposes. The results regarding the main determinants, their dependence and importance, and the costs and benefits of GoA are presented in Sect. 4. Section 5 discusses the results and provides recommendations. In Sect. 6, conclusions are drawn.

2 Related work

2.1 Review of ATO research

Since ATO appears as a technology that can replace manual driving, it is a widely studied area in the literature. According to Yin et al. (2017), there is a large amount of literature on train operation models, train speed control, and speed profile optimization models. In Yin et al. (2017) ATO for main lines and high-speed lines, environmental aspects, and integration with rail traffic control are main areas for future work. In Müller (2020) structural barriers to freight rail innovation were analysed using autonomous freight trains as an example. Although ATO is being considered for use in mainline systems, reliability assessment is a missing aspect, so Corman et al. (2021) explores useful models for ATO. Singh et al. (2021) conducted a detailed survey of the field to identify existing trends, technologies, advances, and challenges in the development and deployment of autonomous trains in rail transportation. The future of transport and the deployment of autonomous vehicles was discussed in the Drive2TheFuture project, and Šoteka et al. (2021) presented the results of opinion surveys on the acceptance of autonomous rail vehicles. Nevertheless, the general challenges and opportunities of introducing autonomous trains in freight rail transport have still not yet been explored, especially from the perspective of experts.

2.2 Delphi studies on railway

The Delphi method has often been used to predict future situations that serve as assumptions and basis for developing further plans and decisions (Preble 1984). Railroad studies using only the Delphi technique can be found in the literature. These studies are based on the collection of opinions or views from experts or other stakeholders dealing with various issues. In recent studies, the Delphi technique has been used to develop an index of switch condition (Rahmani and Seyedhosseini 2020) to identify factors affecting user satisfaction (Ibrahim et al. 2020) and to analyse the possibility of using light rail solutions for freight (Pietrzak et al. 2021). With regard to railway signalling systems, Delphi was used in Aoun et al. (2020a, 2020b) to determine perceptions of the benefits and challenges of the concept of virtual coupling.

After opinions, views, or judgements are identified, the Delphi technique is combined with multicriteria methods to determine the importance of those opinions and to obtain a more accurate assessment. The Delphi technique is often combined with the analytic hierarchy process (AHP), the widely used and well-known multicriteria method. AHP is typically used to determine the weights of indices. For example, in Mlinaric et al. (2018), group AHP was used to calculate aggregate weights for rail performance indicators of ITS using Delphi. In Wenge et al. (2020), AHP was applied to evaluate a sustainable indicator system. In the most recent work (Aoun et al. 2021), Delphi and AHP were combined to calculate weights for criteria used to compare moving block and virtual coupling concepts to signalling.

In addressing some specific cases in the literature, one finds the combination of Delphi technique with more than one multicriteria method. In addition, there are also combinations of Delphi with other analytical tools. Table 2 lists some papers in which Delphi has been used in a traditional or fuzzy context with other methods.

Table 2 Delphi technique with other methods
Full size table

2.3 Contribution to the literature

The AHP method assumes that the criteria or indicators assessed are independent of each other and that there are no relationships between them. However, if this is not the case, the analytic network process (ANP) method is considered more appropriate. The ANP method can be used to create a network of dependencies and importance between criteria as a function of alternatives. In Tavassolirizi et al. (2020) after identifying the factors that cause delays in rail projects, ANP was applied with a 2-round Delphi survey to evaluate the effective factors for delays. Delphi was used to identify key sustainable indicators for evaluating an intercity transportation system, while the DEMETAL-ANP method is used to identify causality between indicators (Rao 2021). In this work, ANP is applied for the first time to explore ATO perspectives in rail transportation through pairwise and mutual priority evaluation of determinants by aggregating expert opinions.

3 Methodology

This study followed a three-stage research approach (Fig. 1). A three-stage approach is proposed to investigate the drivers and barriers to ATO adoption in freight rail operations. The first stage is based on an open Delphi questionnaire to collect experts' opinions on the challenges, risks, benefits, and critical subsystems of ATO in freight rail operations. The open Delphi questionnaire was selected in this study because it is a well-known method used when making a judgement or prediction based on the knowledge and experience of experts (Afshar et al. 2021; Rao and Gao 2021). Since the implementation of ATO is a challenging task for railways, the Delphi open questionnaire is appropriate because it allows for more discussion, iterations and responses to the question “what…can/should be” rather than closed-end surveys with “what…is” type of questions (Gossler et al. 2019; Kumar et al. 2021).

Fig. 1
figure 1

Stages of the research methodology

Full size image

In the second stage, the ANP technique is applied to develop and prioritise the relationship between the determinants essential for the analysis of ATO deployment. ANP is suitable because of its ability to extract judgements from experts, organise judgements by creating relationships between them, and quantify them in terms of priorities, and allows the representation of different opinions after discussions and debates (Saaty and Vargas 2006). ANP is also a popular multicriteria method that derives the weights of determinants by considering dependences, interdependences, and feedback effects between determinants at the same level and between different levels (Huang 2021). The combination of Delphi and ANP techniques is suitable due to the main feature of possible collection and analysis of judgements and the common requirement of reliable results. The main concern of Delphi is to achieve consensus among expert opinions and/or judgements (Hosseini et al. 2022), while in ANP the consistency rate of judgements must be less than 10% (Ivanović et al. 2013).

In the final phase, the experts who did not participate in the second phase were asked to predict the costs and benefits of each grade of automation (GoA) in freight rail. The questions for the final questionnaire were formulated on the basis of the feedback from the first and second phases. All three stages of the research methodology are presented in Fig. 1.

3.1 Delphi technique

The Delphi technique was first introduced by RAND and applied by Dalkey and Helmer (1963) in 1963. It can be defined as an effective procedure for structuring a group communication through interaction between individuals in dealing with a complex problem. The procedure is organised through anonymous iterative processes in which each member of a group establishes his or her view on a problem (Harold et al. 1975; 2002). After the initial process and the processing of new processes, individuals have access to the views of the previous process from other members (Hirschhorn et al. 2018). The goal of group communication from individuals is to generate ideas, gain meaningful insight into a problem and options or alternatives, and reach a consensus on an issue to make accurate predictions and better decisions. The Delphi technique has been used in numerous fields to explore complex and controversial issues related to new technologies, social policy, and environmental concerns. Although it is widely used, there is still no standardised guidance on how to design and implement the Delphi technique (de Loë et al. 2016). However, its advantages, such as the inclusion of geographically distant individuals and the combination with other methods, are recognised in the literature (Hirschhorn et al. 2018; Kluge et al. 2020).

3.2 ANP method

In decision-making, many problems cannot be considered as a hierarchy structure using the AHP method because of the interdependent relationships between elements. To overcome this, Saaty (2006) introduced the analytic network process (ANP) method. ANP has a network structure and considers the relationship between elements within a component and elements between components in the decision-making process. In ANP a component of a decision network is denoted by \(C_{h}\), \(h = 1, \ldots .m.\), and is assumed to have \(n_{h}\) elements denoted by \(e_{{h_{1} }}\), \(e_{{h_{2} }}\),….\( e_{{h_{mh} }}\). The influences of the elements in a component and on any element in the network, as well as the ranking of alternatives, are represented by a priority vector derived from pairwise comparisons in the AHP method (Saaty 2006). The ANP method can conduct the analysis of complex problems that include both quantitative and qualitative values. In the case of qualitative data, the well-known Saaty scale given in Table 3 is used.

Table 3 Saaty scale
Full size table

The influence of the elements in the component and the influence of the elements of the component on other elements in the network is shown in the following supermatrix (Fig. 2). The supermatrix consists of several submatrices, called "blocks of the supermatrix" \(W_{ij}\), which represent the influences between the elements in the component and the influences of the elements between the components. Each column of \(W_{ij}\) is a principal eigenvector of the influence (importance) of the elements in the \(i_{th}\) component of the network on an element in the \(j_{th}\) component. If there are no influences and relationships, \(W_{ij}\) is zero.

Fig. 2
figure 2

a Supermatrix and b submatrix

Full size image

3.3 Delphi-ANP approach

This section explains the development of the Delphi-ANP approach. The first Delphi questionnaire contains eight open-ended questions and one sub-question. Each odd-numbered question asked experts to describe challenges, risks, benefits, and critical subsystems of ATO deployment in freight rail operations. The even-numbered questions targeted which GoAs would cause greater challenges, risks, benefits, and impacts to rail freight subsystems. In this phase, the responses received were processed with the goal of identifying the key determinants of ATO in freight rail operations. The determinants were identified and classified into six groups by the authors.

In order to develop the dependency of the determinants and determine their priority, the ANP method was chosen because of it is able to analyse dependencies and interactions between elements within the model. Other reasons for the ANP method are its ability to trade off between alternatives and other elements, establish priorities among all elements, tangible and intangible, in a system based on the use of a ratio scale created by human judgement rather than arbitrary scales, and involve multiple actors or group decisions with multiple actors (Chang et al. 2009; Ivanović et al. 2013).

The Delphi-ANP process was developed in several stages. The results of the first stage, i.e. the Delphi questionnaire in Fig. 1, serve as input for the development of the ANP model. In the next stage, the “SuperDecisions” software is used to create dependencies and feedbacks between elements at the same or different level of decision-making. The dependency of determinants in the ANP model is developed by the authors. In “SuperDecisions” all determinants, the relationships between them and the alternatives were presented.

According to Turoff (1970), Delphi questionnaires with short answers and a ranking scheme are the ideal solution for data collection. In the third step, the experts were asked to confirm the developed dependence of the determinants and to determine the pairwise and mutual priority of the determinants based on the Saaty scale values (see Table 3). The pairwise and mutual dependency comparison of the determinants is sent to the experts in the form of priority determination tables. The results of each expert are manually summarised, and the average value for the priority of the determinants is presented in a software application. The priorities of the determinants are calculated by the SuperDecisions software. The highest prioritised determinants are used to determine the most suitable alternative. The alternatives evaluated are known grades of automation, i.e. ATO GoA2, GoA3, and GoA4. The process of prioritising the alternatives was the same as for the criteria.

3.3.1 Data used

In the case of missing, unreliable and insufficient data, it is possible to rely on the intuition, knowledge, and experience of individual experts. This alternative is more favourable than being guided only by one's own intuition. In the field of planning, the combination of techniques that use objective and subjective information attempts to achieve the most effective response and the best results in terms of accuracy (Landeta et al. 2008). Therefore, for this study, the qualitative Delphi technique was combined with the ANP method, which can be fed with qualitative data for qualitative determinants (Kheybari et al. 2020).

The initial input data for the Delphi-ANP approach are mainly based on the opinions of the experts who participated in the survey. In the first phase of the Delphi method, an open-ended questionnaire was used to collect data, i.e. subjective information from experts on challenges, risks and benefits, and critical subsystems in the case of ATO implementation in rail freight. The information obtained through this channel was utilised to create the determinants relevant for considering ATO implementation in rail freight. The identified determinants were categorised an ANP model was developed by the authors based on these factors, which was reviewed by the participating experts in the second stage. In the second stage, each expert used the Saaty scale scores (Table 3) to collect data for prioritising the determinants. The experts’ data were summarised, and the average scores were used as input data for the SuperDecisions software of the ANP model. The importance of the determinants within the same and different categories was calculated, and the consistency rate was met. In the final phase of the study, the experts’ knowledge was used to determine the values of costs and benefits of different levels of automation.

3.3.2 Expert selection

The most important step in conducting a Delphi study is the selection of experts to ensure the breadth of knowledge on a given question (Hirschhorn et al. 2018). However, there are no clear guidelines on the number of experts, while a stable number of participants is suggested (Feuerstein et al. 2018). According to the study of de Loë et al. (2016), the number of participants depends on the research topic and the required knowledge of the participants, and 10 to 50 participants were the most appropriate. Since there is no agreement on the number of participants, recommendations, or a clear definition of “small” or “large” groups (Pietrzak et al. 2021), the number of experts varies in each phase of this study.

To ensure a heterogeneous and high-quality study, criteria for expert selection have been proposed in the literature (Feuerstein et al. 2018). Since this study has a narrow topic, the main criteria for participation in the questionnaire were:

  • Work experience with ATO or railway signalling systems; and

  • Research or consulting experience on ATO or signalling topics.

The LinkedIn network and review of articles in scientific databases were used to identify ATO experts. The experts in the literature search were extracted from well-founded articles using the search terms (Autonomous Train Operation OR Autonomous Train OR ATO). The experts on LinkedIn were searched in two rounds using the following search terms: (ATO or automated train operation) and (Railway signalling). After reviewing the articles and LinkedIn profiles of their authors, an invitation to participate in a first round of Delphi was sent via email. The first round of the survey was launched on 20 December 2021. Of the 110 people invited, a total of 40 agreed to participate in the survey. Thirty-four experts who met the inclusion criteria were recruited and 34 responses were collected in the first phase. Due to the lack of responses, 4 experts were excluded. In the next round of this phase, due to incomplete responses, 10 experts were asked to change their responses to achieve consensus among all experts. As a result, the changes of 10 experts were submitted, while only 5 experts were continued in the second phase. Further rounds to improve the quality of the results were not considered because of the risk of experts dropping out (Kluge et al. 2020). After reviewing the responses, the most important factors were identified. Therefore, in total 25 participants were invited to participate in a second phase. However, due to unfamiliarity with the ANP method, many of the participants declined to participate, and 15 dropped out at this stage. Therefore, only 10 participants took part in the survey in the second phase. The first round of survey in this phase was conducted on 30 January 2022. In the final phase, 15 participants who had declined the second phase were asked to participate in the final round. For the final round, which began on 15 March, 15 responses were received on a closing day on 15th April.

4 Findings

4.1 First Delphi phase

This section presents the results of the first round of the Delphi survey. Based on the results of the first round of Delphi, 61 determinants and 13 critical systems were extracted. The most important determinants were grouped as Driver 24, 19 determinants are found under Challenges and 18 determinants under Risks. Regarding the first even question To what extent would challenges increase with a higher GoA? 33 of 34 experts confirmed that higher GoAs would increase challenges, while 1 expert was not sure. The expected higher challenges with the introduction of GoA2 are in the areas of ATO instructions, certification, legal issues, planning, forecasting, warranty, and control algorithms. Greater challenges are particularly related to marshalling operations and monitoring systems. Numerous challenges would increase with GoA3 and GoA4 related to on-time performance, capacity, safety, reliability, capital and operating costs, detection systems, and remote control.

The second even question was: would the move to higher GoA bring higher risks, and at which grade numerous risks can be expected? 23 experts answered positively, 9 experts disagreed, and 2 are not sure. An increase in risks is expected in terms of interactions with signalling systems, unforeseen situations, and the possibility of a higher number of accidents and collisions. The highest risks for GoA3 and GoA4 are related to train reliability, congestions and capacity loss, operational control, cybersecurity, and infrastructure costs. When asked which GoA will bring numerous benefits, the experts cited GoA2 as the most promising, followed by GoA4, GoA3, and GoA1. For this question, experts were presented with a sub-question on the advantages between GoA3 and GoA4. GoA4 was cited as the most advantageous alternative, while 14 experts believe that GoA3 has numerous advantages. However, one expert believes that both levels provide the same benefits, while two experts believe that a higher level of GoA2 is not necessary in freight transportation. In addition, all experts agreed that for GoA3 and GoA4, parts of the infrastructure or subsystems would be most affected by the introduction of ATO. The responses from the first round of Delphi are summarised and categorised and presented in Fig. 3. The drivers for ATO adoption, the challenges, and the risks associated with ATO adoption are separately listed in Fig. 3. The same figure shows which subsystems may be most affected by the introduction of ATO.

Fig. 3
figure 3

Drivers, challenges, risks, and critical subsystems of ATO adoption

Full size image

4.2 Second phase: Delphi-ANP approach

4.2.1 Pairwise priority of determinants

Considering all the factors listed in Fig. 3, key criteria have been identified (see Table 4) that need to be analysed before ATO is implemented in freight rail. According to the experts, the cost of replacing the old devices and equipment with new ones should be considered before introducing ATO. In the technical category, the provision of standardisation, certification, and guidance for new technologies is the most important criterion in planning ATO in freight rail operations. Operational factors related to ATO in rail freight are primarily concerned with efficient operations in yards and the operating and environmental conditions in which trains operate. Safety on open tracks and in marshalling yards, reliability of the traffic management system (TMS) and energy consumption are also issues that should be analysed before ATO in freight rail operations is introduced. Table 4 shows a set of criteria that should be analysed for the implementation of ATO in rail freight. It also shows the classification of the criteria into six categories and the priority of the criteria according to the expert opinions.

Table 4 Importance determinants
Full size table

4.2.2 Mutual priority of determinants

Despite the relationship between determinants and their comparison within a category, there is also an interrelation of determinants between categories. The relationship between the determinants is established by the authors' knowledge, while the prioritisation between them was done by experts. In Fig. 4a, the relationship between operational (OC), safety (SC), reliability (RC), and technical (TC) determinants can be seen, along with the value of priority between determinants. The arrow direction denotes the effect of one determinant over another, while the number on the arrow denotes the importance of one over another. For example, operational determinant 7 (OC7) has an effect on TC1. A. A higher value dependency between OC7 and TC1 means that more capacity (OC7) requires the homogenous use of ATO throughout the network.

Fig. 4
figure 4

Dependence and priority of a operational criteria, reliability criteria (RC), safety criteria (SC) towards technical criteria (TC) and b cost criteria C towards OC, RC, SC, TC, energy criteria (EC)

Full size image

Moreover, it can be seen from Fig. 4a that the criterion of safety level on lines and yards (SC1) is related to TC1, which means that a higher safety level depends on a homogeneous use of ATO throughout the network. As can be seen from Fig. 4a the priority values vary between different pairs of determinants, while their sum should equal 1. Therefore, the sum of all priority values for determinants (OC, SC and RC) associated with TC1 is equal to 1. However, in Fig. 4b it can be seen that the value of dependence is 1. This means that these determinants are connected, but there is no priority between them when considering the introduction of ATO. Figure 4b shows the dependency and priority of the cost determinants on determinants from other categories such as operational, technical, safety, reliability, and energy. In Fig. 4b, it can be seen that the value of dependence between some determinants is equal to 1. This means that these determinants are related to each other, but there is no priority between them when considering the implementation of ATO. For example, the reduction in energy consumption (EC1) is only related to the investment cost (CC1) and has the same priority before the introduction of ATO. The dependency between them also means that the investment costs in ATO can potentially reduce energy consumption in rail freight.

The dependence and priority of each determinant on the reliability, environmental, and safety criteria are shown in Fig. 5. For example, in Fig. 5a the dependence between reliability determinant 5 (RC5) and energy determinants 1 and 2 (EC1 and EC2) can be seen. The arrows indicate that the reliability of running time (RC5) is connected to the energy consumption (EC1) and noise of wear on wheels and rails (EC2). However, the higher priority of EC2 (0.510) indicates that noise is more important than reliability of running time when ATO adoption is considered.

Fig. 5
figure 5

Dependence and priority of a EC towards RC, b OC towards RC, SC, EC and c SC towards RC

Full size image

4.2.3 Ranking GoAs

Figure 6 shows the results of the fourth step, i.e. the preferences for the implementation of ATO GoAs in freight rail operations according to the most important criteria. The preferences for which GoAs should be considered for implementation on rail freight are conducted according to the most important determinants given in Table 4. In terms of expenditures, GoA2 would require the highest investment costs (CC1), while GoA3 and GoA4 would cost less. However, a reduction in operating costs (CC2) is expected for GoA2 and GoA3, while the GoA4 would not have a significant impact. It means that significant energy savings will be achieved in GoA3 and removal of drivers will not bring significant reduction in operational costs. The positive impacts of ATO for rail freight on capacity (OC7), traffic disruptions and congestions (OC3), environmental aspects (EC1), and braking system reliability (RC4) are expected at GoA3. Benefits such as maintenance of safety levels (SC1), efficient shunting operations (OC4), and TMS reliability (RC3) are expected for GoA2, while standardisation and certification requirements (TC4) are expected for GoA3.

Fig. 6
figure 6

Priority of GoAs

Full size image

Since the adoption of the highest level of automation in rail freight (ATO GoA4) is the most questionable and challenging, the most important criteria to be considered are compared in Fig. 7. From Fig. 7, the state of ATO GoA4 according to the highest prioritised criteria can be seen. The most critical field for GoA4 adoption is shunting operation (OC4). Then, the reduction in energy saving with GoA4 should be compared to energy consumption with GoA3 before implementation comparison. Before GoA4 strong recommendations are related to standardisation, certification, and guidance, as well as reliability of rail TMS. Further, doubts related to level of safety and disruptions and congestions exist in the case of GoA4.

Fig. 7
figure 7

Priority of GoA4

Full size image

4.3 Third phase: costs/benefit analyses

A separate questionnaire was used to determine the range of ATO costs and benefits for various GoAs. The results of this separate survey are given in Table 5. Table 5 shows the potential savings for GoA2, GoA3, and GoA4. In Table 5, the survey results are divided into capital expenditures, operating costs, and reasons for implementing ATO in freight rail operations, such as energy, noise, capacity, punctuality, and safety.

Table 5 Potential for capacity increment and reduction in energy consumption, delays and noise
Full size table

To be ready for ATO, old locomotives should be modernised. GoA2 would require numerous changes and there would not be much difference between GoA3 and GoA4 in installations, but likely in procedures. Therefore, the investment costs for GoA2 and GoA3 were examined. For GoA2, 28.6% of the experts indicated that retrofitting locomotives would require € 50–100 k/unit. However, another group of experts (28.6% of experts) indicated € 300–400 k/unit. 42.8% of the experts have no idea about the costs. The same proportion of participants has no idea of the cost of converting locomotives to GoA3. According to the answers of 14.2% of the experts, the implementation of GoA3 would require 100,000 €/unit, while 43% of the experts think that € 500–600 k/unit can be expected.

With the implementation of GoA2, energy consumption can be reduced, resulting in lower operating costs. According to 71.4% of the experts, GoA2 can reduce operating costs by 10–20%, 14.2% of the experts indicated a reduction of 40–50%, while the rest (14.3%) have no idea. For GoA3, 57.2% of experts expect operating costs to be reduced by up to 10–20%, 14.3% indicated a reduction of 25–30%, and 14.3% did not specify. A significant reduction in operating costs is possible in GoA4 through staff reductions. 16.7% of the experts gave no answer, while two other groups of 33.3% indicated a reduction of 15–20% and 25–30%, respectively.

When asked by how much would energy consumption be reduced with ATO grade 2 in rail freight? 71.5% of the experts indicated an energy reduction of 10–20%, while 29.5% of the experts gave no answer. As can be seen in Table 5 for GoA3 and GoA4 potential expectations are varied. Regarding rail capacity, some interviewed experts see no correlation between investigated GoAs and capacity. For the improvement in train punctuality, some experts believe that this depends on other factors and not on particular GoAs. Although the introduction of ATO in freight rail operations, less wear and tear on wheels and rails is possible, reducing noise levels, numerous experts saw no relationship between ATO and noise.

To the question would safety be increased or not with ATO? 57.1% of the experts gave a negative answer. Some experts felt that GoA 2 would not affect safety. For the higher GoAs, the experts believe that safety would remain at the same level, even though there may be inconvenience to the public. In the case of ATO over ETCS, safety is not affected at all levels of automation because ETCS is a standardised form of ATP. Regarding the increase in safety, 42.9% of the experts stated that ATO increases safety. According to 33.4% of the responses, safety can be increased by 50% at all ATO while 66.6% answered that GoA2 can increase safety by 5–10%, GoA3 by 5–20% and GoA4 by 10–20%.

5 Discussion

The study combined Delphi and ANP techniques to examine the determinants of ATO adoption in rail freight. The method helped identify the priority and interrelationships among determinants to be considered prior to ATO deployment. The determinants were identified using an open-ended questionnaire. To identify the priorities among the determinants, the next phase of the Delphi technique was initiated. Since relationships exist among the determinants, the ANP method was chosen for calculating priorities and ranking the grade of automation. The combination of these two techniques eliminates biases that may be caused by the subjective intentions of the experts. Therefore, the results are more realistic and reliable, and such an experimental approach is well applicable.

Moreover, in the last phase of the study, the area of determinants and prediction of benefits for different GoAs was investigated. Based on the questionnaire in the first phase, it can be concluded that the introduction of ATO has a positive impact on freight operations and that it is a technology that can increase the punctuality, safety, capacity, energy consumption, etc. of freight transport. The survey also identifies the negative impacts related to the challenges and risks that could occur. ANP was used to summarise the priorities of the determinants and their interrelationships based on the expert comparisons. For example, Fig. 4a shows the relationship between the homogeneous use of ATO in the overall network (TC1) and some operational criteria. This means that the homogeneous use of ATO in freight operations can contribute to better punctuality, capacity, reduction in disruptions, efficient operation, etc. Figure 5a shows the dependence between energy and some reliability criteria. From the figure, it can be seen that noise reduction, according to the experts, depends on the reliability of running times and braking systems. From another point of view, it is clear that uneven running and emergency braking lead to wear of the wheels and rails, which in turn causes noise.

Based on the results of the ANP methodology recommendations were made regarding cost requirements, technical, reliability, operational criteria, safety, and energy benefits.

(1) Cost requirements: The experts recognised the highest priority for capital costs in freight ATO deployment. The higher priority given to operating costs indicates that ATO helps to reduce these costs. However, in addition to the lowest value for the priority given to maintenance costs, the experts also recognised significant correlations with other determinants (Fig. 4b). The correlation with the technical determinants means that extending ATO to the entire network and upgrading to a higher GoA requires higher maintenance costs. On the other hand, if maintenance expenditures are inappropriate, operational reliability of running, overall traffic management, and safety levels may be jeopardised.

(2) Technical criteria: Before the introduction of ATO, experts gave the highest priority to standardisation, certification, and guidance. However, the second highest priority is the reliance on trackside devices. This priority relates to the ability to use wayside equipment simultaneously with ATO and the cost of removing it. Then, the experts recognised the homogeneous use of ATO as a key determinant, and giving it the third priority. This determinant is particularly important because there are numerous interactions with other determinants such as cost, reliability, and operational benefits. The lower priority determinants relate to computerisation and detection and system testing and trial, which generally cannot be avoided at higher GoAs. (3) Reliability Criteria: In this group, the highest priority to the overall RTMS reliability, reliability of the braking system, and the reliability of the ATO. This is because these elements are essential to provide good train services with ATO to customers. (4) Operational criteria: According to some participants, the feasibility of ATO in railway yards is questionable. The goal of increasing capacity is also questioned. In addition, questions arise regarding train traffic in mixed traffic and uncontrolled operation according to GoA3 and GoA4. Therefore, the experts interviewed gave the highest priority to the questions regarding shunting operations, capacity increase, and operational and environmental conditions. The determinant of traffic disruptions and congestion is not given priority because it should not be related to the implementation of ATO. The rating of other factors such as ensure punctuality (OC6), mode of driving/driving style (OC2), train and freight dynamics (OC5), and high precision (OC8) is poor because they may not need to be considered prior to ATO implementation. For example, high precision (OC8) received a lower priority because it is not important for preliminary considerations and is more related to ATO on urban tracks. However, apart from the low priority, the experts recognised strong relationships between the determinants and other determinants under consideration.

(5) Safety issues: the most important determinant is the level of safety on lines and stations. More attention needs to be paid to this aspect, as ATO, especially a higher GoAs, can affect the safety level in the mixed transport network.

(6) Energy benefits: The potential of energy consumption was analysed as a priority before the introduction of ATO was introduced. It was recognised that it should be considered primarily in terms of capital investment. In addition, to save energy, factors such as operation and reliability should be met.

The introduction of new technologies is expensive, and in the past new innovative solutions have even led to additional costs (EC 2021). Therefore, the positive contributions of new technologies are investigated before their introduction. To investigate the cost-effectiveness of ATO in rail freight, an additional survey was conducted to determine the range of costs of deploying ATO deployment and the values of benefits. Existing studies have cited, examined, and demonstrated many benefits of ATO for passenger rail. These findings come from automation of metro lines already in operation or from automation trials on rail lines. However, the cost of implementing ATO, especially for freight railways, is not discussed. In this study, the retrofitting of locomotives for GoA2 is valued differently by the same expert groups. The higher amount of €300–400 k/unit is consistent with the values in Quaglietta (2020), although the costs were not split between GoAs. For GoA3, a larger group of experts indicated an amount of €500–600 k/unit. Higher investments for GoA3 refer to equipment for remote control of trains, remote monitoring systems for rolling stock, two-way communication, etc. Despite the investment costs, the results of the survey have shown that the operating costs for GoA4 can be significantly reduced by eliminating the train drivers. Furthermore, this reduction is related to the 30% energy savings indicated in the results for GoA3 and GoA4. These savings identified in the study are consistent with values demonstrated for urban rail systems (Judith et al. 2015; Wang et al. 2016). Another advantage of autonomous trains is the expected improvement in safety for railway undertakings (Singh et al. 2021). This benefit is confirmed by the opinions of experts, and each GoA would bring a certain specific level of improvement.

According to the RailFreight Forward initiative, ATO results in less wear on brakes and wheels, and thus less noise. However, the actual value of noise reduction is not known. Based on the survey results, half of the experts surveyed did not recognise this opportunity, although some expected some percentage of noise reduction. The survey results show that capacity increases by 5 to 10% depending on the GoA, while many experts do not see a direct relationship between capacity and GoA. This opinion is confirmed by RFF (2020) which highlights that freight rail capacity at the ATO depends on additional infrastructure requirements, such as the introduction of moving block concepts. According to the report, if the requirements are met, about 10% of the capacity increase can be achieved with ATO. Based on the results, ATO can reduce rail freight delays, with the reduction depending on GOAs, and this benefit was justified with the results of trials with ATO on rail lines summarised in Poulus et al. (2018).

6 Conclusion

This study shows that ATO has numerous positive impacts on rail freight. The degree of positive impact depends on the degree of automation. However, challenges and risks must also be considered during the planning phase of ATO implementation. To reduce or neutralise the challenges and risks, determinants for ATO implementation are identified. Among the 27 identified and weighted determinants, investment cost is also mentioned by the experts as one of the most important determinants for the decision to implement ATO, in addition to significant benefits. Also, among the other determinants, the level of safety on the lines and in the marshalling yards was mentioned as one of the most important priorities before the introduction of ATO. In addition, energy saving is not only proven but still influencing crucial factor that experts believe should be investigated through a cost–benefit analysis before introducing ATO.

To assist decision-makers and policymakers in the strategic planning of ATO in rail freight, the interdependencies of the determinants were identified. Based on these relationships, priorities among determinants from different categories can be identified. For example, from the interdependence of determinants, the impact of technical determinants on the provision of improved operational determinants can be identified (see Fig. 4a). For example, the uniform use of ATO in the network (TC1) may have an impact on the level of safety and regulation of some operational determinants. Based on the results in Fig. 4b, it is possible to see what can cause higher costs and how higher costs can have a positive impact on the reliability of ATO and other connected subsystems. This study can help infrastructure managers and freight rail operators make a decision on whether to implement ATO. In addition, it can be useful to prioritise their strategic actions to address the challenges and risks associated with ATO in order to maintain smooth deployment and traffic operations under ATO. The findings suggest that the challenges of collaboration between infrastructure managers (IMs) and railway companies should be balanced in terms of interests in ATO deployment. In addition, stakeholders should develop effective strategies to address challenges related to driver and customer acceptance of ATO, particularly GoA4. As freight trains need to be retrofitted or upgraded, the rolling stock manufacturers and IMs should decide who will do the retrofitting. Although priorities for this task may vary from country to country, this challenge is important because proprietary solutions in the system architecture can affect interoperability. In addition, cooperation between all stakeholders is important because proprietary solutions can make EU-wide interoperable operation difficult.

The results of the last phase have shown that investment costs depend on GoAs. Nevertheless, there is no consensus among experts on the cost of each GoA. The ATO is also expected to have a positive impact on rail freight. The number of train drivers will be affected by the grades of automation.

Based on the results of this study, the following recommendations are made: (1) Decision-makers should conduct trials and tests of ATO to take advantage of the positive opportunities. (2) Stakeholders should consider measures to mitigate the risks and challenges of using ATO in freight operations. (3) Consider train operations in mixed traffic. (4) Analyse the suitability of specific GoAs based on operating conditions and environment. (5) Repeated testing of systems to ensure system safety before implementation, especially in the case of GoA3 and GoA4. (6) Determination of appropriate strategies and tools to reduce the vulnerability of critical rail subsystems such as level crossings, intersections, and TMS (7) Development of an ATO structure that is interoperable and compatible with signalling systems (ERTMS) of all countries. (8) To reduce scepticism about ATO in rail freight rail, measures should be created for railway undertakings and/or freight customers. (9) To facilitate cooperation and interaction between IMs and RUs, the regulations regarding the costs and benefits need to be reviewed.

6.1 Limitations and future work

In this study, the ANP method was primarily used to investigate the positive and critical aspects of ATO implementation and the grade of automation in rail freight. Although it is a popular multicriteria method, the main problem of this method recognised in the literature is the issue of divergence when absorbing criteria exist (Huang 2021). Due to the complexity and high time consumption of ANP and other multicriteria methods, advanced and hybrid optimization and metaheuristic models have been shown to be efficient for various challenging decision problems (Farughi and Mostafayi 2017; Kavoosi et al. 2020). Consequently, metaheuristic algorithms have been proposed for solving various optimization problems. For the berth scheduling problem, Kavoosi et al. (2020) proposed a universal island-based metaheuristic algorithm. The advantages of their algorithm are that it traverses the search space more effectively and finds superior solutions with a computation time of no more than 306 s for the generated large problem instances, which can overcome the time-consuming ANP problem. Future work can therefore focus on overcoming ANP problems by testing metaheuristic algorithms on the topic of this work or similar topics related to decision-making for innovative solutions and technologies in transportation. The use of ATO in rail freight requires that several objectives for rail performance are achieved, while on the other hand, the cost of ATO implementation should be minimised. In order to find high-quality solutions, multi-objective optimization is proposed to analyse solutions with different characteristics from different perspectives (Zhao and Zhang 2020). When implementing ATO, various constraints from operational and infrastructure perspectives need to be considered. In the future, hybrid techniques should be applied to improve and contribute to the quality of decision-making. For example, such a hybrid technique called Adaptive Polyploid Memetic Algorithm (APMA) is proposed by Dulebenets (2021) for the CTD truck scheduling problem.

This study found that positive impacts increase with higher GoA, but appropriate values are lacking. In addition, the results of the study show that investment costs depend on GoA. Therefore, these results may motivate researchers to further investigate the potential and cost of ATO in freight transportation using real case studies. Selecting appropriate GoAs for a specific rail section, route, or network, investigating the feasibility of ATO in marshalling yards, the suitability of ATO for specific rail products (wagonload, block trains, or intermodal trains), investigating the positive impacts of specific GoAs on capacity utilisation, energy consumption, noise levels, safety levels, and punctuality are some important research areas for researchers. However, the most important issue for this future research is the availability of data and the approach or methodology that should be used. Since it is a difficult task to find the most cost-effective GoA for rail freight, ATO is expected to have various positive effects in rail freight, so using hybrid methods such as multi-objective optimization is the most appropriate approach. An example of an optimization model that can be used is a mathematical mixed-integer linear programming model presented in Kavoosi et al. (2019) for the berth scheduling problem, which aims to minimise the sum of the waiting cost, the dispatch cost, and the delayed departure cost of the ships. Then, a novel multi-objective optimization model presented in Pasha et al. (2022) for the vehicle routing problem can be adapted to find optimal GoAs for different scenarios.

Since this study is based on Delphi surveys, the results of this work are somewhat limited by the data sources and for example not firm cost calculations. One limitation of this study that can be addressed in future work is the inclusion of participants from railway undertakings, railway companies, government agencies, and customers. In case of availability of data collected from various test demonstrations, advanced optimization models such as those used in Rabbani et al. (2022) can be used in the future to find the feasible GoAs that provide numerous benefits at the lowest cost.