key: cord-0983452-pjz2k07i
authors: Mora, Luca; Wu, Xinyi; Panori, Anastasia
title: Mind the Gap: Developments in Autonomous Driving Research and the Sustainability Challenge
date: 2020-09-11
journal: J Clean Prod
DOI: 10.1016/j.jclepro.2020.124087
sha: aaa0b075277b781e60773cc4d5d1b3ffe4855d9b
doc_id: 983452
cord_uid: pjz2k07i

Scientific knowledge on autonomous-driving technology is expanding at a faster-than-ever pace. As a result, the likelihood of incurring information overload is particularly notable for researchers, who can struggle to overcome the gap between information processing requirements and information processing capacity. We address this issue by adopting a multi-granulation approach to latent knowledge discovery and synthesis in large-scale research domains. The proposed methodology combines citation-based community detection methods and topic modeling techniques to give a concise but comprehensive overview of how the autonomous vehicle (AV) research field is conceptually structured. Thirteen core thematic areas are extracted and presented by mining the large data-rich environments resulting from 50 years of AV research. The analysis demonstrates that this research field is strongly oriented towards examining the technological developments needed to enable the widespread rollout of AVs, whereas it largely overlooks the wide-ranging sustainability implications of this sociotechnical transition. On account of these findings, we call for a broader engagement of AV researchers with the sustainability concept and we invite them to increase their commitment to conducting systematic investigations into the sustainability of AV deployment. Sustainability research is urgently required to produce an evidence-based understanding of what new sociotechnical arrangements are needed to ensure that the systemic technological change introduced by AV-based transport systems can fulfill societal functions while meeting the urgent need for more sustainable transport solutions.

The first motor vehicle that pioneering mechanical engineer Karl Benz invented in 1885 has escalated into a global fleet of approximately one billion cars and trucks, which constantly This paper contributes to meet such an objective by reporting on the results of a multimethod bibliometric study which offers a synthetized view of the scientific knowledge produced during the last five decades of AV research . Citation-based community detection methods and topic modeling based on exploratory factor analysis are combined to extract the relevant semantic structures (i.e. main keywords, central topics, and core research themes) hidden in this complex data-rich environment. These sub-information systems of latent variables are then analyzed to give a full account of how the intellectual structure of the AV research field is conceptually shaped.

The paper is structured in four main sections. After presenting the growth rate of the AV research domain, the paper goes into the rationale behind its call for a more consistent effort to reach knowledge summarization in the large network of AV-related scientific publications released during the period under investigation. The introductory discussion ends by focusing on the challenges affecting large-scale exploratory text analytics and the role that digitallyinduced text mining techniques play in facilitating knowledge discovery processes. The paper continues with a detailed description of the methodology adopted to conduct the bibliometric analysis. This second section is followed by a discussion on the latent knowledge extracted during the analytical process, which is used to convey core knowledge from a large volume of scientific publications in a concise but comprehensive way. The paper concludes with a final section in which the insights captured through the analysis is summarized and used to offer recommendations on future research directions. The analysis demonstrates that AV research is not paying sufficient attention to the environmental consequences and socio-economic, cultural, political, institutional, and organizational implications that a mass market for autonomous driving technology can generate. After elaborating on this evidence-based statement, the conclusive section also reports on the limitations of the study and details its contribution. J o u r n a l P r e -p r o o f 6 Scopus data confirms that autonomous driving has become a prominent topic of investigation in the scientific debate on transportation futures. Aware of the difficulties that researchers may experience when attempting to grasp the rapidly expanding intellectual structure of the AV research field, two initial studies have been conducted which attempt to guide summarization by means of knowledge maps. These studies combine knowledge domain visualization and bibliometric techniques to identify relevant semantic links between natural language representation components, whose visualization facilitates the understanding of core concepts embedded in large-scale collections of AV-related bibliographic data sources.

The first study is conducted by Gandia et al. (2019) , who focus on the WoS-indexed literature published between 1969 and 2018. The selection of this timeframe leads to the identification of 10,580 publications, from which two groups of items are extracted: a set of "research-front concepts" (Chen, 2006, p. 359 ) and a number of emergent research categories. The extraction process is conducted by using the software CiteSpace, where the burst detection algorithm designed by Kleiberg (2003) allows the authors to estimate the burst of each emergent concept and research category. When comparing the results, evidence of a trend change surfaces. According to the findings, AV research shows an initial techno-centric focus, with subject areas belonging to engineering and technology disciplines dominating the scientific debate. Research trends begin to evolve around 2015, when a more holistic research environment starts to develop as a result of the growing scientific contribution offered by studies in the social sciences.

The investigation into the conceptual shifts describing the evolutionary nature of the AV research field continues with Rashidi et al. (2020) . CiteScape remains the main supporting tool, but unlike the previous study, this second bibliometric analysis: (1) relies on a smaller bibliographic record composed of Scopus-indexed publications, rather than WoS data; (2) 7 domain is conceptually structured. As Gandia et al. (2019) observe, additional research is needed to complement their findings by increasing the level of detail. This requires looking over the too wide approximation of concepts resulting from their initial lists of emergent thematic areas and deepening the current understanding of the latent knowledge that each thematic area is shaped by. However, fulfilling this refined knowledge extraction process requires coping with the "curse of dimensionality" (Verleysen and François, 2005, p. 761) .

By adopting metrics of proximity or distances (Glänzel et al., 2019) , community detection algorithms can be used to split a research field into clusters of thematically related publications (Panori et al., 2019) . Thematic clusters are high-dimensional knowledge spaces in which huge amounts of textual data is gathered and connected by an intricate network of semantic links (Mora et al., 2019) . As basic entities of natural languages, words offer an initial level of language-dependent understanding of the clusters' contents. However, the presence of meaningless textual components generate noise, making it difficult to extract core information. Reducing the volume of the input variable space is indispensable, by removing as much irrelevant textual components as possible. During the dimensionality reduction process, depending upon the extent of the synthesis and degree of approximation (Yao, 2004) , different levels of knowledge granulation can be reached.

Given the limitations of manual coding techniques in large-scale exploratory text analyses (Kobayashi et al., 2018) , the discovery of latent knowledge patterns requires examining thematic clusters by means of text mining techniques, which make it possible to automatically reduce dimensionality by filtering quality information from high-dimensional sets of textual data. The core knowledge embedded in a large-scale dataset can be expressed as the sum of three complementary sub-information systems (Jing et al., 2017) of latent variables: main keywords, central topics, and core research themes 1 . Sourcing and connecting the different levels of knowledge which are rooted in these subsystems is key to produce a condensed but thorough representation of the intellectual shape of a research area. As a result, a comprehensive knowledge discovery process entails a multi-granulation perspective (Roslovtsev and Marenkov, 2018; Thijs, 2019) .

Topic modeling is one of the most frequently used computer-assisted text data mining applications for knowledge discovery. Its usage makes it possible to automatically identify and look into sub-information systems of latent variables in high-dimensional collections of textual data, producing "insight in properties underlying those knowledge structures" (Tijssen, 1993, p. 111) . Given a collection of unstructured textual data extracted from a cluster of thematically related publications, topic modeling combines a probabilistic approach to unsupervised learning and co-occurrence measures to: extract the words and phrases of 1 The groups of variables are listed in ascending order of knowledge granularity. greatest significance (Level 1: Keywords); arrange these text elements into groups of core topics (Level 2: Topics); and, facilitate the identification of the foremost thematic area emerging from each group of topics (Level 3: Research themes).

Topic modeling allows the integration of latent knowledge sourced from multiple analytical levels (Valdez et al., 2018) , moving from individual keywords to collections of textual components delineating relevant topics and core themes. With researchers selecting from a wide range of different topic modelling techniques and heterogenous data sources, this multi-granulation perspective to large-scale exploratory text analyses has proven successful in examining knowledge structures in different application domains. (2018) sourced textual data from 25,706 publicly available records to map recurrent topics within aviation incident reports. Talavera et al. (2020) discovered behavioral habits by translating the visual content of 100,000 images into textual data. Roy et al. (2012) offered insights into the environmental contributions to early lexical development by examining more than 200,000 hours of audio and video recordings. This data captures the day-to-day linguistic environment in which a newborn child has been immersed during the first three years of life.

The abovementioned studies demonstrate that topic modelling has been successful in replacing laborious manual coding exercises in which the volume of data would have made the analysis impossible to complete without a computer-assisted approach. In addition, this research shows that different types of objectives call for different approaches to topic modelling and variations in the techniques, yet the analytical stages tend to remain the same. Three main phases can be identified, which have been considered in the framework of this study: preparation, topic modelling, and post-processing (Asmussen and Møller, 2019; Kobayashi et al., 2018) .

To reduce dimensionality and better filter quality information describing the intellectual structure of the AV research field, topic modelling was used to analyze each thematic cluster. WordStat's topic modelling function was selected, which is performed on factor analysis with varimax rotation (Péladeau and Davoodi, 2018) . Multiple levels of analysis were combined during the examination, making it possible to progressively source different types of latent knowledge. By considering co-occurrence values, the software was instructed to detect groups of interrelated keywords and assign a topic to each group. The topics represent a set of underlying variables called factors.

Scree plots (Cattell, 1996) and parallel analyses (Horn, 1965) were used for factor retention purposes, to determine the number of topics to consider for each cluster. In an exploratory factor analysis, a scree plot is a line chart which displays the eigenvalues of all the factors identified during the analytical process in a downward curve (Nebel-Schwalm and Davis, 2011) . The inflection point where the slope of the curve levels off divides the factors, revealing those which can be discarded as irrelevant to the analysis (Jany et al., 2020). If considered in the profile of the thematic cluster, these factors "would add relatively little to the information already extracted" (Woods and Edwards, 2011, p. 373) . A number of experiments demonstrate that scree plot tests are easily manageable and tend to produce accurate results (Cattell and Vogelmann, 1977; Linn, 1968; Zwick and Velicer, 1982) .

However, reliability issues can surface, leading to an overestimated number of salient topics (Crawford and Koopman, 1979; Zwick and Velicer, 1986) . Aware of the potential bias that the "subjective quality" (Hoyle and Duvall, 2004, p. 305 ) of this technique can generate, the examination of the patterns of decreasing eigenvalues was conducted by overlapping the results of both scree plot tests and parallel analysis (Ledesma et al., 2015; Nebel-Schwalm and Davis, 2011) .

The topic modelling phase concludes with the identification of the core research themes, which were derived by inductive reasoning. This task was completed by examining the groups of keywords and topics of each cluster, as well as their top ten core documents. The core documents of a thematic cluster are the publications with the highest level of centrality.

The centrality of a document in a cluster is directly proportional to its in-degree value, a social network analysis measure which is calculated by combining the number of citations they have received from other publications belonging to the network. Due to their high connectivity, core documents are the main cognitive nodes of a thematic cluster (Meyer and Beiker, 2014; Mora and Deakin, 2019) and provide most of the information describing its J o u r n a l P r e -p r o o f contents. In this investigation, core documents are deployed as a form of data triangulation to improve construct validity.

The last phase of the knowledge discovery process involved interpreting the results of the topic modelling and validating the proposed observations (Kobayashi et al., 2018) . A concise review of each thematic cluster was proposed, in which the three complementary subinformation systems of latent variables identified during the topic modelling phase were linked (i.e. keywords, topics, and research themes).

Finally, four independent experts were tasked with verifying the validity of the extracted knowledge patterns and significance of the contents used to present them. In this study, domain experts are representatives of public or private organizations who have been actively engaged with research activities in the AV sector and have accumulated at least five years of professional experience. This selection criteria made it possible to ensure that the selected experts had a proven knowledge background in AV research. Each domain expert was invited to undertake a one-hour interview. During the interview, they were initially introduced to the analysis and were then asked to provide feedback on the results. A yes/no binary system was adopted to evaluate the extent to which the experts were in agreement with the proposed overall structure and the contents of each cluster. In case of disagreement, comments motivating the answer were collected and used to refine the topic modelling output. When changes were proposed, before being processed, their validity was checked with all other reviewers.

The network graph in Figure 2 is a document citation network which shows how the AV research field is structured by considering the last five decades of scientific publication output and its main thematic research areas. The network is a combination of edges and nodes. The nodes are Scopus-indexed publications. Each node has a diameter proportional to its in-degree centrality. Therefore, the higher the number of citations received by a publication, the larger its diameter in the graph. The citations are represented as edges, whose weight is directly proportional to the number of citations connecting two nodes. This activity has been implemented by using the core literature as the main reference source, together with the data in Table 1 , which visualizes the temporal evolution of the core research themes and shows how their intensity has evolved over the years. The intensity is a measure of the annual publication output of each cluster. The higher the number of publications added to a thematic cluster during a specific year, the higher its intensity.

During the validation process, all reviewers agreed with the conceptual structure of the AV research field. As a result, only a few changes were suggested, but at the cluster level.

These changes aimed at enhancing clarity in the discussion phase. Therefore, the input collected during the validation process has not only generated construct validity evidence, but it has also helped refine the description of the thematic clusters. Table 1 . Temporal evolution of the core research themes: intensity of the publication output by year

The first cluster mainly focuses on the 2007 DARPA Urban Challenge 2 (Broggi et al., 2016) .

This driverless car race has triggered a significant number of studies that build on its outcome to examine the complexity of AV operation in urban environments and propose approaches to modelling, as well as motion planning, for improving AV operations in uncertain, dynamic and un/semi-structured environments. For example, Urmson et al. (2008) introduce the three-layer planning system of Boss, the driverless vehicle which won the first place of the challenge. The Boss is a 2007 Chevy Tahoe with an artificially intelligent mixedmode system combining: (1) a mission planning layer, which creates various options of trajectories towards the destinations; (2) a behavioral layer that decides the moment for lane-changing and simulates error recovery, and;

(3) a layer of motion planning to avoid obstacles.

Lessons learned from Boss are deployed by Ferguson et al. (2008) 

Planning the real-time motion of multi-AV operations is a key challenge (Frazzoli et al., 2002) . This cluster focuses attention on this subject matter of investigation and introduces various control systems and techniques for multi-AV operations in dynamic road environments, such as sensor network and position measurement for collision avoidance.

For example, based on mathematical programming formulations, Schouwenaars et al. (2001) present an approach to planning the trajectories of multiple vehicles to avoid collisions. In addition, Leonard and Fiorelli (2001) contribute with a framework that coordinates a fleet of vehicles by modelling the vehicles as point masses that contain full actuation. The approach stabilizes flocking motions with vehicles' prescribed group geometry and controls the inter-vehicle spacing by using artificial potentials and virtual beacons.

Advancements in real-time motion planning of multi-AV operations continue with Olfati-Saber and Murray (2002) . The framework proposed by Leonard Inspired by the DARPA Urban Challenge, Wongpiromsarn et al. (2012) put forward an approach that can synthesize control protocols automatically. It ensures system correctness for its specification expressed in linear temporal logic in any operational conditions. Besides, a receding-horizon based framework was presented, which can simplify a computational synthesis problem and divide it into smaller, easy-to-solve problems. Further investigation of the robustness of this framework is expected.

Seeing and understanding road conditions is crucial for AV detection and navigation, which depend on the interaction between sensors and AI-empowered systems. This cluster looks at virtual-based techniques and tests for automated driving.

Sensor devices, processing, and fusion algorithms are crucial components of a data fusion system. Important probabilistic modelling and fusion techniques as well as nonprobabilistic data fusion methods are reviewed by Durrant-Whyte and Henderson (2016) . Their research outlines key principles in data fusion architectures from a hardware perspective as well as an algorithmic perspective. It also reports on two examples of applications: (1) a self-tracking application for AV navigation and (2) Research belonging to this cluster also examines various approaches to implementing navigation systems for AVs. For example, using evidential reasoning, Pagac (1998) analyses the issues of building and maintaining a map of the AV environments to improve its navigation performance. The implemented approach allows support for multiple propositions at a time, which differs from the Bayes approach as it only allows a single hypothesis.

Building upon such studies, Desjardins and Chaib-Draa (2011) come up with a novel method to detect and estimate lanes, which relies on the road image captured by a monocular camera. In other words, the key to the success of this algorithm is the robustness of image processing, which deploys techniques such as Probabilistic Hough

Transform, vehicle lateral localization, road marker estimation. This study thus offers a robust system that takes the perspective image as the only data source. Another example is the system designed in the Blind Driver Challenge 3 (BDC), which allows the safe operation of a vehicle by the visually impaired (Hong et al., 2008) . This system shows the potential to enhance mobility for visually impaired people. It can also be extended to assist in driving for other groups of people.

Having benefited from the advancement of the internet and connected technologies, After recognizing traffic signs, AVs need to make decisions. Regele (2008) proposes a modelling method to improve the decision-making process for autonomous vehicles.

Applying a hierarchical world model, it distinguishes a low-level model from a high-level model as the former one plans vehicle trajectories while the latter one coordinates road traffic. The traffic model is expected to be integrated into traffic management.

This cluster gathers social-oriented research on AVs and its attention is focused on the social effects and public acceptance of AV technology. The publications in this cluster also provide an overview of the potential benefits of AV developments to road safety and driving environments, and they point out the challenges that AV integration brings. Although covering many aspects at a high level, in-depth investigations are less diffused.

For instance, Fagnant and Kockelman (2015) draw a brief overview of the technology and the potential social impacts of AVs, and they discuss the challenges for social deployment.

Their study focuses on aspects such as safety, congestion and traffic operation, travel behavior, vehicle ownership, and parking. Barriers to implementation, which are associated with vehicle cost, AV certification, litigation, liability and public perception, security, and privacy are also discussed. Fagnant and Kockelman (2014) Another important aspect of AV societal research is acceptability, which is closely related to people's opinions on AV technology and SAV service. By means of a survey with 5,000 responses from 109 countries, people's preferences and willingness to different types of AV regarding the level of automation are studied by Kyriakidis et al. (2015) . The results show that nearly 69% of people believe AVs with full automation will reach half of the market share between now and 2050, however, various concerns are revealed at the same time. These concerns are related to aspects such as safety, data privacy, and AV legislation. It enriches stakeholders' understanding of public opinions on AVs and contributes to market strategies.

Likewise, user preference is widely studied and analyzed in multiple regions and cultures, for instance, regional differences between Israel and North America (Haboucha et al., 2017) are explained through the study of user preference of AVs, which may inspire regional policy making in the future.

Some researchers also look at the willingness to pay for AV/SAV services (Bansal et al., 2016; Krueger et al., 2016) . For example, a survey conducted in Austin indicates that people perceive a decrease in car accidents to be the primary benefit and equipment failure as the top concern. The study also finds that participants are more willing to pay for the service that can add a higher level (level 4) of automated technology to their current car than adding comparatively lower automation (level 3).

However, apart from discussing broad social impacts and challenges, these studies are mainly centered around evaluating and improving social acceptance. The marketing strategies and implications from such studies thus imply a research driven out of commercialization. In-depth investigations of other non-technical implications of autonomous-driving technology, such as accessibility, affordability and liability, are largely missing.

This cluster explicates different types of challenges of human-robot (automated vehicles) interaction and discusses some ethical dilemmas.

Five major challenges of human factors research on automated vehicles are pointed out by Sheridan (2016) : (1) a task analysis that considers environmental, economic and other potential factors;

(2) the avoidance of accidental consequences; (3) the mutual models and J o u r n a l P r e -p r o o f shared features between robots and humans; (4) robotic applications for education; and (5) strategies for managing users' concerns and considerations caused by cultural or value difference. In the context of automated driving, complex situations are discussed such as pedestrian behavior (Chang et al., 2017; Rothenbucher et al., 2016) , challenges that a driver with autopilots experience on the roads (Brown and Laurier, 2017), and moral dilemmas in vehicle crash scenarios (Lin, 2015) . Goodall (2014) and Lin (2015) suggest that, even in ideal conditions, automated vehicles cannot always avoid being involved in crashes, and the AV decision that precedes certain crashes has a moral and ethical component. Lin introduces scenarios such as the trolley problem that implicate ethics and illustrate the complexity of AV decision making since this process goes beyond mechanically obeying the existing traffic rules. These studies highlight the importance of ethics for AVs and encourage methods from various disciplines to tackle these challenges. Gerdes and Thornton (2015) attempted to find a mathematical way to pin down the philosophical ethical considerations of AV and address them accordingly by offering better choices of steering, braking, or accelerating under certain circumstances. Efforts on translating between philosophical concepts and mathematical equivalents contribute to simple implementations of ethical rules, however, they simplify the real-world complexity. As suggested by Goodall (2014) , human morality can hardly be encoded into AVs.

To understand and increase users' acceptance and adoption, Pettersson and Karlsson (2015) introduced two methodologies to explore the user's reaction and expectation. The first one encompasses techniques such as interviews while the second one mediates a shift of views over time through setting expectations of the AV use. For a similar purpose of studying pedestrian's reaction and expectation, a breaching experiment was designed and conducted at Stanford (Rothenbucher et al., 2016) , which used a faux driverless car as an intervention in the real-world setting. This study contributed to a new method to investigate interactions between pedestrians and driverless vehicles, and it provided insights on pedestrian behavior and pedestrians' expectations of encountering driverless vehicles.

This cluster mainly discusses testing methods of autonomous driving and cybersecurity risks, which are introduced in an overview by . Their work also covered topics such as the functional testing and verification of AVs and the validation of AV systems.

The core literature also demonstrates some novel ideas and methods. For example, Kalra and Paddock (2016) call for adaptive policies, by pointing out that AVs need to be driven up J o u r n a l P r e -p r o o f to billions of miles to demonstrate their functional feasibility. In addition, applied the Satisfiability Modulo Theory (SMT) and developed an AV verification system, which focuses on making safe decisions based on image management.

Testing the vision-based control systems of AV is a complicated task. To tackle this, Behere and Törngren (2016) described a functional reference architecture for AV operation and explicate several considerations that may affect it. The functions of such architecture do not rely on specific implementation technologies, rather, they are logistically described. The study investigates two aspects, first, how do implementation technologies affect functional architectures, and second, how does the fact of replacing human drivers with computers affect the architectures. Furthermore, the study suggests that to incorporate the processes of such deployment, it is essential to speed up the testing and verification.

Some types of risks around AV operation also trigger discussions. For instance, looked at cybersecurity issues. In particular, they aim to design facial biometric systems that can identify certain attacking behaviors and attackers who try to evade recognition. These systems have been widely applied for surveillance and regulation.

Autonomous Vehicle Storage and Retrieval System (AVS/RS) represents an advanced alternative to the traditional automated storage and retrieval systems. AVs operate as storage or retrial devices while also being able to transfer loads out of the storage racks. The superiority of this new system is that AV systems can match the size of vehicle fleets as well as the number of lifts to the storage system's transaction demand. This cluster focuses on the development and evaluation of AVS/RS, which shows a constant effort in enhancing the technological feasibility of AVs.

Through opportunistic interleaving, a network queuing model was used to evaluate the AVS/RS performance measures (Fukunari and Malmborg, 2009) . This model contains the potential to provide an important component of a decision support system to conceptualize AVS/RS. At the same time, it can combine modelling of cost and resource requirements.

Taking inspiration from the network queuing models, Roy et al. (2012) designed a semi-J o u r n a l P r e -p r o o f opened system to estimate one AVS/RS layer's design trade-offs. After testing, the model was proven to help quantify the trade-offs and such a result further implies its impact on reducing transaction time of AVS/RS. The computationally efficient cycle time model (Fukunari and Malmborg, 2008) is essential in an AVS/RS as it is proved to be useful for the accurate system conceptualization. The model also enables a comparison between the performances of AVS/RS and the traditional AS/RS (Kuo et al., 2007) .

Simulation-based experimental designs were further conducted for AVS/RS studies (Ekren et al., 2010) and strategies are tested in practical projects, for example, a regression analysis of an AVS/RS rack configuration is demonstrated to build warehouse configurations that deploy AVS/RS and AS/RS alternately (Ekren and Heragu, 2009; Zhang et al., 2009 ).

Likewise, a state equation model was introduced by Malmborg (2002 Malmborg ( , 2003 to estimate the usage of dual command cycles in an AVS/RS. Using interleaving, it empowers users with a clear understanding of the computational complexity as well as a rational consideration of the model's accuracy in an early stage of its development.

Autonomous-driving technology has the potential to radically change the automotive industry and generate the system innovation which is needed to boost sociotechnical transitions to a sustainable transportation sector. Driven by the desire to unleash its innovation potential, a fast-growing interest in AV research has manifested across academic disciplines, which has resulted in a sudden increase in the volume of scientific publications. AV-related scientific knowledge is produced and accumulated at a very fast pace. As a consequence, the likelihood of incurring information overload is particularly notable for AV researchers, who can struggle to overcome the gap between requirements for and capacity of information processing.

Inspired by information granularity studies and mixed-methods bibliometric investigations, this paper suggests addressing this issue by adopting a multi-granulation approach to latent knowledge discovery and synthesis in large-scale research domain. The proposed methodology combines citation-based community detection methods and topic modeling techniques to: (1) extract the relevant semantic structures (i.e. main keywords, emergent topics, and core research themes) hidden in large data-rich environments; and (2) use these sub-information systems of latent variables to give a concise account of how the intellectual structure of research field is conceptually structured.

The proposed methodological approach has been successful in providing a synthetized view of the scientific knowledge produced during five decades of AV research . In addition, this study has made it possible to discover that AV research has seriously overlooked the wide-ranging sustainability implications of autonomous-driving technology.

As a result, "the discussions and studies extrapolating AVs technical aspects, by inserting them in a dynamic environment with several agents and implications, are far from being exhausted" (Gandia et al., 2019, p. 22) .

Overall, the findings of this study show that AV research is mainly technology-driven and is much more oriented towards examining the technological developments needed to enable the widespread rollout of AVs, rather than exploring the socio-economic, environmental, cultural, political, institutional, and organizational dimensions of a future sociotechnical transition to sustainable transport systems. The research investigating the non-technical aspects of AVs is significantly underdeveloped when compared to technology-related dimensions. As a result, the current status of AV research exposes a serious lack of attention to the sustainability of large-scale AV deployment. This gap probably explains why no thematic clusters strongly related to sustainability research have been identified. During the last fifty years, little research has been conducted which attempts to assess the sustainability implications of AVs, and this limited effort represents an exclusion of the utmost importance.

When looking at the annual intensity of each core theme, the findings show that the thematic clusters CL.10 and CL.11, which are mainly associated with social sustainability aspects, have grown significantly during the last four years, especially in comparison to the technology-related clusters of the network, where it is possible to assume that a higher degree of maturity has been reached. But the content analysis demonstrates that most of the research is only focused on user experience studies exploring market dynamics and how public acceptance of AV solutions can be enhanced. This suggests that the research focusing on the social and economic sustainability of a potential sociotechnical transition to

AVs is not only scarce, but it is also mainly driven by a market-oriented approach. Despite being pivotal in the search for a sustainable approach to AV deployment in urban J o u r n a l P r e -p r o o f environments, broader topics such as accessibility, affordability, liability, trust, business models, travel behavior, political and cultural implications, and socio-economic impacts remain unexplored. The solutions to some of the most relevant barriers to AV implementation and mass-market penetration relate to these unexplored areas of research (Fagnant and Kockelman, 2015; Hess, 2020) .

In addition, the analysis shows that social and economic implications are not the only The politics of regime change is an additional factor that plays a major role in building the institutional and regulatory systems needed to sustain sustainability transition contexts (Gonzalez de Molina, 2013) , but it also represents an explored area of research when observing the AV research field. Sociotechnical transitions require political action to readjust public policies, so that they reflect more sustainable trajectories (Goyal and Howlett, 2020) and arbitrate when competing propositions, resistance, and powers struggles hinder the transformation (Wironen and Erickson, 2020). As a result, formal institutional frameworks are The project Aramis is emblematic of the importance that the politics of regime change has in sustainability transitions. Launched in the 1970s, Aramis attempted to introduce a revolutionary Personal Rapid Transit (PRT) system in Paris. This government-funded experimental project was expected to improve road capacity by placing small cars on an automated highway system and using them as trains. The system was never completed, and research investigating the development process has proved that inconsistent political support was one of the major failure-causing issues, together with a serious lack of strategic planning (Latour, 1996) . In addition, when looking at changes in organizational settings regulating the functioning of transport systems, it is also important to highlight that AV-related sustainability studies are also required for exploring how AV development will be affected by the Covid-19 pandemic and what new barriers and opportunities have been brought to light. For example, while people are forced to observe social distancing measures to prevent the virus for spreading, AVs have been deployed for non-contact, low-speed delivery services, in particular in areas that were subject to lockdown measures (Okyere et al., 2020) . This approach to deployment shows that driverless technology may provide an opportunity to reduce biosafety risks. In addition, the pandemic has caused a drop in transit ridership, making it difficult for public transport systems to address transport needs in some urban areas (Short et al., 2020; Wang et al., 2020) . The pressure exerted by this unfortunate event may help accelerate the development of mass autonomous transit systems, by lowering the existing resistance to change (Zeng et al., 2020). However, shrinking economies and health concerns may also challenge the AV industry, by reducing funding opportunities and slowing down real-world testing. How self-driving technology development and deployment will be reorganized in the post Covid-19 era is another subject matter worthy of investigation.

This methodological approach has been effective in achieving the proposed research objectives, but it is important to acknowledge the presence of a number of methodological limitations, which themselves present future research opportunities. Second, this study only focuses attention on peer-reviewed publications. Therefore, AVrelated grey literature was not taken into account. It would be interesting to evaluate whether this type of literature, which is not subject to a formal peer-review process, has influenced the academic debate and the shaping of the intellectual structure of the AV research field.

Finally, the evaluation phase of the topic modelling output can be further enhanced. This phase of the analytical process has proven successful in generating construct validity evidence and input for refining the description of the thematic clusters. However, due to resource constraints, validity was ascertained by means of a limited number of domain experts and the interrater reliability was measured by considering qualitative rather than qualitative approaches, which would provide more robust measures. Therefore, additional research involving large-scale data collection tools would be beneficial for further testing and refining the results of the topic modelling. 

Testing advanced driver assistance systems using multi-objective search and neural networks

Testing Vision-Based Control Systems Using Learnable Evolutionary Algorithms

Internet of vehicles: From intelligent grid to autonomous cars and vehicular clouds

Springer Handbook of Science and Technology Indicators

Using hybrid methods and 'core documents' for the representation of clusters and topics: the astronomy dataset

Information management and improvement of citation indices

Agroecology and Politics. How To Get Sustainability? About the Necessity for a Political Agroecology

Ethical Decision Making during Automated Vehicle Crashes

Who learns what in sustainability transitions?

Driverless Technologies and Their Effects on Insurers and the State: An Initial Assessment

Parallel and diagonal parking in nonholonomic autonomous vehicles

Incumbent-led transitions and civil society: Autonomous vehicle policy and consumer organizations in the United States

Development of a semi-autonomous vehicle operable by the visually-impaired

A rationale and test for the number of factors in factor analysis

Determining the Number of Factors in Exploratory and Confirmatory Factor Analysis

Finding multiple lanes in urban road networks with vision and lidar

Autonomous vehicles testing methods review

Safety Verification of Deep Neural Networks

Obstacle detection for self-driving cars using only monocular cameras and wheel odometry

Seeking protection from precarity? Relationships between transport needs and insecurity in housing and employment

Large-scale mapping in complex field scenarios using an autonomous car

Overcoming barriers to adoption of environmentally-friendly innovations through design and strategy

Navigation, guidance and control of an overactuated marine surface vehicle

Preliminary factor and psychometric analysis of the Motivation for Fear (MOTIF) survey

A technical and environmental comparison between hydrogen and some fossil fuels

Positive externalities of an epidemic: The case of the coronavirus (COVID-19) in China

Flocking for Multi-Agent Dynamic Systems: Algorithms and Theory

Distributed cooperative control of multiple vehicle formations using structural potential functions

An evidential approach to map-building for

Part I: Externalities and economic policies in road transport

Mixed integer programming for multi-vehicle path planning

Accessorize to a Crime: Real and Stealthy Attacks on State-of-the-Art Face Recognition

ACM SIGSAC Conference on Computer and Communications Security

Human-Robot Interaction

Public Transport and COVID-19: How to Transition from Response to Recovery

Innovation studies and sustainability transitions: The allure of the multi-level perspective and its challenges

Intelligent autonomy for unmanned naval systems

Deep interventions for a sustainable transport future

Driving in the Dark: Designing Autonomous Vehicles for Reducing Light Pollution

Real-time motion planning with applications to autonomous urban driving AR 66

Path planning for autonomous vehicles in unknown semi-structured environments AR 58

Intelligent vehicles BC 58

Lane-change fuzzy control in autonomous vehicles for the overtaking maneuver AR 46

Motion planning in urban environments AR 46

Motion planning in urban environments BC 43

Autonomous driving in urban environments: Approaches, lessons and challenges AR 43

MPC-based approach to active steering for autonomous vehicle systems AR CL

Real-time motion planning for agile autonomous vehicles AR 57

Real-time motion planning for agile autonomous vehicles CP 57

Real-time motion planning for agile autonomous vehicles CP 43

Virtual leaders, artificial potentials and coordinated control of groups CP 37

Mixed integer programming for multi-vehicle path planning CP 32

Flocking for multi-agent dynamic systems: Algorithms and theory AR 23

Receding horizon temporal logic planning AR 16

Nonholonomic source seeking with tuning of angular velocity AR CL

Preparing a nation for autonomous vehicles: Opportunities, barriers and policy recommendations AR 131

The travel and environmental implications of shared autonomous vehicles, using agent-based model scenarios AR 65

Preferences for shared autonomous vehicles AR 63

Public opinion on automated driving: Results of an international questionnaire among 5000 respondents AR 51

Auction-based autonomous intersection management CP 35

Assessing public opinions of and interest in new vehicle technologies: An Austin perspective AR 29

A multiclass cell transmission model for shared human and autonomous vehicle roads AR CL

The social dilemma of autonomous vehicles AR 44 Goodall N. 2014 Ethical decision making during automated vehicle crashes AR 32

Why ethics matters for autonomous cars BC 27

Ghost driver: A field study investigating the interaction between pedestrians and driverless vehicles CP 13

Implementable ethics for autonomous vehicles BC 11 Pettersson I., Karlsson I.C.M. 2015 Setting the stage for autonomous cars: A pilot study of future autonomous driving experiences AR 10 Brown B., Laurier E

Social LSTM: Human trajectory prediction in crowded spaces CP

Communicating awareness and intent in autonomous vehicle-pedestrian interaction CP

Eyes on a car: An interface design for communication between an autonomous car and a pedestrian CP

Learning a deep neural net policy for end-to-end control of autonomous vehicles CP

The social behavior of autonomous vehicles CP CL

Driving to safety: How many miles of driving would it take to demonstrate autonomous vehicle reliability?

Safety verification of deep neural networks CP 13

Coordinating hundreds of cooperative, autonomous vehicles in warehouses AR 13

Coordinating hundreds of cooperative, autonomous vehicles in warehouses CP 12

A functional reference architecture for autonomous driving AR 10

Autonomous vehicles testing methods review CP 9

Testing advanced driver assistance systems using multi-objective search and neural networks CP

Testing vision-based control systems using learnable evolutionary algorithms CP

Accessorize to a crime: Real and stealthy attacks on state-of-the-art face recognition CP

DeepTest: Automated testing of deep-neural-network-driven autonomous cars CP

Conceptualizing tools for autonomous vehicle storage and retrieval systems AR 31

Design models for unit load storage and retrieval systems using autonomous vehicle technology and resource conserving storage and dwell point policies AR 24

Variance-based approximations of transaction waiting times in autonomous vehicle storage and retrieval systems AR 23

A network queuing approach for evaluation of performance measures in autonomous vehicle storage and retrieval systems AR 23

An efficient cycle time model for autonomous vehicle storage and retrieval systems AR 23

Interleaving dynamics in autonomous vehicle storage and retrieval systems AR 18

Simulation based experimental design to identify factors affecting performance of AVS/RS

Performance modelling of autonomous vehicle storage and retrieval systems using class-based storage policies AR

Performance analysis and design trade-offs in warehouses with autonomous vehicle technology AR

Analytical model to estimate performances of autonomous vehicle storage and retrieval systems for product totes AR

Simulation based regression analysis for rack configuration of autonomous vehicle storage and retrieval system CP

article; BO: Book; BC: Book chapter

Steering Control; Steering Angle; Steering Wheel; Front Steering; Wheel Steering; Steering Controller

Vehicle Wheels; Vehicle Dynamics; Automobile Parts and Equipment

Decision; Behavioral; Make; Human; Interaction; Behavior; Research; Action; Decision Making

Markov Decision Process

Image; Vision; Processing; Camera; Detection; Computer; Computer Vision; Image Processing

Obstacle Detection; Optical Radar; Vision System; Detection and Tracking; Obstacle Detector; Stereo Vision; Lane Detection; Computer Graphic

Control system Control; Controller; Lateral; Design; Track; Simulation; Predictive; Steering; Control System

Driver Assistance System; Human Driver; Driver Model; Driver Behavior 2.59 1334 39.42% Uncertainty analysis Uncertainty; Disturbance; Robustness; Robust; Uncertainty Analysis

Disturbance Rejection; External Disturbance; Robust Controller

Advanced; Challenge; Research; Describe; Environment; Urban Environment; Urban Challenge; DARPA Urban Challenge; Urban Planning; Grand Challenge; Urban Traffic 2.26 1441 55.69% Optimization problem Problem; Solve; Optimization; Optimal; Programming; Solution; Optimization Problem; Optimal Control; Control Problem; Optimal Trajectory

Real-time motion planning of multi-AV operations Position, velocity and convergence Convergence; Constant; Seek; Velocity; Local; Numerical; Modeled; Position; Signal; Position and Velocity; Angular Velocity; Convergence of Numerical Methods

Motion planning Planning; Trajectory; Path; Motion; Constraint; Programming; Compute; Optimization; Planner; Optimal; Motion Planning; Path Planning; Motion Control

Stability analysis Loop; Lyapunov; Stability; Close; Law; Stability Analysis; Control Law; System Stability; Closed Loop; Lyapunov Method; Control Theory; Closed Loop Control; Graph Theory; Loop System

Unmanned aerial vehicle Unnamed; Aerial; Unmanned; Aircraft; Air; Flight; Unmanned Aerial Vehicle; Unmanned Vehicle; Aerial Vehicle; Aircraft Control; Air Navigation; Unmanned Autonomous Vehicles

Intelligent Vehicle-Highway System Traffic; Highway; Transportation; Intelligent; Safety; Road; Drive; Intelligent Vehicle; Intelligent Vehicle Highway; Intelligent Transportation; Intelligent Robot; Roads and Streets

Sensor network Detection; Data; Sensor; Environmental; Map; Search; Development; Sensor Network; Environmental Monitoring 2.80 412 40.77% Collision avoidance Avoidance; Collision; Obstacle; Avoid; Free; Collision Avoidance; Obstacle Avoidance

Multi-sensors and fusion systems Road Road; Street; Lane; Transportation; Traffic; Road and Street; Traffic Control; Intelligent Transportation System

Lidar Lidar; Optical; Radar; Cloud; Point; Optical Radar; Point Cloud; Lidar Data; Optical Flow; Lidar Sensor; Light Detection and Ranging

Neural; Convolutional; Network; Deep; Train; Learning; Dataset; Neural Network; Deep Learning; Convolutional Neural Network; Machine Learning; Learning System; Deep Neural Network; Learning Algorithm

Stereo image processing Stereo; Image; Camera; Processing; Dense; Estimate; Estimation; Map; Vision; Visual; Compute; Match; Image Processing; Computer Vision; Stereo Image Processing

Global; Localization; Location; Trajectory; Positioning; System; Position; Mobile; Estimation; Mobile Robot; Motion Estimation; Autonomous Mobile

Virtual-based testing Virtual; Reality; Agent; Behavior; Rule; Modeled; Coordinate; Surface; Virtual Reality; Autonomous Agent; Virtual Environment

Motion planning Planning; Path; Motion; Avoidance; Collision; Compute; Constraint; Obstacle; Motion Planning; Path Planning; Collision Avoidance; Obstacle Detection; Obstacle Avoidance; Obstacle Detector

Autonomous car drive Car; Drive; Driverless; Driving; Autonomous Car; network Neural; Deep; Network; Train; Learning; Dataset; Neural Network; Deep Learning; Convolutional Neural Network; Deep Neural Network; Learning System

Liadar Radar; Optical; Lidar; Cloud; Point; Light; Optical Radar; Point Cloud; Light Detection and Ranging

Slam; Simultaneous; Mapping; Localization; Robotic; Vehicle Localization; Localization and Mapping; Simultaneous Localization and Mapping; Localization Method; Localization Accuracy; Localization System

Localization Error

Road marking detection Marking; Street; Road; Mark; Lane; Road and Street Marking; Lane Marking; Road Surface; Road Marking 3.21 1222 47.72% Safety Technology; Future; Development; Develop; Safety; Research; Accident Prevention

Automotive Industry

Global; Inertial; Positioning; System; Position; Accurate; Global Positioning System; Inertial Navigation System; Vehicle Position; Global Navigation Satellite System; Inertial Measurement; Inertial Sensor; Position Estimation; Inertial Measurement Unit

Autonomous car drive Driver; Assistance; Advance; Automobile; Advanced Driver Assistance System

Stereo image processing Vision; Image; Camera; Computer; Monocular; Visual; Stereo; Processing; Computer Vision; Image Processing

Motion planning Planning; Path; Motion; Motion Planning; Path Planning; Motion Estimation; Path Planner; Highway Planning; Local Path; Path Tracking; Control System

Proposed Method; Detection Method 2.41 775 87.83% Odometry Scale; Large; Odometry; Outdoor; Collect; Visual Odometry; Large Scale

Motion planning for agricultural machinery Scene segmentation Segmentation; Classification; Scene; Outdoor; Visual; Ground; Terrain; Perception; Natural; Feature; Selection

Unmanned; Operating; Outdoor Environment; Perception System; Unmanned Vehicle; Image Segmentation

Image processing Image; Vision; Camera; Detect; Processing; Detection; Stereo; Detector; Computer; Machine; Row; Computer Vision

Machine Vision; Image Processing

Path tracking Straight; Average; Curve; Steering; Proportional; Guidance; Angle; Error; Equipment; Path; Successfully; Develop; Path Tracking; Automobile Steering Equipment; Guidance System

Control Velocity; Orientation; Adaptive; Relative; Nonlinear; Linear; Feedback; Trajectory; Curvature; Distance; Follow

Autonomous Vehicle; Control Approach; Tracking Control; Control System

Agricultural machinery Agricultural; Agriculture; Precision; Machinery; Increase; Farm; Agricultural Machinery; Agricultural Vehicle; Precision Agriculture; Agricultural Robotics; Agricultural Environment

Vehicle behavior Behavior; Solve; Problem; Plan; Nonholonomic; Mobile; Practical; Constraint; Deal; Robot

Assistance; Driver; Advance; Advanced; Automobile; Technology; Case; Advanced Driver Assistance System

Automobile Driver

Fuzzy control Controller; Fuzzy; Logic; Simulation; Design; Proportional; Tune; Fuzzy Control; Control System

Simulation Result; Fuzzy Logic Control; Controller Design; Autonomous Vehicle Control computing Cloud; Distribute; Computation; Edge; Complexity; Unit; Computing; Require; Assist; Edge Computing

Hoc; Ad; Vehicular; Network; Communication; Lead; Vehicle to Vehicle Communication; Vehicular Ad Hoc Network

Network Security; Millimeter Wave

Lane detection Image; Detection; Lane; Detect; Camera; Vision; Transform; Condition; Line; Extract; View; Edge; Road; Computer Vision; Road and Street; Lane Detection; Vision System

Thing; Internet; Service; Smart; Quality; Cloud; Life; City; Internet of Things; Internet of Vehicles

Control Verify; Lateral; Angle; Controller; Introduce; Reference; Modeled; Track; Good; Follow; Lane; Side; Steering; Comfort

Automobile Steering Equipment; Lane Detection; Lateral Control; Computer Vision; Hough Transform

Cybersecurity Security; Cyber; Attack; Cooperative; Safety; Secure; Physical; Connect; Network Security

Kalman filter Filter; Kalman; Estimation; Estimate; Method; Image; Kalman Filter; Kalman Filtering; Image Processing; State Estimation; Feature Extraction; Image Segmentation

Traffic; Behavior; Transportation; Street; Safety; Capability; Traffic Control; Traffic Congestion

Road Traffic; Autonomous Car

Spatial; Temporal; Sample; Resolution; Sampling; Oceanography; Scale; Data; Observation; Spatial and Temporal

Underwater intervention Manipulator; Intervention; Recovery; Submersibles; Man; Knowledge; Object; Project; Learning; Class; Equip

Demonstration; Dock; Highlight; Open; Recent; Exist; Float; Free; Human; Address; Survey; Task; Capability; Future

Mix; Initiative; Support; Team; Laboratory; Heterogeneous; Operational; Command; Type; Air; Include; Number

Infrastructure; Technology; Management; Framework; Requirement; Command and Control

Motion planning Numerical; Derive; Drive; Modeled; Wind; Efficient; Scheme; Methodology; Speed; Level; Finally; Dynamic; Method; Energy; Presence; Path

Model predictive control Formation; Decentralize; Predictive; Nonlinear; Action; Constrain; Operative; Local; Computational; Model; Avoid; Strategy; Constraint; Formation Control; Operative Control

Kalman filtering Filter; Localization; Kalman; Position; Accuracy; Measurement; Fusion; Navigation; Error; Measure; Extend; Kalman Filter

Sonar obstacle detection Detect; Autonomously; Detection; Advantage; Link; Forward; Combine; Moor; Robust; Map; Forward Looking Sonar avoidance Reach; Planning; Avoid; Goal; Unknown; Path; Motion; Controller; Behavior; Obstacle; Variety; Environment; Function

Obstacle Detection; Obstacle Detector; Obstacle Avoidance; Motion Planning; Unknown Environment 28

Saccadic vision Saccadic; Expectation; Action; Perception; Capability; Hierarchical; Hardware; Representation; Active; Multi; Mission

Architecture; Knowledge; Perform; Decision; Head; Control; Complex

Night-time operativity Night; Effectiveness; Light; Procedure; Feasibility; Segmentation; Locate; Front; Condition; Robustness; Fast; Estimate

Line detection Edge; Mark; Marking; Curve; Street; Extraction; Detection; Fit; Width; Road; Interest; Lane; Region; Stage; Extract

Lane Detection; Detection Algorithm; Vehicle Detection; Edge Detection; Image Segmentation; Road and Street Marking cities Future

Hoc; Attack; Intrusion; Ad; Vehicular; Security; External; Semi; Communication; File; Service; Cooperative; Network

Neural Network; Vehicular Ad Hoc Network; Network Security; Vehicle to Vehicle Communication; External Communication

Traffic sign recognition Color; Recognition; Region; Candidate; Image; Traffic; Recognize; Light; Shape; Segmentation; Classifier; Sign; Classification; Method; Feature; Detection; Gradient

Semantic; Relationship; Aid; Language; Mobility; Infrastructure; Platform; Capture; Motor; Context; Key; Simple

Ontology; Domain; Dynamic; Map; Concept; Scene

Deep; Convolutional; Learning; Neural; Training; Dataset; Network; Train; Classifier; Detector; Prove

Deep Learning; Machine Learning; Convolutional Neural Network; Vehicular Ad Hoc Networks

Motion planning Path; Planning; Avoid; Collision; Motion; Obstacle; Motion Planning

Intelligent Vehicle; Intelligent Vehicle Highway System; Advanced Driver Assistance System

Intelligent Transportation System; Vehicle Control System

Social impacts and integration of AVs Intersection management Intersection; Delay; Stop; Control; Signal; Management; Collision; Cross; Traffic; Protocol; Propose

Traffic Congestion; Control System; Traffic Management; Intersection Control; Street Traffic Control

Demand; Fleet; Service; Share; Operation; Mobility; Size; Ride; Trip; Urban; Transport; City; Travel

Transport Vehicle; Shared Autonomous Vehicle; Autonomous Mobility

Acceptance Perceive; Acceptance; Survey; Factor; Perception; People; Influence; Participant; Trust; Affect; Public

Risk Perception; Stated Preference; Technology Acceptance; Public Attitude; Public Road; Public Transportation

Optimization issues Programming; Linear; Problem; Program; Solve; Optimization; Optimal; Constraint; Schedule; Solution; Minimize

Optimal Control; Integer Linear; Mixed Integer; Optimization Problem; Control Problem; Integer Linear Program; Linear Programming; Optimal Solution; Numerical Experiment

Human-computer interaction Interaction; Human; Trust; Machine; Participant; Task; Design; Computer; Simulator

Human Engineering; Human Factor; Human Driver; demand Estimate; Travel; Choice; Trip; Travel Time; Travel Behavior; Travel Demand; Mode Choice; Shared Autonomous Vehicle

Ethical; Ethics; Philosophical; Moral; Aspect; Dilemma; Argue; Make; Decision; Principle; Legal; Situation; Philosophical Aspect; Ethical Decision; Make Decision; Moral Dilemma

Deep; Neural; Network; Camera; Learning; End; Steering; Image; Visual; Learn; Vision; Performance; Input; Deep Learning; Neural Network

Public concern Concern; World; Technology; Future; Public Concern 4.08 150 43.80% Motion planning Avoidance; Obstacle; Collision; Path; Motion; Planning; Algorithm; Simulation; Navigation; Collision Avoidance; Motion Planning; Navigation System

Human; Man; Machine; Interaction; Trust; Robot; Computer; Engineering; Interact; Human Factor 3.30 1374 72

Testing and risk assessment Verification and validation Verification; Correctness; Verify; Decision; Property; Formal; Respect; Check; Make; Tool

Neural networks and deep learning Neural; Deep; Image; Input; Network; Adversarial; Technique; Learning; Robustness; Camera; Training; Recent

Deep Neural Network; Deep Learning; Machine Learning; Learning System

Testing Testing; Test; Generation; Automatically; Reality; Virtual; Drive; Automatic; Demonstrate; Car; Software; Generate

Autonomous Driving; Software Testing; Driving Car; Software Engineering; Safety Testing

Graph; Bond; Modeled; Dynamic; Wheel; Theory; Model; Fault; Deal; Validate; Intelligent; Intelligent System; Intelligent Autonomous Vehicle; Bond Graph; Graph Theory

Artificial; Intelligence; Machine; Learning; Attack; Security; Physical; Network; Neural; Neural Networks; Artificial Intelligence Attack

Computer vision Computer; Time; Unit; Platform; Program; Real; Processing; Run; Vision; Computer Vision

Graphics Processing

Risk assessment Risk; Assessment; Safety; Hazard; Run; Support; International; Automotive; Situation; Safety Engineering; Safety Critical Systems; Automotive Systems; Risk Assessment; Vehicle Safety

Cyber-physical systems Cyber; Physical; Embed; Smart; Virtual; Embedded System; Virtual Reality; Cyber Physical System 3.25 192 33.00% Automobile steering equipment Steering; Equipment; Track; Automobile; Wheel; Path; Automobile Steering Equipment

Automated Storage and Retrieval System (AVS/RS) Transport logistics Pick; Production; Shuttle; Move; Order; State

High-density storage areas Density; High; Transfer; Area; Flexibility; Effect; Aisle; Detail; Capacity; Throughput; Cycle; Parameter; Address; Location; Vertical; Warehouse; Tier; Unit; High Density Storage Area

Arena; Commercial; Software; Average; Complete; France; Rack; Number; Configuration; Variable; Study; Determine

Vehicle movement Horizontal; Vertical; Lift; Analyze; Movement; Travel; Insight; Network; Semi; Solve; Transaction; Decomposition; Tier; Improve; Queue

Unit Load Automated Storage & Retrieval System Unit; Load; Design; Automate; Technology; Transaction; Queuing Network

Agent-based simulation Environment; Agent; Order; Recent; Dynamic; Implement; Efficient; Tool; Flexibility; Agent Based Simulation

Rail-guided vehicles Guide; Rail; Tool; Problem; Include; Propose; Address; Optimal; Rail Guided Vehicles 3.64 64 66.10% Transition cycle-times Time; Cycle; Transaction; Aisle; Cycle Time; Storage and Retrieval 3.29 127 76.27% Queuing network Handle; Material; Technology; Alternative; Automate; Automation; Key; Open Queuing Network; Queuing Network

Durrant-Whyte H., Henderson T.C. 2016 Multisensor data fusion BC 22Menze M., Geiger A. 2015 Object scene flow for autonomous vehicles CP 18Desjardins C., Chaib-Draa B. 2011 Cooperative adaptive cruise control: A reinforcement learning approach AR 14 Pagac D., Nebot E.M., Durrant-Whyte H. 1998 An evidential approach to map-building for autonomous vehicles AR 11Hall D.L., Llinas J. 1997 An introduction to multisensor data fusion AR 10 Pereira J.L.F., Rossetti R.J.F. 2012 An integrated architecture for autonomous vehicles simulation CP 10 Häne C., Sattler T., Pollefeys M. 2015 Obstacle detection for self-driving cars using only monocular cameras and wheel odometry CP 10Al-Shihabi T., Mourant R. R. 2003 Toward more realistic driving behavior models for autonomous vehicles in driving simulators AR 9Moras J., Cherfaoui V., Bonnifait P. 

On the detection of grey hole and rushing attacks in self-driving vehicular networks CP