Governing beyond innovation: Exploring the impact of connected and automated vehicles on the organization of vehicle accident investigations

Abstract Connected and automated vehicles (CAVs) bring new risks and safety benefits to future road transport. This paper explores the impacts and consequences of CAVs on the organizational, governance and knowledge practices of vehicle accident investigation (VAI). A focus on these practices of VAI draws attention to its politics, a key aspect behind accident investigations that receives little attention. We do so through three lines of inquiry. First, we read the accident investigation and safety science literature to list the characteristics of a state-of-the-art ideal-typical VAI organization. Second, we draw from the literature to identify seven challenging characteristics from CAVs to existing VAI designs and practices: automation, mixed traffic, connectivity (including cyber security), data, regulatory gaps, safety culture and international connections. Lastly, we describe the current state of affairs of VAI in the United Kingdom, Japan and the United States, highlighting differences and commonalities. In the conclusion, we bring these lines of inquiry together, conclude that there is a need for further governance work on VAI and CAVs, and list several avenues of future study.


Introduction
Connected and Automated Vehicles (CAVs) are road vehicles that to various extend can communicate with other vehicles, infrastructure or traffic participants and which are equipped with driving systems that take on some (at Level 1 and up) or all of the driving tasks in fully automated vehicles (at Level 5) (Guo et al., 2019;SAE, 2018).CAVs have gained rapid interest over the last decade, with increasing levels of automation in vehicles sold and various higher level automated vehicles tested on public roads in pilot projects.This interest rests primarily on the expectation that connectivity and automation increase road safety as they reduce driver error (Fagnant & Kockelman, 2015;Marchau et al., 2019).CAVs, so the argument goes, do not get tired or distracted nor partake in drugs or alcohol. 1  Nevertheless, CAVs are part of fatal and non-fatal incidents (DSB, 2019;NTSB, 2019NTSB, , 2020)).In this respect, Grabbe et al. (2020, pp. 2-3) nuance the driver error argument by pointing to the complexity of accidents and the multiple causes and factors underpinning them.This includes the uncertainties and potential failures of CAV systems and their integration in line with the literature on driver-takeover moments (Straub & Schaefer, 2019), fender benders (Song et al., 2021) or the real-life road-testing of early models (Kalra & Paddock, 2016).While not disputing that there are risks to life road-testing, proponents of CAVs in turn warn that there is "a cost of waiting for nearly perfect automated vehicles (Kalra & Groves, 2017)" leading authorities to try and find a balance between the risks of this new technology and its development (Shladover & Nowakowski, 2019).
This balance between safety and innovation in transport is not just a matter of risk management but is inherently political and thus a focus of the transport governance literature (Docherty et al., 2018).Among others, a transport governance reading of CAV development would point to the various networks, institutional designs, funding, decisionmaking, distributed responsibilities and public policies that help govern road mobility (e.g. Lee & Hess, 2020;Taeihagh & Lim, 2019).Hardly ever discussed in this line of work are studies focusing on the politics and governance around CAV incidents and responding vehicle accident investigations (VAI) (an exception being: Bissell, 2018).The transport safety literature, similarly, barely discusses the organizational design and policy choices around VAI, let alone how the latter are affected by increasing automation and connectivity (an exception being Stanton et al., between accident investigations and consumer trust (Smart, 2004), which is seen as a challenge for CAV adoption.Secondly, it is noteworthy because CAVs entail dynamic interactive systems where faults and errors are distributed across national and international agents, organizations and technology in transport, energy, software and communication systems, and thus further break with Roed- Larsen andStoop's (2012, p. 1392) traditional summary stating that 'road traffic accidents have generally been treated as single accidents [ … ] investigated by the police'.
The aim of this paper is therefore to offer a first exploration of the potential effects from CAV development on the governance and organizational design of VAI.For this, Sec. 2 introduces a research approach based on a three-part review process: two separate scoping underpinned narrative reviews and one narrative review of three example countries.Section 3 then offers a discussion of the VAI literature to identify an ideal-typical organizational design touching on the definition of accidents, the purpose and methodologies of investigations, and the position of investigation bodies.Section 4 continues with a discussion of the challenges that CAVs could impose on VAI organizations by briefly identifying and discussing seven challenges.Section 5 in turn takes a closer look at the debates and ways that vehicle accident investigations are organized in the United Kingdom (UK), Japan and the United States (US).Section 6 brings these lines together and concludes.

Research approach
Given the explorative nature of this paper, the approach resembles a scoping literature review as proposed by PRISMA (Tricco et al., 2018).A scoping review is used to answer broader research questions and is, among others, well suited to: 'summarize findings from a body of knowledge that is heterogeneous in methods or discipline (Tricco et al., 2018)'.However, for this paper a systematic scoping review along guidelines set by PRIMSA has proven problematic, given that reflections on the organization, institution, mandates and practices of VAI are most often an implicit element in papers focusing on other aspects of road safety, accident investigation methods or accident investigation case studies, let alone that papers touch on this in conjunction with CAVs.Similar to Li et al. (2022), we therefore broke up our topic into multiple review steps that link to our subquestions: 1) What is VAI and how is VAI ideally organized and institutionalized? 2) How are CAVs affecting or expected to affect VAI?And 3) how is VAI organized, evolving and impacted by CAVs in the UK, Japan and the US?
Specifically, for sub-questions 1 and 2 we started following the procedure of a scoping review.For these searches we utilized two separate inclusive yet narrow keyword searches (see below), with subsequent screening and exclusions, across SCOPUS and Web of Science dated from 01-1990 to 09-2021.In both cases, however, we ended up adding missing literature (more than) doubling the initially resulting corpus.That, together with the haphazard discussion of insightful reflections in these corpora, led us to conclude that the best way to present the findings was via narrative review (e.g., a topical general discussion based on pivotal papers known by authors, UAB Libraries, 2020).For the country discussions, given the empirical literature required, we moved straight to narrative review.This is valid as our goal is not to systematically synthesize findings across these literatures, but to argumentatively explore new links between them.In doing so, we aim to confirm whether CAVs challenge VAI institutions and identify future research needs (De Vos & El-Geneidy, 2022).
For the first sub-question, we utilized the following keyword search: (TITLE-ABS-KEY (vehicle OR road OR transport Ã OR car OR traffic OR motor) AND TITLE-ABS-KEY ("accident investigation" OR "collision investigation" OR "crash investigation" OR "safety investigation") AND TITLE-ABS-KEY (branch OR board OR organization OR institute OR policy OR governance OR learning OR funding OR independence)) AND PUBYEAR > 1989 AND (LIMIT-TO(LANGUAGE, "English")).This resulted in 626 results, split 312 (SCOPUS) and 314 (Web of Science).After removal of duplicates 494 articles remained.We subsequently read titles and abstracts to exclude road safety studies (e.g.statistical studies into road safety based on accident data, case studies or discussions of safety methods) and studies on other modes of transport (except when they offer general insights on VAI practices).Instead we focused on papers with discussions on historical reflections and/or trends in safety analyses, international collaboration, and reflections on VAI practices, processes and institutions.This resulted in 29 articles.Next, we utilized backward and forward reference tracing to find missing key texts (e.g.ESReDA, 2015;Gomez Buquerin et al., 2021;Jakobsson, 2011;van Vollenhoven, 2002) and build a final corpus of 65 articles that at least partly offer reflections on trends in VAI organizations, institutions, mandates, practices and international collaboration.
For the second sub-question on VAI and CAVs (Sec.4), we utilized the search chain: (TITLE-ABS-KEY (vehicle OR road OR transport Ã OR car OR traffic OR motor) AND TITLE-ABS-KEY("accident investigation" OR "collision investigation" OR "crash investigation" OR "safety investigation" OR "safety board" OR "incident investigation") AND TITLE-ABS-KEY(CAV OR automated OR automation OR autonomous OR connected OR self-driving)) AND PUBYEAR > 1989 AND (LIMIT-TO(LANGUAGE, "English")).This resulted in 103 results, split 40 (SCOPUS) and 63 (Web of Science).After removal of duplicates 79 articles remained.We subsequently skimmed titles and abstracts to exclude studies on other modes of transport and specific accident investigation methods-except when they offered general insights on CAV challenges.This resulted in 12 semi-relevant papers.We subsequently utilized forward and backward reference tracing to add additional literature on CAV safety and key grey literature (e.g.Alvarez, 2017;DSB, 2019;Read et al., 2022;Stanton et al., 2019), resulting in 41 articles used narratively to identify seven interrelated dimensions where CAVs potentially challenge VAI.
Lastly, for the third sub-question (Sec.5), we only conducted narrative reviews by exploring primary and secondary literature such as legal statutes, websites, blogs and year-reports to sketch the current state of affairs of VAI in three countries with a focus on institutional design, expertise and independence.This too took place late summer of 2021.In terms of country selection we chose the UK, Japan and US.The choice for these countries is both strategic and pragmatic.The US is an acknowledged leader in CAV development and has the oldest permanent, independent, multi-mode accident investigation bureau that has been running internationally recognized CAV accident investigations.The UK similarly aims to be a global CAV developer but is witness to an interesting debate to reorganize its road accident investigation-the only mode of transport that does not have its own permanent branch. 2 Japan lastly is understudied within the literature and has its own particular VAI structures and staff agreements.These three countries thus offer a most-likely (US), least-likely (UK) and underexplored example (Japan).Furthermore, these choices are supported by authors' language proficiency as these debates are often held in the countries' respective languages.

Vehicle accident investigation
In this section we discuss the ideal organization and institutionalization of VAI as per the literature on VAI and road safety.

Defining accidents
The question of what VAI entails starts with the politics of designating an event as an accident.In fact, the term accident itself is already politicized as some prefer the term crash or collisions to indicate preventability (Hagenzieker et al., 2014) or incident to denote events without harm or damage.While the safety literature (Otte et al., 2018) categorizes events as minor accidents (<24 h hospital), severe accidents (>24 h hospital), fatal accidents (within 30 days) and incidents (systematic near-misses with safety learning potential), statutes behind VAI leave the precise criteria to invest in an investigation unnamed.For VAI, the naming is part of the process itself.For instance, the Canadian Transport Safety Board's official objective is to study "selected transport occurrences" without delimitating these occurrences (Canadian Transportation Accident Investigation and Safety Board Act, 1989, 7[1]a).The Dutch Safety Board similarly is authorized to investigate any 'occurrences', except for a few listed exceptions (e.g. law enforcement or military actions), with occurrence defined as an "event causing a person's death or injury, or causing damage to an object or to the environment, as well as an event giving rise to the risk of such a consequence (Kingdom Act, 2004, 1[1]f)."Alternatively, in Finland the events studied are defined annually by the Ministry of Transport and Communications, which ordinary includes all accidents with fatalities but can add special focus areas like heavy duty or e-bikes incidents (Salo et al., 2016).The openness in definition is thus deliberate as it allows safety experts to include new occurrences and start investigations not just based on legal requirements or severity (and public outcry), but on their evaluation of potential consequences, likelihood of recurrence and expected safety learning benefits (HSE, 2004).

Purpose
Accident investigations entail organizational learning from past events (Hale, 1997).It is one of the central characteristics for achieving a more resilient and safer system (Cook & Woods, 2006).Vald es and Comendador (2011) insist that identifying root causes and producing recommendations to prevent reoccurrences are essential to ongoing safety improvement.Another core investigative objective is standardized accident data collection and collation (Ahmed et al., 2019;Imprialou & Quddus, 2019) in order to conduct statistical analyses to minimize risks of future events (e.g.Song et al., 2021).VAI thus differs from risk management approaches in transport safety research, although this is not a hard distinction.
Practically, VAI takes place within broader traffic management frameworks geared to clean up the sites of accidents and have traffic flowing as fast as possible again (Steenbruggen et al., 2012).Within this challenging context, accident investigations generally are divided into six phases: (1) make sure that occurrences are properly and quickly reported; (2) decide which occurrences require further investigations and subsequently allocate resources and time; (3) collect evidence to determine what happened; (4) analyze evidence and develop conclusions as to why it happened; (5) develop effective, specific and applicable follow-up recommendations; (6) and then disseminate those recommendations to prevent the same from happening again (via reports, media, hearings, etc.), with follow-up if necessary (DOE, 2012;ESReDA, 2015).A broadly shared perspective is that investigations do not attribute criminal or financial blame (as in judicial and insurance investigations) but are meant to determine causes and prevent recurrence of accidents (van Vollenhoven, 2002).Final conclusions therefore are shared as recommendations instead of obligations, although some authorities legally have to respond to the reports (e.g., Dutch public authorities, US Department of Transport).

Methodology and expertise
Methodologically, VAI is shifting from a focus on random events, independent cause and effect studies and a microlevel understanding that highlights 'road safety as a lack of safety' to a grounding of road safety in systems thinking (e.g.Salmon et al., 2012).Based within the wider move in safety science to sociotechnical systems, such system perspectives advocate a deeper understanding of road accidents by noting shared responsibilities between interactive road users, vehicles, infrastructure, environment and regulation in complex systems (Larsson et al., 2010).In line with this 2 At the time of revision (November 2022), the UK government has decided to initiate the Road Safety Investigation Branch (DfT, 2022).
shift, accident investigations have expanded to the full complexity of sociotechnical systems and include a focus on agents, institutions, regulations, engineering and technology, materials, company strategies, global developments and wider environmental factors (Le Coze, 2017).This coincides with a growing variety of theoretical and methodological safety and accident investigation frameworks, concepts and methodological models (for overviews see Alvarez, 2017;Grabbe et al., 2020;Stanton et al., 2019).
With accidents becoming more intricate, there is a growing need for analyses that are inter-and multi-disciplinary.Larsen (2004) describes how multi-disciplinary analyses give detailed knowledge of factors leading up to accidents and are employed in a growing number of countries.Inter-and multi-disciplinarity helps investigators become aware of epistemological and ontological differences in their approaches, which enhances their 'creativity and ability to "see" what we do not know-and therefore, generate new assumptions and new models' (ESReDA, 2009, p. 17).However, with inter-and multi-disciplinarity come increased demands on investigator's expertise and education, which ultimately determine the quality and validity of investigations.Given increasingly complex system-level events, modern accident investigator teams must have a range of competencies across disciplinary fields beyond engineering and safety science, like psychology, law, political science, organizational theory and ergonomics (Cedergren & Petersen, 2011), as well as skills in leadership, communication, victim support and so on (Nixon & Bentley, 2006).Generally, expertise is a frequently mentioned challenge to accident investigations together with a lack of resources and time (ESReDA, 2015).To address this, VAI organizations regularly invite external experts on a case-by-case instance to help investigations while trying hard to minimize conflict of interest concerns.

Institutional design
Both the active learning and expertise requirements have led to calls for the establishment of permanent multi-modal investigatory bodies (Stoop et al., 2017;Stoop & Dekker, 2012;van Vollenhoven, 2002).For example, Baxter (1995) insists that one institution responsible for investigations of all transport modes is the most efficient way to conduct accident investigations because it enables for the sharing of administrative and training capacities, the creation of a pool of specialist expertise and an internal learning capacity around investigative techniques and best practices across transport modes.It also allows for a scaling of staff in case of major events as the basic evidence gathering practices (interviews, etc.) are comparable across transport modes.Of course, the advantages of multi-modal investigation boards only ensue when these synergies are actually exploited (Cedergren & Petersen, 2011).In general, the number of multi-modal investigation bodies like the NTSB that explicitly include road transport (e.g.Finland, Sweden, Norway, Argentina, Taiwan) is growing, although there are also countries that exclude road transport from their multi-mode accident investigation bodies (e.g.Australia or Canada) or study it ad-hoc as a part of their nonspecific occurrences category (e.g.Netherlands).Other countries have permanent but distinct investigation agencies for air, rail and marine, in line with international agreements, but exclude road transport (e.g.UK or Germany) or bundle it with rail (e.g.France).

Independence
The prioritization of learning and future transport safety over criminal investigations, which aim to identify who is responsible for accidents, calls for a clear division of roles between accident investigation organizations, law enforcement authorities and other involved parties.Without full independence pressure from external bodies can affect recommendations.Likewise, it could make individuals and companies less likely to cooperate openly, and it can lead to tensions between stakeholders such as industrial regulators, industrial bodies and trade unions (Vuorio et al., 2014).It is the independence and in particular the resulting transparency (van Vollenhoven, 2002) that gives investigations their credibility and the subsequent recommendations the moral force to be adopted.Earlier calls to have accident investigation bodies write legal rules and regulations have been rejected on the grounds that it would make them legislative bodies instead of investigative bodies, and thus open to political manipulation and interests (Baxter, 1995).VAI should be impartial, objective, and structurally, functionally and organizationally independent of any public bodies as well as commercial parties (Roed- Larsen & Stoop, 2012), while having the proper investigative powers to conduct their work (access to information, places and people; obligation to have questions answered, etc.).
In reality, major accidents often see multiple investigations occur simultaneously led by various law enforcement agencies, public transport safety authorities, the insurance sector and the vehicle manufacturers themselves.Hence, sharing access, evidence and technical information is frequently set out in protocols and memorandums of understanding.In case of clear indications of criminality, law enforcement takes control (and carries the financial burden of evidence gathering), but otherwise accident investigations have precedence.In these latter cases the focus on safety learning generally justifies, for example, that witnesses cannot incriminate themselves.Naturally, this relationship between law enforcement and safety learning is constantly evolving, as are the investigations themselves.

In short
Ideal-typical VAI prioritizes safety learning over blame attribution.It benefits from a pragmatic delineated mandate that does not exclude events in advance and a perspective that places accidents within a wider sociotechnical system.The organizational structure is optimized when accident investigations are set up as a permanent organization that crosses transport modes.And investigations themselves are best served with an organization that has the proper mandates and investigative power to conduct investigations quickly and appropriately.

Challenges coming with connected and automated vehicles
This section discusses seven dimensions where CAVs potentially impact VAI: automation, mixed traffic, connectivity (and cyber security), data, regulatory context, safety culture and internationalization.

Automation
The introduction of advanced driver systems results in a redistribution of responsibility between machines and human drivers for all or some of the driving tasks, which leads to challenging handover moments between them (Alvarez, 2017) as well as potential lacking situational awareness for (remote) operators (Bogusławski et al., 2022).
Lessons might be gained from aviation and rail accident investigations (Read et al., 2022), which have experience with automation and the consequences of driver distraction at handover moments (with professionals, not private drivers).In the long-term, automation could further result in a reduction in driving skills as drivers gain an overreliance on automation (Banks et al., 2018;Strauch, 2018).The challenge for VAI will be to determine which of these is the root cause, especially given various levels of automation and producer variation of advanced driving systems with diverging human-machine interactions, distractions and software.
Investigators will need expertise and access to advanced driving systems and their back-end structures.

Mixed traffic
CAVs will partake in an open road system with mixed traffic (Alvarez, 2017;Larsson et al., 2010;Yuan et al., 2021).
As such, they will share road space with vehicles with other levels of automation (or none), pedestrians, bicycles, motorbikes, animals, and so on.These interactions between a vehicle and its environment are regulated by traffic rules, both formal and informal, and local customs.While CAVs can be programmed to follow formal rules, they have limited capacity to handle informal and customary aspects of traffic, communicate with other road-users beyond blinkers (F€ arber, 2016), or improvise and break with traffic rules if/when situations demand it.Studying these interactions is challenging, but fits the sociotechnical system trend within accident investigations that keeps expanding from a microlevel focus in order to include how 'constraints, recourses, demands and incentives are produced by regulatory, administrative and organizational factors, rather than the "errors" of individual humans in isolation (Cedergren & Petersen, 2011, p. 1243)'.

Connectivity
Cars are increasingly connected with the outside world and literally drive on millions of lines of code (El-Rewini et al., 2020).Such connectivity, especially with other road participants and road infrastructure, has the potential to enable various safety crucial applications such as cooperative collision warnings and avoidance, intersection control or cooperative cruise control (Elliott et al., 2019;Shladover, 2018), just as it enables other services like toll payment and traffic control (Guanetti et al., 2018).On the other hand, connectivity carries the risks of technical disruptions, cyberattacks and nonrandom interference (Liu et al., 2020;Morris et al., 2020).El-Rewini et al. ( 2020) describe how such malicious activities can be aimed at the sensors, the actual connectivity or the control(s) of the connected car.In terms of VAI, Shladover (2018) argues that connectivity will have important consequences for staffing at transport agencies, including VAI, as connected vehicles require skillsets from the information and telecommunication sector that so far have not been required in vehicle transport.Importantly, not all incidents will be nefarious.Furthermore, aviation and rail frequently deal with incidents wherein communication problems between vehicle-to-vehicle or vehicle-to-infrastructure play a role (like frequency interference in airplanes or signals passed at danger for rail).

Data
All sensors in CAVs collect data, which is stored either in the vehicle or off-site.This is another challenge for VAI in terms of analytic skill-sets and access.VAI and road safety experts face three difficulties: (1) most data is encrypted and stored in proprietary formats and therefore not accessible to parties outside the company, ( 2) not all effective safety data are always stored, or they are stored in a distributed way across supplier modules hindering recovery, and (3) the number of SAE Level 4 CAVs or higher on the roads is too small to provide enough data for road safety analyses (DSB, 2019; Grabbe et al., 2020).There is hence a discussion between car producers and third-parties (authorities, insurers, repair/aftermarket, consumer and privacy organizations, etc.) about access to in-vehicle data.This includes VAI, where a key challenge entails how much and what data should be recorded and submitted to investigations (Shladover, 2018).Debates are ongoing around event data recorders (EU, 2022) and the inclusion of ADAS (and future ADS subsystems) data in national vehicle registration processes so that they can be included in accident registration and analysis (DSB, 2019).There is further concern about those cases when vehicles with automation for whatever reason do not register something as an incident and hence do not engage emergency data capture protocols.Avoided accidents (breaking or disengaging) are not registered beyond the car company either.There is thus little public data about near-misses involving ADAS and ADS, except when vehicle companies are obligated to report disengagements, as required by the California Department of Motor Vehicles (Favar o et al., 2018).In other words, vehicle accident investigators will not only need powers of investigation to access vehicle and off-site stored CAV sensor and algorithm data, but the mandate to study ignored and near-miss incidents.Naturally they also require the ability, expertise and protocols to process said data, properly, purposefully and confidentially.

Regulatory context
Another aspect are the organizational and structural factors of CAV regulation.Within accident investigations these are increasingly acknowledged as causaly related to accidents but so far have received little attention (Cedergren & Petersen, 2011).Alvarez (2017) for instance highlights the feedback loops between regulators, companies and the design of automated vehicles and how their various responsibilities justifies inattention to factors not directly under their purview, thus generating unsafe conditions (NTSB, 2019).A related challenge is the need for regulation to keep a balance between safety and innovations (Hansson, 2020;Shladover & Nowakowski, 2019) as overly strict political rules or 'excessive' safety costs placed on manufacturers could undermine the developments of CAVs (Awad et al., 2018) while a lack of safety endangers lives and undermines consumer trust.We cannot answer this balance, except to reaffirm that this tension confirms the importance of political and commercial independence of accident investigative institutions even as they maintain a working relationship with law enforcement and vehicle manufacturers (van Vollenhoven, 2002).

Safety culture
The above is especially important when taking into account CAV producers' safety cultures.Every organization has a safety culture, including vehicle manufacturers.When it comes to vehicle automation and CAVs, however, in addition to traditional vehicle manufacturers, new entries into the market seem to have a strong background in software development and ICT, such as Google, Uber and so on (Lin et al., 2018;Merat et al., 2014).These companies have a relative short vehicle production history, are characterized by fast growth and heavy competition, and originate from a sector with a safety culture based on the idea of software updates.Whether this shift in mentality and practice from exhaustive pre-market testing (with recalls) to repeated updates and over-the-air add-on packages (like a one month sport-mode package) will impact the respective safety cultures for better or worse is something future investigations will show, but first indications are cause for concern (DSB, 2019;NTSB, 2019).

Internationalization
Compared to traditional road vehicles, CAVs have an even further increased international element.This follows, among others, from cross-border software updates, national driving cultures, locally differing traffic regulations and from new business models with centralized ownership in case of shared CAVs.However, while there is some international cooperation among multi-mode accident investigating bureaus (ITSA, 2022), this remains mostly ad-hoc and voluntary in case of road transport.This differs from air where investigation branches at the point of accident report their findings to the countries where the airplane was registered, operated, designed and produced.Findings are also shared with the International Civil Aviation Organization (ICAO), which maintains an international Accident/Incident Data Reporting database (ADREP).For vehicles, international data sharing so far is voluntary, like IRTAD at the Organization for Economic Co-operation (OECD) or IGLAD, which is supported by the European car manufacturers (Winkle, 2016).Policy makers might want to consider more structural arrangements when thinking about accident investigations of CAVs, with their international management and software updates.

In short
Together, these seven interrelated factors confirm that CAVs do have the potential to affect VAI.While CAVs do not affect the purpose and overall methodology of VAI, they do affect particular methods, institutional design and independence through requirements placed on staffing, expertise and internationalization.The above dimensions thus reinforce the need for systemic accident investigations and multi-and interdisciplinarity with a (new) focus on ADAS and ADS software, artificial intelligence, cyber-security and behavioral studies for mixed traffic and connectivity.The section further stresses the need for supportive domestic and international regulatory and legislative frameworks so that VAI can access required in-vehicle data and cooperate across borders.

Evolving vehicle accident investigations
Knowing the core ideal-typical VAI characteristics and the potential challenges from CAVs leads to the question whether current VAI agencies are able and capable of taking on CAVs.In this section, VAI organizations from the US, the UK and Japan are compared.

The United Kingdom
The UK has three separate permanent Air, Rail and Marine Accident Investigation Branches (AAIB, RAIB, MAIB) in line with international agreements and pre-Brexit European Union directives.It does not yet have such an agency for road vehicles (at the time of writing). 2Instead road collision investigations in the UK are shared between local authorities (police and coroners) and project-based programs.
The police is the leading agency for VAI and have the primary responsibility for the site and investigations (APP, 2020;NPIA, 2007).The road policing lead investigators play a key role in the management of the initial response and start of the investigation, under the guidance of central guidelines and processes (APP, 2020;NPCC et al., 2019;NPIA, 2007).The investigators process and analyze the data and create (digital) reconstructions.In case of fatal or life changing accidents the police also generate 'police fatal collision reports' to be used for criminal prosecution and the coroner's office.Police road collision investigations (and recommendations) are generally quite localized, with only adhoc cooperation across districts (Barry, 2017).Furthermore, while coroners are allowed to make recommendations around accidents with fatalities, few do so systematically (Gooding, 2017).
The UK also has project-based VAI programs.RAIDS (Road Accident In-Depth Studies) is funded by the Department of Transport (DfT) and contracted to the Transport Research Laboratory and the Transport Safety Research Center at Loughborough University since 2012.RAIDS coordinates specialist collision investigators and injury causation experts to investigate and gather event information.Although more specialized, RAIDS does not have official investigative powers and is limited to three specific UK regions.In fact, road accident data and analysis is spread out over a variety of organizations in the UK: from the police with STATS19 and fatal investigation reports to hospital data, Health and Safety reports, RAIDS, Highways England, local authorities, universities, NGO's like the RAC Foundation, and private companies.
Within the UK this lack of a nationwide systemic overview as well as evolving vehicle technology (incl.CAVs) have led to increasing calls for a permanent Road Collision Investigation Branch (RCIB) (Cuerden, 2017;Gooding, 2017;PACTS, 2017).The arguments in favor of such an organization are that it could develop a systemic national evidence base across a wide range of sources, support best-practice police investigations, undertake specialist investigations (based on severity and/or safety learning potential) and recommend national road-safety policies (Cuerden, 2017).Gooding (2017) similarly calls for attention to the missing link in road accident investigations within the UK, especially root cause analysis and shifting patterns of road safety accidents over time, but highlights how ideas for a RCIB are rejected based on costs and the already extensive powers of the police and coroners in the UK.While the debate is ongoing on whether and how such a RCIB should be set-up, the DfT has supported a study by the RAC Foundation into the 'business case' of investing more resources and the potential benefits of more systematic VAI (Barrow, 2019) and recently published the results of an open consultation on this topic (DfT, 2021).
The RCIB would reflect the UK's AAIB (1915), RAIB (2005) and MAIB (1989), which are all permanent accident investigation branches.They are part of the Department of Transport and coordinated by the Accident Investigation Chiefs' Council (AICC).This means that chief inspectors are civil servants appointed by the Secretary of State for Transport and budgets are provided based on the number of accident investigations a year.Being part of DfT, as Gooding (2017) notes, does not affect the political and corporate independence of these boards, as their independence derives from their statutes, role descriptions and procedures instead of their organizational structure (as in the case of the NTSB below).In terms of expertise, the branches work with a core staff and accredited agents-in the case of RAIB this includes screened staff from railway companies-to perform preliminary investigations and secure evidence at remote locations until RAIB arrives or is able to decide that a full investigation is not warranted.And in terms of investigative powers, AIB inspectors in the UK can enter property, seize evidence, require records and information, and enforce people to answer their questions.As elsewhere, the RAIB and AAIB are required to share a draft of their report for consultation with affected parties to comment on while also sharing technical and factual information with other (law enforcement) investigations (AIB & NPCC, 2018).

Japan
The Japan Transport Safety Board (JTSB) is the main multimode accident investigative body.The JTSB resulted from a merger between the Aircraft and Railway Accidents Investigation Commission and the Marine Accident Inquiry Agency in 2008 after a major rail accident.The JTSB has expert investigators in a variety of disciplines, with full-time and parttime investigators often having commercial experience or also working at universities.As in the UK, the JTSB falls under the Ministry of Land, Infrastructure, Transport and Tourism (MLIT), and MLIT appoints the board members and allocate its budgets (Fukuyama, 2009).The JTSB manages the rest of its staff independently from MLIT and is mandated to give direct recommendations to private parties.The JTSB consults and cooperates with the police, but priorities between them are decided on case-by-case bases.If JTSB is prioritized, the police are obliged to share their data with JTSB-and vice versa.
Similarly to the UK, however, the JTSB does not study road accidents.These are carried out by the police, the Institute for Traffic Accident Research and Data Analysis (ITARDA) and the Accident Investigation Committee for Business Vehicles (AICBV).The objectives of the police are similar to those in the UK and the US: to secure the site of the incident, gather basic evidence, identify legal responsibility and criminalize if necessary (AlFS, 2020).Within the police organization, the responsibility for VAI lies with the department of traffic (e.g.traffic police) at each prefectural police headquarter or district police office (Okatamo, 2012).Police investigations target serious and malicious accidents (i.e.drink-driving) and accidents arising from complicated causes (NPA, 2020).The analyses fall under the responsibility of the Director for Traffic Accidents.At a national level, the Traffic Bureau of the National Police Agency (NPA) oversees the training of traffic officers, and establishes procedures and processes.
The two other organizations, ITARDA and AICBV, conduct scientific analyses of the factors leading up to accidents and offer proposals to prevent reoccurrences.ITARDA was created in 1992 following a rise in accidents and a growing safety awareness in the 1980s and 1990s.The accident investigation institute employs academic staff covering a variety of disciplines, ranging from brake systems to cognitive functionality (ITARDA, 2020; Kinoshita, 2019;Saito, 2019).
ITARDA was set up as a public-interest corporation with approval from Prime Minister and relevant ministers, and officially designated as an investigation center by the National Public Safety Commission.Its board members originate mainly from relevant industry associations.In 2020, more than 70% of ITARDA'S budgets came from the results of their research, with direct subsidies making up less than 3%.ITARDA's research is not shared with the police, insurance companies or court investigations.Its reports include a variety of topics and recommendations (e.g.measures to elderly drivers or cooperation between accident analysis and medical services).
In turn, the AICBV studies severe accidents outsourced from the Ministry of Land, Infrastructure, Transport and Tourism (MLIT, 2020) and the NPA involving commercial vehicles such as (tourist) buses and heavy vehicles.It was established around 2010 in response to several commercial vehicles crashes that were found to be mediated by the corporate culture and organizational structure of their operators.The AICBV consists mainly of academic experts from universities, with expertise ranging beyond traffic and engineering to other fields such as labor, law, construction and health.Their investigation outcomes involve recommendations to operators on issues like the health of drivers, proper instructions for drivers and a reduction of burdens on drivers.

The United States
In the United States, public road accident investigations are shared across local, state and federal police, the National Highway Traffic Safety Agency (NHTSA) and the National Transportation Safety Board (NTSB).
As elsewhere, local and state police are often first on the scene and thus involved in initial response.Besides ensuring safety, the key accident investigation objective of law enforcement is to determine fault and liability and to provide evidence for insurance claims or injury lawsuits.On a national level, the NHTSA is involved with accident investigations as an agency of the Department of Transport offering vehicle safety ratings, safety standards and vehicle recalls (NHTSA, 2016).It does so through various programs, most of which are focused on traffic safety, not accident investigations.However, the NHTSA is responsible for emergency communications (the 911 number) and runs the Crash Injury Research and Engineering Network (CIREN), which conducts detailed engineering and medical examinations on crash vehicles and the injuries that victims sustain to prevent and improve (emergency) treatment of injuries.Noted challenges include regulating rapidly changing technologies (CAVs, cyber security), updating the vehicle recall system (Malone & Creamer, 2016), and political interference-the NHTSA has been without administrator since 2017 (Mulero, 2019).
The NTSB is a permanent, multi-mode and multi-disciplinary accident investigation board set up in the late 1960s which gained independence from the Department of Transportation under the 1974 Independent Safety Board Act (Baxter, 1995).With 5 board members and 400 plus staff, the NTSB covers accidents in air, rail, highway, pipeline, and other accidents that the Board deems salient.Within the NTSB, specialists with a wide range of expertise are assigned on a rotational basis to so-called Go Teams.Depending on the severity of the accident, these teams head to the scene and lead the investigations under a Board's Investigator-in-Charge (IIC).Only during the on-site evidence gathering phase do Go Teams include expertise from contracted parties in government and industriesy (like the FBI, FAA or EPA).The NTSB is not organizationally affiliated with the Department of Transport but falls directly under the US Congress and is officially independent from local, state and federal judicial investigative agencies.Although the NTSB has priority over other simultaneous investigations, it does not have exclusive authority, cannot offer witness immunity, and is obliged to share evidence.During and after the investigations, the NTSB publishes recommendations aimed at the organization best able to address issues, including regulators, operators or users.It communicates and tracks these recommendations with its 'Office of Safety Recommendations and Communications', adjusting communication strategies to increase (public) support and pressure organizations to implement their recommendations.
Although a global leader, the NTSB highlights various challenges to its work and organization.These challenges include the selection of accidents to investigate given available resources, recruitment and training of staff, the development of more impactful recommendations and generally needing to "gain familiarity, increase technical knowledge and engage with industry leaders in the emerging technologies of [space, drones, rail], automated vehicles, database management systems, and data analytics (NTSB, 2017, pp.17-18)."

In short
In comparison a number of things stand out from the way VAI is organized across these three countries.Firstly, the three examples show that accident investigation structures have strong local flavors that are hard to adapt (cf.van Vollenhoven, 2002) and are based on historical decisions, path dependency and disaster responses (JTSB merger).The brief overviews also show, however, that VAI in these countries is not static but keeps evolving, implying that decisions are and keep getting made on institutional design and responsibilities.Secondly, the way independence is organized differs across the three countries and is either based on organizational independence (NTSB) or mandated and statutory independence (AIBs in UK and Japan).All three countries have initiated independence at an investigation level, but the relationships between VAI and law enforcement or transport companies are evolving constantly.Thirdly, the countries seem to have different staff constructions but nevertheless share an appointed board, hired staff and detached/accredited agents.CAV software and cyber expertise seems to rely more on the latter group, thus complicating and potentially conflicting with VAI independence.In short, there are clear indications that respective VAIs in these countries are affected and challenged by automation, let alone CAVs.

Conclusion
The three lines of inquiry above highlight that VAI is constantly evolving, and that CAVs will demand even more of it.The first line of inquiry summarized the core characteristics of ideal-typical VAI.It reinforced the need for VAI to have a pragmatic mandate and proper investigative powers; a permanent and independent multi-modal institution; an ever wider set of specialists expertise; and a clear focus on safety learning through a sociotechnical system perspective that crosses people, organizations, regulations and technology.The second line of inquiry touched on seven challenges that CAVs pose to existing VAI, ranging from the technical aspects of advanced driving systems (automation, connectivity, data) to mixed traffic and national and international organizational and regulatory contexts (regulatory contexts, safety culture, internationalization).And the third line of inquiry briefly described the current state of affairs in three countries, highlighting how VAI is witness to various institutional designs, levels of (in)dependence, staff recruitment approaches and geographical coverage-and thus a valid field of study for transport governance.
As stated, the aim of this paper was to explore some of the initial consequences of current developments around CAVs for VAI governance.We fully acknowledge that this paper is limited in empirical depth and range of countries, which is a core reason why we refrain from policy suggestions and generalizations about VAI design beyond the ideal-type summarized in Sec.2: a permanent, fully independent, multi-modal accident investigation organization with proper investigative powers aimed at finding the factors leading up to the accident and offering recommendations to prevent them.That said, CAVs do draw attention to VAI mandates, expertise and absent (inter)national bodies.This paper's contribution thus lies in confirming that there is a need to extend the discussion about the back-end learning and organization of vehicle accident investigations, not just the upfront safety of these vehicles.
In terms of future research, this paper justifies an indepth empirical study into the interaction between CAVs and VAI governance.Specifically, we envision: (1) a wider and more systematic empirical engagement with VAI governance in other countries; (2) more empirical attention to the politics, design, staffing and internationalization of (vehicle) accident investigation organizations and how independence is achieved under changing conditions; (3) an indepth study into the staffing and expertise challenges that VAIs experience because of new road mobility technologies (beyond a focus on data, protocols and processing capacity); and (4) an empirical comparative study into the investigative powers of various accident investigation branches and the consequences of their absence.These and similar studies would offer important academic insights and policy reflections regarding the governance of VAI, new mobility technologies, road safety and consumer adoption.Learning how vehicle accident investigations are organized, institutionalized, funded and conducted can and should be done actively, not just reactively.