Call for evidence outcome

Automated vehicles: statement of safety principles - summary of responses and next steps

Updated 17 June 2026

Introduction

The automated vehicles: statement of safety principles call for evidence was published on 10 June 2025 and ran until 1 September 2025. This is a summary of the responses we received to the call for evidence.

It covers:

• the background to the call for evidence

• a breakdown of respondents by stakeholder group

• the analysis methodology

• a summary of the responses

• next steps

Background

The Automated Vehicles Act 2024 enables the implementation of a new legal framework to enable the use of automated vehicles (AVs) on Great Britain’s roads.

Included within the act is a requirement for the Secretary of State for Transport to prepare a statement of safety principles (SoSP). This will allow government to establish core safety principles that all automated vehicles must adhere to so they can be used on Great Britain’s roads.

This call for evidence sought information to support our understanding of how:

• the safety principles might be used
• the safety standard might be described
• safety performance can be measured

Breakdown of respondents

Responses could be provided via an online survey which was accessible via the call for evidence webpage, via email or post.

The department received 80 responses to the call for evidence. 37 responses were received via the online survey and 43 responses were received via email or post.

Responses were received on behalf of 15 individuals and 65 organisations. A breakdown showing the type of organisation of those responses received on behalf of an organisation is set out in Table 1.

Table 1: Breakdown of respondents by stakeholder group

Type of organisation	Number of respondents	Percentage of organisation respondents
Academia	0	0%
Disability group	0	0%
Emergency services and police	2	3.08%
Highway authority	1	1.54%
Insurance	9	13.85%
Legal	5	7.69%
Local government	4	6.15%
Local representative group	0	0%
Public sector	2	3.08%
Research, consultancy and professional organisation	10	15.38%
Trade body	7	10.77%
Safety and road user group	11	16.92%
Vehicle operator	4	6.15%
Vehicle supply chain	0	0%
Vehicle technology developer and/or vehicle manufacturer	7	10.77%
Another area:	3	4.62%

Analysis methodology

The call for evidence survey comprised a total of 39 questions and included both closed and free text responses. All responses received were analysed by DfT officials. We are grateful to all the organisations and individuals who responded to the call for evidence with their views and suggestions.

A thematic analysis was undertaken against questions to draw out key themes raised through the responses alongside analysis of responses to closed questions. This document sets out the main themes raised through responses and the government’s response.

Not all participants answered all the questions. The data and analysis presented is based on the responses received to each question.

Where respondents did not respond directly to a yes/no question, these responses have not been categorised as agree/disagree/don’t know or yes/no/don’t know and therefore the statistics for each question will not represent those responses, although any comments provided as part of these responses will have been considered.

Uses for the safety principles

1) In your view, are there any other uses for the safety principles we have not identified?
2) In your view, what other uses might there be for the safety principles and why? Provide evidence if possible.

There were 55 responses to question 1. Of those who responded to question 1, 69% of respondents answered yes to the question of whether there were other uses of the SoSP that we had not identified. 16% disagreed and 15% didn’t know. There were 45 responses to question 2.

Respondents identified a range of other uses for the safety principles beyond those that we identified in the call for evidence.

Supporting international standards

Multiple respondents to this question suggested that the SoSP should act as a benchmark for international harmonisation, aligning UK standards with United Nations Economic Commission for Europe (UNECE) processes and other global frameworks. Many referenced how this would support cross-border deployment and reduce regulatory fragmentation, minimising duplicate evidence burdens for manufacturers and operators.

Several noted that this alignment should avoid duplication of technical compliance roles and avoid creating parallel regulatory processes, instead mirroring UNECE standards for data storage and reporting.

Interoperability testing and cybersecurity considerations were also flagged as essential for global alignment. In addition, aviation and rail were noted as good examples of industries to mirror in terms of aligning innovation with regulatory goals to set safety benchmarks.

Public education / trust

A key theme across several of the answers was the opportunity for the SoSP to build public understanding and confidence in automated vehicle deployment. The key message was that it should serve as a tool for public education and communication to provide adequate assurance that safety is set at an acceptable level.

One respondent highlighted how the safety principles could also be used to monitor misalignment between societal expectations and actual AV performance, informing improvements for manufacturers and operators to reassure the public.

Another key element raised by several respondents was the importance of data transparency in building public trust. Respondents recommended achieving this through standardised data collection and publishing, with a clear audit trail and consistent application across manufacturing and post-deployment.

One respondent suggested developing a UK wide incident and near miss data base structured around the SoSP to enable transparency. Another respondent noted the risk that early AV deployment may increase collisions and that this must be transparently acknowledged by the government.

Insurance

Many responses also noted how the SoSP could play a critical role in insurance and liability frameworks, by providing clear definitions of acceptable risk levels for automated vehicles. Many noted that the SoSP should be used to provide assurance to insurers and underwriters that vehicles meet safety requirements. One respondent highlighted direct lessons from electric vehicles, where higher premiums have partly been the result of a lack of data, which has damaged public trust - they stressed the need for early clarity to avoid similar issues. Another respondent was critical of the SoSP, outlining that they lack clarity in responsibility and liability definitions and measurable safety parameters.

End-to-end development

Many of the respondents suggested dynamic application of safety principles across pre and post-deployment stages. One respondent recommended extending the SoSP to inform the supply chain and third-party service providers with end-to-end quality assurance through establishing clear expectations and safety requirements.

Others focused on how the SoSP should guide designers and manufacturers during the pre-deployment phase, including vehicle safety systems and control technologies.

Others suggested extending the SoSP beyond approval checkpoints to cover lifecycle aspects, to ensure safety is continuously maintained,such as:

• software updates
• recalls
• re-authorisation after major changes

Using statement of safety principles at pre-deployment

3) Do you agree or disagree with our characterisation of how the statement of safety principles might be used at pre-deployment?
4) Why do you think this (providing evidence if possible)?

53 respondents answered question 3. Of those who responded to this question, 45% agreed with our characterisation of how the SoSP might be used at pre-deployment and 23% disagreed. 32% were neutral. 44 people responded to question 4.

While many respondents expressed agreement with the proposed approach, they often stressed that the principles must be supported by clear implementation guidance, robust validation methods and greater transparency about how compliance will be assessed.

Many of these respondents highlighted the value of the SoSP in shaping a consistent framework that places responsibility on manufacturers to demonstrate that automated vehicles meet defined safety expectations in real-world conditions. Several cited ongoing work from initiatives like the Sensor Assurance Framework as evidence that comprehensive pre-deployment safety validation requires a combination of virtual modelling, track testing and live road trials, especially in varying environmental and weather conditions.

There was particular emphasis on the need for testing fidelity, sensor performance assessment and traceability across test environments. Others stressed that the SoSP should not operate in isolation but should sit within a wider authorisation process that incorporates independent third-party validation and considers the operational design domain of the system being approved.

Some noted the importance of using the SoSP to set clear thresholds for safety that are meaningful both for regulators and the public, with a few respondents suggesting that the principles could also support public communication and trust-building ahead of widespread AV deployment.

Those who disagreed raised a range of objections, often focused on perceived gaps in accountability, methodological limitations, or regulatory design. Some respondents criticised the approach for relying too heavily on manufacturer-provided safety cases without sufficient independent scrutiny, comparing this unfavourably to the licensing of human drivers or the regulation of pharmaceuticals.

Others questioned the realism of test-based evidence when deployed vehicles must operate in uncontrolled, dynamic environments, raising doubts about whether simulated or structured testing can fully reflect real-world challenges.

There was concern that critical areas, such as cybersecurity, system reliability and the specific risks faced by vulnerable road users (VRUs) like motorcyclists, were not sufficiently integrated into the proposed approach.

Some respondents challenged the idea that traditional type-approval processes could simply be extended to AVs, warning that the absence of a driver fundamentally changes the nature of vehicle operation and introduces new requirements that may not be addressed under existing frameworks.

There was also criticism that safety and security considerations appeared to be treated as downstream compliance issues, rather than embedded from the earliest stages of design. A small number of respondents argued that the SoSP’s application should be subject to greater democratic scrutiny or questioned whether the proposed framework sufficiently addresses public concerns about control, liability and long-term safety oversight.

Informing pre-deployment safety requirements

5) Do you agree or disagree with our characterisation of how the  statement of safety principles might be used to inform pre-deployment safety requirements?
6) Why do you think this (providing evidence if possible)?

There were 54 responses to question 5. Of those who responded to this question, 41% agreed and 26% disagreed. 28% neither agreed nor disagreed, with a further 6% indicating that they didn’t know. There were 42 responses to question 6.

Many respondents agreed with the proposed use of the SoSP as a foundation for assessing safety prior to deployment. These respondents generally saw value in grounding safety assessments in a principled framework that offers consistency, regulatory clarity and a clear benchmark for developers.

Some highlighted the SoSP’s potential to reduce subjectivity in decision-making, encourage early alignment with regulatory expectations and support public confidence. Others noted the benefit of aligning UK requirements with international standards, particularly UNECE regulations, to avoid duplication and facilitate cross-border deployment.

Several submissions referenced evidence-based models as a way to give substance to the “careful and competent” standard and ensure measurable performance. One example was the ‘speed, space, surprise, consequence model’ which provides measurable criteria such as dynamic speed management and consequence-aware decision-making.

However, even among those broadly supportive, many stressed the importance of applying the principles contextually; for example, by tying assessments to the specific operational design domain in which the vehicle is intended to operate and ensuring that safety is not only demonstrated under ideal conditions but also in adverse environments such as poor weather or low-visibility situations.

There was also strong emphasis on the need for simulation, track testing and real-world validation to be used in combination, with some calling for a Euro New car assessment programme (NCAP) style rating system or periodic reassessment to ensure continued safety throughout the vehicle’s lifecycle.

Conversely, a significant minority of respondents disagreed with the characterisation of the SoSP’s role, raising concerns about conceptual clarity, feasibility and potential unintended consequences. A common criticism was that benchmarking against human drivers, particularly the “average” driver, risks cementing an unambitious or poorly defined standard.

Several argued that public expectations for AV safety are far higher than for human drivers, drawing comparisons with other transport modes like rail and aviation, where system-led safety is held to stricter scrutiny. Others noted that the “absence of unreasonable risk” standard could be too vague to operationalise effectively or may even conflict with statutory language in the AV Act that refers to “acceptably safe” and “acceptably low risk” operation.

Concerns were also raised about the lack of a practical, evidence-backed methodology for implementing human driver comparisons at scale, particularly given the volume of data and testing this would require. Some respondents proposed that pre-deployment assessments should instead focus on whether good design, testing and safety engineering practices have been applied, with comparative benchmarking used retrospectively as more in-service data becomes available.

Additional themes included the need to address lifecycle risks such as component wear and software degradation and the importance of reflecting the broader “Safe System” approach, in which AV safety is seen as a shared responsibility involving operators, manufacturers, infrastructure providers and regulators. There were also calls to explicitly consider the needs of vulnerable road users and ensure the SoSP reflects the full complexity of real-world driving environments.

Information at pre-deployment

7) What information do you think would need to be provided to the relevant authority pre-deployment to demonstrate consistency with the statement of safety principles?

There were 61 responses to this question. The majority of respondents highlighted 2 key types of information that should be provided to the relevant authority:

• safety case
• testing results

Safety case

Respondents argued that a detailed safety case evidencing the safety of the autonomous system should be provided. Several respondents highlighted specific data points they would expect to see outlined within the safety case, such as:

• information on how the vehicle has been validated against human driving benchmarks
• a profile of expected risks and mitigations
• the organisation’s safety management system
• how the AV will interact with other road users, including vulnerable road users

The majority of respondents also highlighted that information on the vehicle’s expected operational design domain (ODD) and evidence that the system is able to safely operate in these conditions, should be provided either as part of the safety case or as additional documentation.

Testing results

Second, the majority of respondents argued that results from any testing done on the automated system should be provided. Results from real-world testing were seen by many respondents as the most valuable here. However, several acknowledged that it would not always be possible to conduct real-world testing on a large enough scale to be able to solely rely on this.

The vast majority of respondents therefore also highlighted virtual and closed-track testing as important data sources here. Some respondents recommended that these different types of testing should form a multi-stage approach, with virtual testing taking place first, following by closed-track and then real-world tests.

Many respondents also highlighted the sort of data they would expect to see validated through testing, such as:

• incident rates
• the system’s handling of routine and edge-case scenarios
• interactions with vulnerable road users
• near-miss and critical incident rates

Other comments

Several respondents called for close alignment with the approach to pre-deployment assessment taken in the upcoming UNECE automated driving system (ADS) regulations. These respondents argued that the requirements in the UNECE regulations should serve as the basis for pre-deployment safety and that any pre-deployment safety requirements introduced should align with these.

They also called on government to avoid introducing any new or different pre-deployment requirements and to avoid unnecessary duplication of the UNECE requirements. They argued that doing so would prevent potential overregulation and inefficiency and would ensure the UK remains aligned with international approaches to AV regulation.

Meeting safety expectations at pre-deployment

8) In your view what considerations should be taken into account when assessing at pre-deployment whether automated vehicles meet the expectations set by the statement of safety principles?

There were 62 responses to this question. Respondents highlighted a range considerations that should be taken into account. Many argued that any pre-deployment assessment needs to include an objective, technical assessment of the ADS’ capabilities and highlighted a range of factors that this technical assessment could consider.

These included the capacity of the ADS to handle common and rare scenarios, its ability to adapt to new risks and edge cases, how it handles interactions with other road users and in particular vulnerable road users and how the ADS’ behaviour compares with human driver performance benchmarks.

Many respondents also argued that considerations around the vehicle’s operational design domain (ODD) need to be taken into account. This included consideration of whether the ODD’s parameters had been specified in proper detail, whether the ADS’ capability had been tested within these specific ODD parameters and whether assessment of the ADS’ interactions with existing infrastructure within the ODD had taken place.

Several respondents highlighted that any framework for assessing compliance will need to be dynamic and evolve, as learnings from real-world AV deployments emerge, technologies and infrastructure develop and new risks are identified. Several respondents also called for alignment with the upcoming UNECE ADS regulations.

These respondents argued for using the safety case and type approval as the basis for assessing pre-deployment safety and highlighted that alignment with the UNECE regulations will ensure that compliance with the SoSP will be well documented within the safety case. They also argued that UNECE alignment will maximise the interoperability and safety of systems across countries and will also help to avoid inefficiencies and regulatory burden.

Two other considerations received multiple mentions. First, several respondents highlighted considerations about data. These included arguments that:

• an assessment will be needed of whether the authorised self-driving entity (ASDE) or no-user-in-charge operator (NUICO) will be able to provide sufficient data to enable post-deployment monitoring and investigations
• incident data will need to be made readily available to insurers to enable fair claims handling

Second, several respondents highlighted considerations around vehicle operators and manufacturers, arguing that assessments need to consider the reputation and financial standing of these organisations and whether they have embedded robust operational processes and strong safety cultures.

Using the statement of safety principles at post-deployment

9) Do you agree or disagree with our characterisation of how the statement of safety principles might be used at post-deployment?
10) Why do you think this (providing evidence if possible)?

There were 60 responses to question 9. Of those who responded to this question, 57% agreed and 13% disagreed. 28% neither agreed nor disagreed and 2% indicated that they did not know.

There were 40 responses to question 10.

The vast majority of respondents highlighted additional considerations and questions around the way in which AV performance will be reported post-deployment and how this will be used. Several respondents argued that both leading and lagging metrics should be used to assess AV performance, but some added caveats about how these should be used.

For example, some argued that comparable metrics for human driver performance, such as STATS19, should be used to benchmark AV performance against, whereas others highlighted potential challenges in comparing AV and human driver performance, such as the lack of ODD-specificity in human driver metrics and therefore felt that lagging metrics were more important.

Some respondents argued that AV performance reporting should include information about how the vehicle has detected and responded to new or intensified risks over time, including changes in ODD or environmental conditions and information about how AVs have interacted with other road users, in particular VRUs.

Some respondents also raised questions about how issues identified post-deployment will then affect vehicle approvals and authorisations and called for clearly defined processes for the alteration of approvals and authorisations in response to post-deployment assessments or incident investigations.

Many respondents raised specific concerns about the reporting of incidents. Some respondents highlighted near-misses, arguing that the reporting of near-misses should be made mandatory as these could provide useful warning signs of emerging risks before collisions occur and because they felt that AV systems might not fully capture near misses.

Several respondents argued for broader mandatory reporting of all incidents (including, for example, collisions, near misses, system disengagements, non-injury or damage collisions, harassment by vehicle passengers). These respondents argued that only mandatory reporting of all incidents would allow government to gain a full picture of AV performance.

They often felt that minor incidents in particular would otherwise go under-reported, either because the ADS may fail to capture them or because companies may deliberately not report them and that mandatory reporting of all incidents will be important in allowing trends to and potential future safety issues to be identified. Some respondents also raised questions about what counts as a reportable incident, with some specifically calling for government to define what counts as a near miss.

There were 2 other common points raised by respondents. First, many respondents expressed support for the establishment of statutory inspectors and no-blame incident investigations.These respondents typically also emphasised the importance of establishing a culture of learning and the sharing of safety learnings between manufacturers, operators and ADS developers to enable continuous improvements to safety.

Second, several respondents called for the post-deployment framework to be aligned with UNECE ADS regulations and argued the UK should avoid imposing any new or different post-deployment requirements to those set out in the UNECE regulations.

Providing information at post-deployment

11) What information do you think would need to be provided to the authorities post-deployment to demonstrate consistency with the statement of safety principles?

There were 61 responses to this question. Respondents highlighted a range of information they thought should be provided to the authorities post-deployment to demonstrate consistency with the SoSP.

The vast majority of respondents identified a range of metrics on automated vehicle performance within their response. The metrics identified included:

• data on vehicle miles travelled
• any incidents involving the vehicle
• the vehicle’s compliance with road rules
• telematics data
• data on how the vehicle has interacted with other road users, in particular VRUs
• information on any minimal risk manoeuvres
• data on the vehicle’s performance in its defined ODD
• any unexpected environmental conditions
• both leading and lagging metrics

Some respondents argued that the data provided should be compared against human driver metrics in order to assess AV performance. Others suggested that the data provided should allow for the assessment of both the performance of individual vehicles and the identification of trends across vehicle fleets.

Many respondents focused specifically on the information about incidents they felt should be provided. The types of information identified included data on near misses, system disengagement, killed or seriously injured (KSI) incidents, collisions and minor non-injury and non-damage incidents.

Of these, information on near misses was most commonly identified, with the majority of respondents who discussed incident reporting requirements arguing that this should be provided to the authorities. Some respondents argued that data on the conditions both inside and outside the vehicle during an incident will be required to be able to fully reconstruct the external environment and decision-making processes that led to the incident.

Some argued that mandatory reporting of all incidents involving AVs should be required, especially in the earlier years of deployments, to effectively ensure safety. Some also stated that information on what the regulated body and/or vehicle manufacturer are doing to address any issues identified during investigations to prevent incidents reoccurring should be required.

Several respondents argued that information on any changes made to vehicles which may affect the vehicle’s ability to meet pre-deployment requirements should be required. This included information on any software updates and whether these have been installed properly and any hardware modifications.

Several respondents also highlighted human-reported information they felt should be provided. This included any feedback or complaints received from the public, other road users and owners/users of automated vehicles. Some specifically felt that it would not always be possible to get a full picture of an incident through AV-captured data alone and so human reporting would be required to fill in any gaps in knowledge on these incidents.

Several respondents argued that any post-deployment reporting requirements should be aligned with those set in the forthcoming UNECE ADS regulations. These respondents felt that the in-service monitoring and reporting (ISMR) and data storage system for automated driving (DSSAD) requirements within the UNECE regulations provide sufficient post-deployment oversight and so the UK should avoid adding any new or different post-deployment requirements to ensure international harmonisation.

Several respondents also argued that the data provided must be easily accessible and in a standardised format across all manufacturers to allow for effective incident investigations. Some of these respondents stated that the data provided should be stored in a secure central location to allow for easy access by regulators and incident investigators. Some of these respondents also called for the mandating of independent verification of the data provided by regulated bodies to ensure accuracy and accountability.

Meeting safety expectations at post-deployment

12) In your view what considerations should be taken into account when assessing at post-deployment whether automated vehicles meet the expectations set by the statement of safety principles?

There were 62 responses to this question. Respondents put forward a range of considerations which they felt should be taken into account.

Many respondents raised concerns about how performance metrics are used after deployment and said that these need careful consideration.

For example, several people stressed that performance should be assessed in the context of the environmental conditions the vehicle is operating in, rather than in isolation.

This is in order to properly contextualise vehicle performance, ensure that performance is being measured across a variety of environmental conditions, assess whether the vehicle has remained within its ODD and whether ODD conditions have remained consistent and assess whether the vehicle has been able to safely handle any changes in environmental conditions.

Similarly, several respondents argued particular consideration must be given to the ways in which AVs are interacting with other road users and any feedback received from AV users and other road users. Some respondents felt that this was important for capturing additional feedback about the real-world performance of AVs and identifying any emerging risks from human-AV interactions, such as over-trust of AV systems and misunderstanding of AV interfaces.

Others argued that this would allow for the identification of any impacts AVs are having on other road users and their behaviour, such as if particular groups of road users are modifying their travel behaviours because of a lack of trust in AVs.

Some respondents argued that considerations around the organisational safety of regulated bodies should be taken into account. This includes:

• consideration of how the regulated body is responding to its regulatory requirements, incident investigations and public/regulator/stakeholder feedback
• how well the regulated body is maintaining its safety management systems
• whether the regulated body is actively implementing a learning culture and proactively implementing and sharing learnings following incidents

Some also argued that consideration must be given to:

• whether vehicles are being kept updated with the latest software
• whether failure to do so is causing incidents
• how regulated bodies are dealing with situations in which owners are failing to install safety-critical updates or where older vehicles are no longer eligible for software updates

Finally, several respondents again argued that any post-deployment requirements introduced should be aligned to those contained in the draft UNECE ADS regulations.

 Performance of human drivers

13) Provide any evidence you are aware of on the current performance of human drivers?

There were 55 responses to this question.

When answering the question respondents regularly noted that they considered the performance of human drivers to be variable. Several factors were identified in this including:

• age

• driving experience

• impairment

• attentiveness and compliance with road rules

• road design

• vehicle capabilities

Eight respondents identified that human error is a factor in collisions.

Road user representatives also provided evidence on the impacts poor driving performance has on groups of vulnerable road users.

In response to the question, we also received suggestions on a range of specific datasets to explore further. These included:

• government statistics e.g. Department for Transport road casualty statistics, Home Office statistics on fixed penalty notices issued to motorists and sentencing council motor offence data tables

• datasets collected by Highways authorities and local transport authorities

• datasets collected by groups representing road users

Some respondents also identified specific pieces of research for consideration, as well as expert organisations and individuals. These included:

• Japan Automobile Research Institute

• naturalistic driving studies UDRIVE and SHRP2

• IAM Roadsmart

Other respondents identified particular academics with relevant policy expertise:

• Oliver Carsten, University of Leeds – road user behaviour, vehicle design, speed management

• Nicola Christie, UCL – transport safety and injury prevention ·

• Lisa Dorn, Cranfield University – driver behaviour

• Dr Sarah Jane Fox (University of Leicester, Co-Director Institute for Digital Culture) - Road and Aviation expert (law, policy strategy - including autonomous/technology enhanced systems)

• Jonathan Flower, UWE Bristol – autonomous vehicles, street design, road safety

• Siddartha Khastgir, University of Warwick – safe autonomy

• Sally Kyd, University of Leicester – road traffic offences

• Phillip Morgan, Cardiff University – human factors, human-machine interface, transport and intelligent mobility

• Damian Poulter, University of Greenwich – human behaviour in road traffic environments

• Nick Reed, Reed Mobility – psychology, human factors, road safety, connected and autonomous vehicles

• Helen Wells, Keele University – connected and automated vehicles, policing

• Mark Young, University of Southampton – human factors in transport

Human driving competence

14) In your view, does human driving performance improve with competence?
15) Why do you think this (providing evidence if possible)?

There were 60 responses to question 14. Of those who responded to this question, 65% agreed that human driving performance improved with competence. 15% disagreed and 20% didn’t know. There were 53 responses to question 15.

There was general agreement that driving performance increases with competence, with the majority of respondents to question 14 agreeing with this statement.

34 respondents provided further information on this. They noted that performance generally improves as competence develops, in particular for hazard perception, decision-making under pressure and implementation of strategies to manage distractions or impulses.

There was also a belief that increased competence, or experience, would translate to a driver having greater ability to navigate or avoid difficult situations. However, some respondents also noted that greater levels of competence can lead to over-confidence and an increase in risky manoeuvres.

These responses identified the fact that a driver’s competence is variable and that there are other factors that can affect performance. Some respondents noted the principles adopted in the safe system approach that human error is inevitable and so competency will not directly relate to performance. Examples of these factors were given such as:

• illness

• hunger

• stress

• tiredness

• attitude

• temperament

Finally, some respondents identified evidence gathered by the department that demonstrate that younger and less experienced drivers are involved in more collisions. For examples, noting that 20% of collisions involve a driver aged between 17 and 24 years old. They also noted that men under 24 years old are 4 times more likely to be killed or seriously injured in collisions compared to drivers aged 25 or older.

Respondents also noted that driver experience and age generally decrease the risk of collisions. For example, DfT’s Cohort 2 study found evidence that safety significantly improved in the months after a new driver passes their test. However, it was noted that this trend reserves itself in older age when aging can lead to longer reaction times and other changes which decrease driving performance.

Characteristics of careful and competent human driving

16) In your view what characterises careful and competent human driving and why (providing evidence if possible)?

There were 63 responses to this question.

Respondents highlighted a range of characteristics they viewed as comprising careful and competent human driving. Many highlighted specific driving behaviours they’d expect a careful and competent driver to exhibit, such as good vehicle control, predictable driving, smooth driving which avoids harsh braking and cornering and not driving impaired or under the influence.

Several respondents listed driving behaviours specifically relating to hazard perception, such as successfully perceiving hazards and reacting to them, being able to anticipate hazards and take action to avoid them, being aware of your surroundings and making context-specific evaluations of risks and hazards.

A number of respondents highlighted obeying all road traffic laws and safety-critical road signs as a key characteristic here. For some, fully obeying the Highway Code was seen as the standard that should be achieved by a fully careful and competent driver, or at the very least could be considered a good starting point for evaluating careful and competent driving.

 A number of respondents also highlighted being able to successfully interact with other road users as a key characteristic. This included being aware of other road users and their actions, being able to communicate effectively with other road users, including:

• understanding hand signals
• being able to anticipate the actions of other road users
• paying particular attention and care to vulnerable road users

Some respondents highlighted the ability to account for and drive safely in, all environmental conditions as a key characteristic. This includes being able to account for prevailing vehicle, environmental and weather conditions and being able to adapt your driving style to changes in these conditions.

Some respondents also listed a range of non-driving tasks they thought careful and competent human drivers successfully undertake. These included ensuring that all passengers are wearing seatbelts, ensuring the vehicle’s lights are on as necessary, being able to judge their personal fitness to drive, being able to identify and assess vehicle defects and careful route planning.

A small number of respondents argued that it is difficult, or not possible, to identify specific characteristics because the careful and competent standard is hard to define. These respondents argued that ‘careful and competent’ is not an objective measure, that it will mean different things to different people and that it can be applied differently depending on context.

A small number of respondents within this group highlighted that case law does not show any absolute standard for dangerous or careless driving and so by extension does not provide any absolute standard for careful and competent driving either.

Careful and competent automated driving

17) Do you agree or disagree with the considerations we have outlined in thinking about careful and competent automated driving?
18) Which consideration do you disagree with and why (providing evidence if possible)?

There were 56 responses to question 17. Of those who responded to this question, 27% agreed with all the considerations identified; a further 46% agreed with the majority of considerations. 11% disagreed with the majority and 2% with all considerations. 13% responded that they didn’t know. There were 41 responses to question 18.

Nearly half of respondents agreed with the majority of considerations. Of those who agreed with the majority of considerations, several respondents argued they didn’t agree will all of the considerations because of the limitations of AV systems. The limitations highlighted included the fact that:

• machines can also have lapses in attention due to computational constraints
• AVs may struggle to handle unexpected scenarios and may perform below safety standards in some scenarios
• not all AVs will be able to react to objects beyond human eyesight – in particular camera-based sensor systems, which can be badly equipped for poor weather conditions

Many of those who agreed with the majority of considerations also argued that it is difficult, or in some cases wrong, to directly compare human and automated driving performance.

The reasons given for this included the fact that:

• careful and competent human drivers also have limitations and so shouldn’t be set as the standard for AVs
• it is difficult to directly compare the two due to the lack of a standard definition for careful and competent driving
• the unique characteristics of AVs mean they shouldn’t be expected to replicate human driving
• assessments of AV safety performance should be outcomes-focused, rather than based on comparisons

A small number of these respondents also argued that, due to the above factors, it would be more valuable to assess AV performance based on the safety case submitted under the UNECE ADS regulations.

A smaller number of respondents who agreed with the majority of considerations argued that a key additional consideration should be how AVs interact with other road users and ensuring the introduction of AVs does not negatively impact the behaviours of other road users or whole system safety.

Amongst those who disagreed with the considerations, the reasons given for this were similar to those given by respondents who agreed with most/all of the considerations. Several of these respondents also highlighted the limitations of AVs as their reason for disagreeing with the considerations.

Limitations given by these respondents included the fact that:

• AVs might demonstrate less variation in behaviour than human drivers and so might be consistently wrong in certain scenarios
• the abilities and behaviours of AVs will be constrained by biases in training data and will be dependent on how perception and decision-making are engineered
• that AVs may not be able to recognise when their understanding of a given scenario is insufficient

 A smaller number of respondents who disagreed with the considerations also argued that it was difficult or wrong to make a comparison between human and automated driving. The arguments given for this included that assessing careful and competent AV driving should rely on oversight of how systems perceive, interpret and act; the fact that careful and competent is not widely defined; and that it would be more valuable to assess performance using the safety case approach set out in the UNECE ADS regulations.

Comparing human and automated careful and competent driving

19) In your view, how might the assessment of careful and competent driving differ between human drivers and automated vehicles?

There were 57 responses to this question.

Many respondents focused their response on the differences in the way performance is assessed between human and automated drivers. Several of these respondents highlighted that assessments of AV performance are quantitative and focused on technical ability, whereas assessments of human performance are more qualitative and focused on the judgement, adaptability and proficiency of the driver. Some noted that AVs can also be tested to a much greater degree than human drivers as AVs can be tested in simulation and are not limited by fatigue or distraction.

As a result of these differences in assessment, some respondents argued that assessments of careful and competent driving should focus on measurable outcomes, rather than on direct comparisons between human and automated drivers. Several other respondents listed specific differences in the way AVs are assessed they think should be made to account for this.

For example, they argued that AVs should be judged on system reliability and safe performance within the ODD, that AV assessments must consider operational behaviour and system-level factors, that AVs should be assessed on the reliability and transparency of their decision-making algorithms and that AV-specific issues, such as maintenance and software errors, must be considered.

Several respondents argued that AVs should be held to a higher standard of careful and competent driving than humans. Some respondents in this group gave an explanation for why they thought this, such as that AV introduction should deliver a net positive impact on road safety, that society is more accepting of human errors than those made by a machine and that AVs can be expected to have access to more information and are able to assimilate that information more quickly.

Most respondents this group focused their response on specific outcomes or performance metrics they thought AVs should be able to achieve to demonstrate a higher standard of careful and competent driving. These included:

• not exhibiting unsafe human behaviours such as distraction, impairment, speeding or harsh braking
• being able to easily pass a driving test and recognise and obey safety critical road signs
• being able to exceed human performance in identifying hazards, assessing risks and planning accordingly
• exceeding the 1.6 second perception-response time considered by courts as reasonable for human drivers
• delivering a reduction in KSIs

Several respondents didn’t highlight specific differences in the assessment of careful and competent driving but instead outlined factors and behaviours unique to automated and human drivers which they felt should be considered here. Amongst the factors unique to AVs given were:

• the greater availability of data to AVs to make decisions
• the fact that AVs won’t be subject to bias, poor judgements on gap acceptance, or poor perceptions of approaching vehicle speed
• the fact that AVs may be able to perceive risks humans aren’t aware of, such as traffic signal timing or congestion information
• the fact that AVs won’t be subject to the limitations of human driving, such as distraction, impairment or fatigue

Amongst the factors unique to human driving were:

• the fact that humans can pick up on contextual clues, contextualise risks and the actions of other road users and adapt their behaviour accordingly
• the fact that human drivers are better at reacting to new situations
• that humans are informed by their direct senses and make judgements based on their situational awareness

A small number of respondents argued that it was too difficult to make any meaningful comparison between the assessment of careful and competent driving for humans and AVs. These respondents argued that because careful and competent driving is not clearly defined and because there is not set assessment of careful and competent driving for humans, it is difficult to identify how any assessment of this should differ for AVs.

Careful and competent safety standard

20) In your view, what are the implications of setting a safety standard equivalent to careful and competent human drivers?

There were 58 responses to this question. A number noted that they felt careful and competent was too low as a safety standard. Reasons for this included:

• Careful and competent was not seen as a high standard for human drivers and suggested government acceptance of the current number of fatalities and serious injuries on UK roads.

• AVs were seen as having superior capabilities to humans (such as in their ability to detect objects and not get distracted). A safety standard of careful and competent was seen as unambitious in the context of these capabilities.

• The public may be less accepting of AVs unless they are seen to be significantly safer than human drivers as they are likely to hold machines to a higher safety and moral standard. Public acceptability is therefore dependent on AVs representing a step change in safety, which an equivalent level of safety to human drivers doesn’t secure.

Others felt that a standard of careful and competent would be realistic and reasonable and represented an appropriate balance between realising the safety benefits of early deployment of AVs whilst allowing sufficient space for innovation. Some felt that careful and competent represented a higher standard than an average driver.

Some commented that careful and competent should be the minimum standard AVs had to reach. Others commented that it shouldn’t be a static standard but one that was reassessed regularly to allow for developments in technology.

Several respondents commented on the definition of careful and competent and the metrics used to assess it. Alternative suggestions to careful and competent included the SSSC model, vision zero, quantitative comparisons to human drivers (e.g. being 10 times safer) and embedding careful and competent within the safe systems approach.

Some respondents had concerns regarding the approach of comparing AVs with humans. One respondent noted that any assessment needed to reflect the specific context in which an AV was operating. Another felt that the careful and competent standard was not a viable assessment criterion for type approval or authorisation. A couple of respondents noted the role of the UNECE ADS regulations.

Higher than careful and competent safety standard

21) In your view what characterises a standard higher than careful and competent human driving (providing evidence if possible)?

There were 48 responses to this question.

Respondents provided a range of suggestions that could characterise a standard higher than careful and competent human driving. Suggestions included:

• better safety outcomes for all, defined through specific metrics such as reduced number of collisions or injuries

• hazard detection: earlier and more reliable detection of hazards

• reaction to hazards: taking appropriate action to pre-empt hazards and quicker reaction time when hazards are identified –his includes identifying and reacting to vulnerable road users such as horses and motorcyclists

• adherence to traffic rules

• adaptability to unexpected circumstances and ability to perform better than humans in adverse conditions, such as poor visibility or adverse road conditions

• continuous learning and improvement at a system level through sharing learning across the AV fleet

• ability to detect and respond to faults within the vehicle

• behaves in a predictable and courteous manner with consideration of other road users

• transparent and risk-based decision taking

• consistent performance, where performance is not reduced due to tiredness or stress

A few respondents noted that they disagreed with introducing a standard higher than careful and competent, or felt that such a standard was not realistic or measurable. One respondent noted that there may be differences in approach between reference to behavioural competences that be adopted under a careful and competent standard and the approach adopted within the UNECE ADS regulations.

Implications of a higher safety standard

22) In your view, what are the implications of setting a higher safety standard than careful and competent human drivers?

There were 55 responses to this question.

A number of respondents felt that a higher safety standard would result in road safety improvements in the form of fewer accidents, injuries and fatalities (including for specific vulnerable road users such as horse riders).

A few noted that a higher safety standard would be necessary to achieve wider road safety goals, such as Vision Zero or Safe Systems. A number of respondents noted that a higher safety standard would be necessary in order to promote public trust in the technology and one noted that it provided an opportunity for the UK to be a global leader in AV safety.

A number of respondents noted that a higher safety standard would be likely to have a detrimental impact on developers. This included longer and more onerous pre-deployment checks, slower deployment, higher costs to developers including in collecting data for more in-depth ongoing monitoring, higher costs may limit the number of market entrants and innovation may be stifled.

Some respondents noted that a higher standard may make the UK less attractive for AV investment, with one noting that a higher standard would not align with the careful and competent standard within emerging international regulations. A slower deployment of AVs may then mean that the economic growth, mobility benefits and road safety benefits are reduced.

A few respondents noted that there was a need to balance road safety with ensuring that the standard is achievable and plausible. One suggestion was for the regulatory framework to evolve as the technology develops.

One respondent commented that a higher safety standard may risk AVs driving in a manner other road users find excessively cautious; another respondent noted the importance of AVs driving in a predictable manner so that it doesn’t negatively impact humans driving which could consequently reduce overall safety levels.

Assessing safety impact on other road users

23) In your view what evidence should be used to assess the safety impact that automated vehicles will have on other road users through the hierarchy of road users (providing specific evidence to support your response)?

There were 58 responses to this question.

Many respondents stated that AV deployments should have a positive safety impact for all road user groups, rather than just leading to an overall average improvement in road safety which fails to see improvements (or potentially even sees decreases in safety) for certain road users.

The majority of respondents therefore highlighted evidence demonstrating that AVs are able to ensure safety for all road user groups. Some highlighted pre-deployment data sources, such as test and simulation data showing that AVs are capable of safely interacting with all road user groups and evidence from trials with in-vehicle safety drivers of how the vehicle interacts with other road users.

Others highlighted post-deployment data, such real-world performance data on interactions with other road users and data about incidents involving other road users (e.g. collisions, near misses, traffic infractions). Several respondents argued that incident data must be disaggregated by road user type in order to show whether any particular user group is experiencing an increased risk of certain types of incidents due to AV deployments.

Some respondents focused specifically on vulnerable road users in their responses. These respondents highlighted the same pre- and post-deployment data outlined above but narrowed it down to focus specifically on data around AV interactions with and AV incidents involving, vulnerable road users.

Some respondents argued that current road safety and casualty statistics, as gathered by the police and regulators, should also be collected for AVs and could thus serve as a point of comparison to show whether any road user groups are seeing an increase in incidents due to AV deployments.

Finally, a small number of respondents argued that the requirements in the upcoming UNECE ADS regulations will provide sufficient evidence here. Specifically, the safety case, as assessed through type approval, should demonstrate that the impact on other road users will be positive, whilst the safety management system will allow for continuous monitoring of vehicle performance to show that this is being achieved.

Protected characteristics

24) What evidence are you aware of about the safety impact that automated vehicles will have on groups with protected characteristics?

There were 51 responses to this question.

Respondents highlighted a range of different evidence sources and considerations to be taken into account here. The majority of respondents highlighted concerns and evidence around the ability of AVs to properly perceive people with protected characteristics. Several respondents highlighted known issues with existing AI facial recognition technologies failing to properly perceive non-white faces and argued AV perception systems could present this same issue.

Similarly, several questioned whether an AV will be able to properly identify wheelchair users, those with mobility aids or guide dogs, or those who have different movement patterns. Others raised age as an issue, highlighting that:

• AV sensors may have difficulty in recognising children due to their low height
• older adults are over-represented in pedestrian fatalities and may face difficulties interpreting AV behaviours

Some respondents highlighted potential issues with testing of AVs which could contribute to the above problems. For instance, they highlighted that vehicle systems are currently not tested on dummies of diverse behaviour or appearance and argued that if training datasets or test scenarios are not fully representative of certain groups of protected characteristics, this could lead to biases in AV behaviours.

Some respondents highlighted evidence to show passenger safety concerns in no-user-in-charge (NUIC) vehicles with no on board driver/attendant. For example, one highlighted a report of an incident in the US where a single female rider in an AV was targeted by 2 men who stopped the vehicle and harassed the woman for several minutes. Similarly, the CCAV APS driver roles and passenger inclusivity research found that 21 of the 66 driver roles identified are difficult to fulfil without a driver/attendant onboard the vehicle.

Some respondents also argued that AVs will need to ensure they can meet accessibility requirements for passengers with protected characteristics, such as providing accessible communications during the journey and allowing for the easy securing of wheelchairs. Some highlighted that the CCAV emergency tasks research showed that disabled, older and neurodivergent passengers often have distinct needs and vulnerabilities which will need to be explored further to ensure AVs are able to meet them.

Some respondents highlighted evidence to show the potential benefits AVs may deliver for those with protected characteristics. Some argued that they will increase mobility and independence for these groups, both by removing barriers posed by traditional driving and increasing the availability of transport services. Others highlighted that young drivers are disproportionately represented in road traffic deaths and serious injuries and argued AVs provide an opportunity to address this by reducing the need for younger people to drive or undertake complex journeys themselves.

Finally, some respondents argued that there was little direct evidence available of the impact AVs have on those with protected characteristics and highlighted areas for future research they felt would be beneficial. These included research on:

• AV accessibility design
• biases in AV perception
• community trials of AVs including passengers with protected characteristics
• direct engagements with vulnerable road user and protected characteristic groups to understand their perceptions of AVs and the impacts AVs are having on them as deployments increase

Equalities and fairness safety principle

25) Do you agree or disagree that an equality and fairness safety principle should be included within the statement of safety principles?
26) Why do you think this (providing evidence if you can)?

There were 59 responses to question 25. Of those who responded to this question, the vast majority (85%) agreed with the inclusion of an equality and fairness safety principle. 10% disagreed, with 3% neither agreeing or disagreeing and 2% didn’t know.

There were 29 responses to question 26.

The vast majority of respondents agreed that an equality and fairness safety principle should be included within the statement of safety principles. Those that agreed commented that the inclusion of such a principle would:

• ensure that the benefits of AVs are being shared equitably and that AVs do not create disproportionate barriers to particular groups, such as those with protected characteristics.

• be consistent with existing government policies regarding protecting vulnerable road users

• help to develop public trust in the technology

Comments from those who disagreed with the inclusion of an equality and fairness principle in the statement of safety principles included:

• the need to consider that equality and fairness was already present in the emerging UN ADS regulations –further requirements meant additional complexity with no additional safety gains

• that fairness and equality was not relevant to road safety

• the need to focus the safety principles on all citizens regardless of an equality and fairness principle

A couple of respondents also noted that AVs may result in other inequalities, such as between those who can afford services and those who can’t. Another respondent noted the barriers to travel that disabled people can face and argued that disabled people need to be represented from the start in service design.

The scope of an equality and fairness safety principle

27) Do you agree or disagree that an equality and fairness safety principle should focus on all road users?
28) Why do you think this (providing evidence if possible)?

There were 55 responses to question 27. Of those who responded to this question, the vast majority (80%) agreed that an equality and fairness safety principle should focus on all road users. 2% disagreed, with 11% neither agreeing or disagreeing and 7% didn’t know.

There were 49 responses to question 28.

The majority of respondents to these questions agreed that an equality and fairness safety principle should apply to all. A number of respondents who agreed highlighted issues such as the need to ensure that AVs are safe for all, achieve equitable outcomes and avoid specific negative impacts on specific groups.

The fact that AVs will be operating in a mixed environment with other road users was noted by some respondents. A couple noted that an equality and fairness safety principle would help to build public confidence in AVs.

One respondent noted that AVs shouldn’t be allowed to introduce discrimination, because this would be a backward step compared to current transport services. However, a couple of respondents felt that AVs may need to behave differently around some vulnerable road users, such as children, in order to provide them with additional protection.

One respondent suggested that an equality and fairness safety principle could refer to issues such as:

• accessible human-machine interface (HMI)
• boarding/alighting assistance policies
• service coverage in underserved areas
• fair pricing

Another commented that the principle should reflect wider social goals, such as:

• transport decarbonisation
• modal shift

Those who didn’t answer question 34, or who neither agreed nor disagreed, tended to highlight specific issues regarding the application of the principle or its scope. One respondent supported applying the principle universally but argued there needed to be effective SSSC implementation.

Another highlighted the need to address training bias in AI based systems. Other respondents argued that the principle needed to be extended. One respondent pondered whether it should be extended to animals and whether it captured humans away from roads, such as car parks. One respondent thought AVs should be held to a higher standard in relation to vulnerable road users such as pedestrians.

Monitoring an equality and fairness safety principle

29) In your view, what metrics, if any, should be considered to support monitoring and evaluation of performance against an equality and fairness safety principle?

There were 47 responses to this question.

A significant number of respondents felt that collision data, focused on fatalities and serious injuries, would be relevant to monitoring the equality and fairness safety principle. In particular this could be monitored against different road user groups although a number of respondents felt that the protected characteristics of those involved in an incident should be recorded.

Such data could be used to identify whether any individual road user group or protected characteristics was disproportionately impacted by the deployment of AVs.

A couple of respondents noted that there was already existing data relating to accidents and argued that there was no need for AVs to be treated differently to conventional vehicles.

They commented that any metric needed to present contextual information, to avoid giving the public a false impression about the performance of AVs. Another respondent noted that in-service monitoring and reporting data from AVs may be used to compare human driver safety versus deployed AV safety on the public roads.

A few respondents felt that near misses should also be recorded and that this should record incidents with protected characteristics or vulnerable road users. One respondent highlighted the types of manoeuvres that might be captured in near miss data, such as sudden braking, or evasive manoeuvres triggered by vulnerable road users – including motorcyclists.

A few respondents commented on the need for indicators relating to user outcomes. Suggestions included:

• satisfaction of relevant users – potentially segmented by protected characteristics
• confidence surveys, or indictors of wider public trust of AVs such as how comfortable different road users feel around AVs or complaints about AVs

A few respondents suggested monitoring whether AVs were serving communities in an equitable manner. Suggestions included:

• accessibility of AV services for people with disabilities
• geographic deployment across deprived and non-deprived areas
• consistency of safety outcomes across demographic groups

A couple of respondents highlighted other accessibility considerations they felt should be monitored, such as how the percentage of vehicles being accessible or the accessibility of service stops, needed to be monitored.

A few respondents suggested there should be metrics relating to the performance of an AV, in particular relating to an AV’s ability to detect and respond to different road user types. This could help to identify any potential biases in AI based training.

Outcomes

30) In your view, what outcomes should be considered for the monitoring and evaluation of performance against the statement of safety principles?

There were 52 responses to this question.

In terms of lagging metrics, respondents suggested that the number of collisions involving AVs should be reported, disaggregated by severity (such as whether there were fatalities, serious injuries, minor injuries or property damage) and by road user group.

This data may need to be related to suitable exposure metrics which take into account the operational design domain. Another suggestion was to report on collisions per distance travelled.

Many respondents suggested that leading metrics should also be reported on. The most suggested metrics were the number of breaches of traffic rules and near misses.

Other suggestions included:

• reporting on erratic vehicle behaviour
• harsh braking or swerving
• minimal risk manoeuvres
• speed
• disengagements events, such as when a remote driver was required

A few respondents suggested measures relating to system performance, such as reliability of safety critical components or software changes.

Some respondents, notably from industry, highlighted that the post-deployment monitoring should focus on assessing whether the claims in the safety case assessed against the UNECE ADS regulation were appropriate. These respondents noted the need for harmonisation (and avoiding duplication) with the ADS regulation and proposed using data from ISMR and DSSAD as information sources.

Other suggestions for metrics included gathering data on the public perception of AVs, customer satisfaction or complaint levels and the impact of AVs on people’s willingness to use other sustainable travel modes.

Other comments made in response to this question included that it could take years to gather enough data to assess AV safety performance, suggestions for how reporting information should be published and the cost of AV deployment to councils and government.

Information to monitor outcomes

31) In your view, what sources of information could be used to monitor and evaluate performance of these outcomes?

There were 50 responses to this question.

The most common suggested source of information from respondents was data from AVs directly. This included data collected by the vehicles themselves, such as data recorded directly from the vehicle’s sensors and on the DSSAD, as well as video. One respondent suggested this should include weather related information, such as use of windscreen wipers.

It also included data provided by ASDEs and NUIC operators. Suggestions included:

• data provided under the in-service monitoring and reporting process
• fault and service history of the vehicles and operational records

Some respondents explicitly suggested that this data should cover near misses as well as incidents. One respondent suggested that only data required under the ADS regulation should be collected.

Another common suggested source of data related to collisions. Many respondents suggested collision data was a source of information. Specific sources suggested the STATS-19 database, data from police and Highways authorities and insurance companies. A couple of respondents also suggested hospital records linked to AV incidents may be a useful source of information.

In addition to collision data, some respondents suggested information about traffic infractions should be sourced, including data from traffic cameras, CCTV cameras and telematic data.

Other sources of data identified included:

• existing inspection and surveillance frameworks, such periodic technical inspections and market surveillance activities

• information from garages who are authorised to service and repair NUIC vehicles

• number of complaints or feedback received from AV users

• public feedback –suggestions included having a mechanism for the public to report incidents, public or stakeholder surveys, focus groups, perceptions of AV safety through national travel attitudes survey

• a mechanism to enable police to provide feedback to relevant authorities

• pre-deployment information might include simulated data, test track studies and on-road testing

There were a range of other comments received as part of this question including:

• the need for information to be collected or verified independently

• a suggestion that AV safety performance should be compared across different AV companies

• a suggestion AV data should be compared against other countries

• Identifying a number of potential challenges with accessing data, such as quality, interpretability and universality of the information received and how it will be retrieved and used

• the need for government to ensure that information is accurate, timely and comprehensive

Evidence to compare human and automated driving

32) In your view, what evidence sources could be used to compare the safety performance of human drivers and automated vehicles?

There were 46 responses to this question.

A number of respondents commented that comparisons between human drivers and AVs may be difficult to make. Respondents highlighted that AVs have different abilities to human drivers and will have different capabilities which may vary across their operating environments.

One respondent considered whether it was possible to identify careful and competent human drivers in order to make the comparison; another felt that AVs should be compared to all drivers and not just those who are careful and competent.

A number of industry respondents noted that the comparison with human drivers was unhelpful and unnecessary given the role of the ADS regulations in ensuring AV safety. They highlighted that manufacturers would be likely to consider human driver related safety models as a baseline in their safety case and suggested that the safety case may be a useful source of evidence in making a comparison with human drivers. Another industry representative felt that it would be effective to consider safety through the safety case rather than comparing with human drivers.

In the event that a comparison was to be made, a number of respondents noted the need to ensure it was fair. In particular, a number of respondents noted the need to ensure that sufficient context was provided – for example, using exposure data (such as vehicles miles travelled by road type, time of day, weather) which would allow for situational factors that might influence the safety outcomes.

A couple of industry respondents commented that AV performance will depend on their specific organisational design domain and capabilities, in which case non-standardised data risks being misconstrued without the appropriate context. These respondents argued that industry data needed to be aggregated and anonymised

A significant number of respondents noted that publicly available statistics would provide an evidence source for making the comparison. Data sources proposed included collision rates from STATS-19, killed and seriously injured statistics, police/court data and data from insurance companies. One respondent noted the limitations of STATS-19 (for example risk of under-reporting) and another noted the need to ensure that it recognised AVs as a distinct user category.

A number of respondents identified that operators of AVs would need to provide data; suggestions included incident logs, near-miss events and traffic violations. A number of respondents noted that data would be provided under the in-service monitoring and reporting requirements within the UNECE ADS regulations.

Other suggestions for evidence sources included data from connected cars, CCTV and analysing traffic camera footage. A few respondents suggested alternative means of making a comparison – these included using naturalistic driving studies, theoretical scenarios, public focus groups considering individual incidents, or considering lessons from other industries such as air, maritime or aviation.

One respondent noted that any comparison needs to go beyond considering collision outcomes and instead consider whether AVs are driving safely. The metrics proposed to measure this were:

• no at fault crashes, as per insurance i.e. crashes should be avoided where possible
• an appropriate level of care and respect for other road users, as per the Highway Code rules for road users requiring extra care
• a KSI / million miles driven rate significantly lower than the average human driver

Reporting on annual performance

33) In your view, what metrics comparing the safety performance of human drivers and automated vehicles should be annually reported on by the Secretary of State?

There were 47 responses to this question.

A significant majority of respondents proposed using collision or accident statistics to facilitate a comparison. These suggestions included:

• Collision rates, including total number of collisions and collusion rate per mile travelled. Some respondents had suggestions for how this data might be broken down further, including by severity, geography, collision type (e.g. rear-end shunt) and other road users impacted.

• Killed and seriously injured statistics. This might include the total of KSI caused by AV, KSI per mile travelled or per 100,000 km travelled, road type, severity of injuries and road user types.

In terms of sources of data for these metrics, respondents suggested court/police records, data from insurers, DfT statistics, analysis of in-service reports provided by manufacturers, ASDEs and/or NUiCO and insights from statutory safety investigations under Section 38 of the automated vehicles Act 2024.

A number of respondents felt that collision and KSI statistics for AVs can be compared with the same statistics for human drivers and presented in a similar way. One respondent suggested that the accident rate per 100,000 km should be compared with professionally qualified human drivers with equivalent sized vehicles and that data should be collected separately for NUICO and safety driver operations.

One respondent noted that the data on AVs should be slotted into the existing statistics and should not got beyond what is collected for STATS 19, while another questioned if a new system would be needed for analysing and reporting on such a comparison.

Many respondents suggested near miss data (including near misses per distance travelled) and leading metrics should be reported. Suggestions included:

• harsh braking, G forces, swerving, minimal risk manoeuvres

• disengagement of the ADS system

• erratic or unexpected vehicle manoeuvres, which could be disaggregated further to provide context (e.g. road type, environment)

• frequency of traffic infractions

• operational design domain compliance, including performance across different weather, lighting and road conditions

A number of industry respondents disagreed with the inclusion of leading metrics such as near misses, arguing that they were neither useful or practical for evaluating ADS performance. Another respondent noted that near miss data would not usually be available for human drivers so a comparison would not be possible. The challenges in comparing human and automated driving were noted by another.

A number of respondents noted the importance of providing context in publishing the data and in particular highlighting occasions when an AV was involved in a collision but was found not to be at fault.

A couple of respondents suggested that the use of AVs should be recorded, such as:

• the total mileage operated by autonomous NUIC
• autonomous with safety driver and human driven vehicles
• AV penetration (number of vehicles listed, in active operation) vs number of non-ADS-approved vehicles

One respondent noted that DfT’s road safety data provides context and explanation and argued that this approach should be replicated with AVs. Another respondent suggested that fair reporting criteria will include:

• objective severity tiers to ensure like-for-like comparisons

• the use of ODD-specific benchmarks to reduce reporting bias and further enabling like-for-like comparisons for different collision types

• a form of information presentation that is based on miles driven, rather than just overall numbers

Respondents suggested a number of other metrics that might be included with the Secretary of State’s annual report. These included:

• Outcome focused measures. One example was providing more details regarding vehicle accidents, such as Driver licence status, driver intoxication (drink or drugs), Owner driver or thief, weather conditions and location. Another is evidence of AV compliance with the Highway Code’s hierarchy of road users.

• Behaviour focused measures. One example is the anticipatory behaviour of an AV and their ability to foresee when a situation is becoming risky, although one respondent noted this may be difficult to assess. One respondent suggested that AVs should be assessed in accordance with the 27 driving tasks human drivers must pass in order to be granted a driving licence.

• System focused measures. This might include standardised data on software updates and safety-critical system failures made by the AV vehicle manufacturer (potentially including the rationale for the change) and Evidence of user in charge behaviours/conduct in respect of software updates.

A few respondents suggested metrics relating to the impact of AVs on others, such as:

• Surveys from different user groups and disability groups/elderly about issues they face using the road around AVs – one respondent was concerned that perceptions of safety around AVs could impact people’s choices to travel by sustainable modes such as walking, cycling and public transport

• user/consumer complaints to manufacturers and key road user stakeholders

• health impacts on victims in AV related incidents (severity and frequency)

• monitoring human driving standards over time – one respondent was concerned that AVs could reduce human driving competency if they behave in certain ways

In terms of publishing, a couple of respondents suggested that data should be based on a weekly or monthly basis which could be supplemented by quarterly or annual reporting to capture lagging factors or cumulative impacts.

A few suggested that AV operators should be required to publish regular public facing reports on AV safety, to help build public trust in the technology. One respondent suggested that the National Highway’s stocktake on smart motorways may provide a useful format.

Other respondents suggested that the process should be as closely aligned with the DfT’s annual statistical release as possible, providing the benefit of consistent format and a well understood methodology. One respondent suggested that the data on AVs should be drawn from mandatory data submissions but the need to aggregate and standardise this data was noted. One respondent requested that government agree with industry how the monitoring report will be presented.

Excluded principles

34) Do you agree or disagree with our proposed approach to these potential principles?
35) Why do you think this (providing evidence if possible)?

There were 61 responses to question 34. Of those who responded to this question, 61% agreed with our proposed approach. 26% disagreed and 13% didn’t know.

There were 45 responses to question 35.

Many of those who agreed with our approach agreed with our explanation in the call for evidence as to why the principles should be excluded from the SoSP.

Specifically, many agreed that the ability to drive without human monitoring or control and cyber resilience were better considered and assessed through type approval and authorisation.

Some stated that they agreed with these principles being considered elsewhere in the regulatory framework but highlighted that they must still be given due consideration and that their omission from the SoSP mustn’t lead to any reductions in safety, security or transparency.

Amongst those that disagreed, the vast majority did so on the basis that they felt a principle on cyber resilience and/or explainability should be included in the SOSP. A range of reasons were given for this.

For example, those that felt a cyber resilience principle should be argued that type approval is a fixed checkpoint, whereas cyber resilience requires continuous adaption and so should be built into the SoSP to allow for continuous monitoring of resilience. Some also argued that UNECE R155 is designed for human driven vehicles and so on its own may not be enough to ensure cyber resilience.

Those that felt an explainability principle should be included tended to focus on the impact not including it would have on public trust. For example, they argued that the public needs to understand why AVs have behaved the way they in order to trust them, that removing explainability could weaken ethical oversight of safety decisions and could create challenge in ensuring equity and fairness, which would in turn impact public acceptance.

Finally, a small number of respondents highlighted other principles they felt should be included in the SoSP, such as a principle on maintenance workforce competence. A small number also argued that the SoSP should be regularly reviewed as AV deployments increase and the technology develops in order to assess whether additional principles need to be included.

Inclusion of other principles

36) In your view, are there any other principles you think should be included within the statement of safety principles?
37) What other principles do you think should be included and why (providing evidence if possible)?

There were 55 responses to question 36. Of those who responded to this question, 45% felt there were additional principles that should be included within the statement compared to 22% who disagreed. 33% didn’t know.

There were 34 responses to question 37.

Respondents who answered ‘Yes’ to Q42 provided varied responses to Q43, with only a few minor convergences on the additional principles they felt should be included. Some felt that a principle to ensure transparency on the part of regulated bodies and AV manufacturers should be included, with arguments for this including that full transparency around safety data and close working with regulators by AV companies would be crucial to achieving public trust in AVs.

Several respondents argued a principle around integration with the wider transport system, or integration into a safe system approach, should be included. They typically felt that this principle should examine how well AVs comply with safe system principles and integrate into the wider transport ecosystem and argued implementing this principle would ensure AVs contribute to safety across the transport network.

A smaller number of respondents argued for the inclusion of additional principles focused on whole lifecycle safety of AVs. The additional principles proposed by these respondents included ones on maintenance technician competence, recognition that safety assurance should recognise vehicle component degradation, requirements for safety cases to be continually updated as real-world learnings emerge and requirements for regulated bodies and manufacturers to demonstrate pro-activity in safety improvements through continually learning from their operations.

A small number of respondents argued a principle on data should be included, such as one requiring that data captured or used by AVs is exclusively stored and processed in the UK, or one that ensures data from AVs can be leveraged for safety benefits beyond individual vehicle performance (for example, by allowing local authorities to use aggregate near miss data to identify potential incident hotspots).

A small number of respondents also highlighted additional equality and fairness principles which should be added, such as one ensuring that AVs deliver improved safety for vulnerable road users.

Amongst respondents that answered ‘no’ to Q43, several followed this up by stating that it was not possible to comment on whether any additional principles should be included without first seeing a draft of the SoSP.

Further evidence

38) Provide any further evidence you wish to submit for consideration on what safety expectations should be set for the deployment of automated vehicles.

There were 25 responses to this question.

This question received mixed responses, with respondents presenting a range of additional evidence for consideration. Several respondents highlighted the importance of creating a culture of shared learning and transparency within the AV industry to ensure continuous improvements in safety.

These respondents acknowledge the importance of knowledge sharing and transparent data reporting amongst regulated bodies, vehicle manufacturers and regulators and other involved stakeholders in achieving this and highlighted the aviation and rail industries as examples of good practice. Several respondents also highlighted the importance of transparent data reporting in building and maintaining public trust in AVs and emphasised the importance of high public trust in ensuring successful AV deployments on UK roads.

Other respondents highlighted the need to ensure AVs remain roadworthy throughout their use and suggested mechanisms through which this could be achieved. For example, some respondents suggested an annual MOT test or equivalent AV-specific check could be used to verify the vehicle and ADS remain fit for use. Several respondents specifically also commented on the need to ensure the vehicle’s software remains up to date, highlighting that users will need to be made aware of whether their vehicle’s software is up to date and raising questions about who should be responsible for this being the case.

Several respondents also highlighted the impact AVs might have on VRUs and argued this should be given special consideration as part of the development of the SoSP to avoid AVs disproportionately impacting VRUs. For example, some respondents highlighted the unique safety considerations posed by motorcyclists and horse riders and raised concerns about whether AVs will be sufficiently capable of detecting, interacting with and navigating around these types of road users.

Other comments

39) Any other comments?

There were 47 responses to this question.

This question received mixed responses, with respondents highlighting a range of additional considerations. Several respondents put forward the need for the UK to align with the development of international AV regulations at the UNECE and the forthcoming UNECE ADS regulations. These responses argued that the SoSP should not seek to duplicate the UNECE regulations or apply new or different requirements to those already being established through at the UNECE.

Several respondents also argued that the SoSP will need to be dynamic and evolve over time. These respondents thought that much of the evidence used to inform the safe development and deployment of AVs will come from real-world testing and so the SoSP will need to be able to evolve in response to findings from real-world testing and subsequent technological developments.

Many of these respondents also warned that government should not place undue delay on the implementation of the SoSP and wider regulatory framework to ensure that learnings from real-world deployments can start being captured as soon as possible.

Finally, several respondents also highlighted key themes that have emerged in their responses to previous questions. These included the impacts that AVs may have on VRUs and specific road user types, and the importance of ensuring public trust in AVs and how transparent data reporting will be required to achieve this.

Next steps

Thank you to all respondents for their feedback to this call for evidence.

The feedback received has informed the development of the statement of safety principles and this is set out in more detail in the statement of safety principles consultation.

The results from the call for evidence will also be used as we continue to take forward the development of the wider regulatory framework for AVs, including developing our approach to the general monitoring duty.

In terms of next steps, the draft statement of safety principles will be subject to public consultation. Following the consultation, a draft will be laid before Parliament, who must approve it before it can come into effect. The statement of safety principles will be used alongside the wider regulatory framework for AVs, which is expected to be in place in the second half of 2027.