Autonomous vehicles: how safe is safe enough?

The potentially enormous safety benefits of self-driving vehicles have long been considered to be among the technology’s biggest assets.

Numerous research projects have found human error is a contributing factor in between 85% and 95% of current road collisions.

The conventional thinking has been that if you remove human error through the use of fully autonomous technology, then the collision rate would fall by a similar amount.

This has been a strong selling point for self-driving vehicles to a public which, so far, seems unwilling to trust the technology.

For example, research conducted last year on behalf of the Institution of Mechanical Engineers found 60% of people said they would always prefer to drive themselves rather than use a self-driving vehicle, while two-thirds of people are uncomfortable with the idea of travelling in a driverless car.

Part of this could be down to unfamiliarity with a technology which is still being trialled and developed, and is many years away from being a common sight on the roads.

But the way the mainstream media overlooks the many hundreds of thousands of incident-free miles travelled in self-driving vehicle trials around the world while sensationally covering collisions also has an impact, argue autonomous vehicle (AV) advocates.

“The headlines go ‘whoosh’ (if there is a collision),” Ben Boutcher-West, head of mobility at kerbside management company AppyWay, told a Westminster Energy, Environment and Transport Forum conference on autonomous transport in the UK.

“The way the media handles some of those events make it very difficult for any OEM to put their name forward and push out a service.

“It’s the way they (driverless cars) are perceived. That for me is all about media and education and the moment these vehicles put a foot wrong, they will be battered like crazy by people who maybe don’t understand the full situation of what actually occurred.”

A study released in America this month by the Insurance Institute of Highway Safety (IIHS – see panel below), found the perceived safety benefits of AVs could be significantly lower than commonly believed by the wider AV sector.

It claimed self-driving vehicles might prevent only one-third of crashes if automated systems drive too much like people.

“It is likely that fully self-driving cars will identify hazards better than people,” says Jessica Cichone, vice-president for research at IIHS and a co-author of the study. “But we found this alone would not prevent the bulk of crashes.”

The study was criticised by companies and organisations working on self-driving vehicles who argue that it underestimates the technology’s capabilities.

 

No mistakes can be made

However, any negative publicity can reinforce opposition to the technology and Brian Wong, director at specialist transport law firm Burges Salmon, warns: “If the societal acceptance (of self-driving vehicles) is going to change, then nobody, and least of all the connected and autonomous vehicle (CAV) industry, can really afford for mistakes to be made.”

This places extra importance on the success of AV trials, a number of which have already been carried out in the UK.

These include the Nissan-led HumanDrive project which, in November, saw a modified Nissan Leaf electric car cover 99% of the 230 miles between Milton Keynes and Sunderland in fully autonomous mode, and Driven, led by software developer Oxbotica.

This £13.6 million project ran from April 2017 to December 2019 and focused on completing fully autonomous routes within the complex urban environments of London and Oxford.

The Government has produced a code of practice to provide guidance on trialling AV technologies on public roads or in other public places in the UK.

It makes recommendations on how to maintain safety and minimise potential risks, and was this year supplemented by two new key documents.

PAS 1881 Assuring Safety of Automated Vehicle Trials and Testing was released by the British Standards Institution (BSI) in its role as the UK’s national standards body, and Zenzic, the organisation dedicated to accelerating the self-driving revolution in the UK. In addition, Transport Research Laboratory (TRL) created an updated Safety Case Framework Report 2.0.

PAS 1881 has been delivered in conjunction with the Centre for Connected and Autonomous Vehicles (CCAV), Department for Transport (DfT) and Innovate UK.

Document author Camilla Fowler, head of risk management at TRL says it aims to accelerate the safe use of connected and autonomous vehicles with guidance and technical standards.

It includes a safety case which details the aim of the trial and what technology is being used. This filters into a risk assessment as well as identifying what action can be taken to mitigate any risks.

“Until this standard was released, there hadn’t been any regulations or standards that document what should be within a safety case,” she says.

“Building trust is about addressing fears over safety and security and one of the key things we need to make sure of is that we are very
transparent in our approach to managing those fears.

“Publishing safety cases will go towards helping public trust in AV trials and testing so they can understand what it is that is happening, how many vehicles and where is it happening, what the test objectives are, what are the key risks and what are the control measures.”

New risks

However, risks – as with all new technologies and road transport – will remain.

“There are still more than 27,000 people killed or seriously injured on our roads each year and while CAVs have real potential to reduce that number significantly, they also could bring new types of risks” says Catherine Lovell, deputy head of the Government’s CCAV.

“The sensors could fail to properly gain information about the environment around them, the vehicle could fail to correctly interpret that and choose a safe driving course.

“Or they might be vulnerable to things like cyber-attack in a way that current vehicles are not. So, in CCAV, we’re trying to sort of bring those benefits forward as fast as possible while also being aware of those risks and tackling them.”

It is clear that setting the right expectations for the safety of self-driving cars is an important factor in winning public acceptance for the technology.

And while it would be possible for AV developers to strive for close to zero risk of causing a collision, injury or fatality, it would take a very long time to develop and prove that systems are at that level, says David Hynd, chief scientist for safety and investigations at TRL.

“There is a balance to be made,” he adds. “If you wait that long, a lot of people will have been injured and killed in the meantime, so part of the idea is to find a good balance between what you are really aiming for long-term and being able to save lives and serious injuries as you go along that journey.”

So, how safe is safe enough for an AV?

“It sounds like a very simple question, whereas it’s a really big and quite a difficult question to answer,” says Hynd.

“A lot of people talk about defining safety in terms of a comparison with human drivers, so you could say it’s got to be at least as safe as human drivers.

“It’s got to have no more collisions, no more serious injuries, no more deaths than we currently have on the road network.

“But, if you think about the number of collisions that involve a human component or some kind of failing from the human driver such as drink-driving or speeding, if the car is doing the driving task then it automatically doesn’t have any of those things.

“For me, the target has got to be – as a minimum – that it does at least as well as a very good, alert human driver who is paying attention to the driving task. That is still quite a woolly definition, but is quite a lot safer than humans on average because everybody sometimes is not as awake as they should be or is not paying as much attention as they should be, and we have other poor behaviours on the road as well.”

 

Crash reduction likely to be less than expected, says new report

Self-driving cars could reduce collisions by a significant amount less than commonly-held industry expectations, according to new analysis by America’s Insurance Institute for Highway Safety (IIHS).

The technology has sometimes been touted as key to reducing crashes to almost zero, but the research group, funded by US insurers, found self-driving car technology may actually cut collisions by just a third.

“Building self-driving cars that drive as well as people do is a big challenge in itself,” says Alexandra Mueller, research scientist at IIHS and lead author of the study. “But they’d actually need to do better than that to deliver on the promises we’ve all heard.”

For the study, researchers analysed more than 5,000 police-reported crashes and determined the driver-related factors contributing to those.

They imagined a future in which all the vehicles on the road are self-driving. They assumed these future vehicles would prevent those crashes that were caused exclusively by perception errors or involved an incapacitated driver.

That is because cameras and sensors of fully autonomous vehicles could be expected to monitor the roadway and identify potential hazards better than a human driver and be incapable of distraction or incapacitation.

Crashes due to only sensing and perceiving errors accounted for 24% of the total and incapacitation 10%.

The study concluded these collisions might be avoided if all vehicles on the road were self-driving – though it would require sensors that worked perfectly and systems that never malfunctioned.

The remaining two-thirds might still occur unless autonomous vehicles are also specifically programmed to avoid other types of predicting, decision-making and performance errors.

However, the autonomous vehicle industry in the US says its cars are programmed to prevent a vastly higher number of potential crash causes, including more complex errors caused by drivers making inadequate or incorrect evasive manoeuvres.

Taking those design choices into account, autonomous vehicles could avoid some 72% of crashes, countered Partners for Automated Vehicle Education, a consortium of self-driving technology companies.

The group says it is “fundamentally speculative” to determine crash avoidance rates. 

It adds: “We believe that reducing traffic fatalities by even a third would be something to be proud of. We aim to do even more.”

 

Other benefits of driverless cars

More accessible transport

In theory, driverless cars mean no driving licence, so people of all ages and abilities could access mobility. There is great potential for enabling older people and those with disabilities to travel.

“I can see AVs being very useful for people who maybe have health issues and are unable to drive, as it may increase their mobility and freedom,” says Camilla Fowler, head of risk management at TRL.

 

Reduced emissions

Widespread adoption of self-driving vehicles also has the potential to reduce energy consumption and emissions.

This can be done by optimising traffic flow for fuel consumption and platooning where AVs travel very close to each other to reduce aerodynamic drag.

If used as smart taxis or autonomous ride-share, AVs could require a much smaller fleet to service travellers’ needs.

“People often talk about the safety aspects, but I think there are very clear potential benefits in terms of minimising the use of energy to get people from A to B,” says David Hynd, chief scientist for safety and investigations at TRL.

“These benefits – in terms of efficiency and energy consumption – might actually come to be seen as the bigger wins for autonomous vehicles in the long-term.”

 

Cheaper transport

The costs of drivers and safety requirements (driver rest breaks etc.) are a major outlay for transportation companies. Vehicles that drive themselves would cost less to operate, enabling more, cheaper taxi and ride-sharing-type services.

KPMG says roughly half the cost of on-demand private hire vehicles relates to the driver and, as a result, estimates that AV mobility as a service provision could be up to 40% cheaper than private vehicle ownership by 2030.

 

Congestion reduction

In theory, driverless cars could organise themselves to optimise road use by ‘platooning’ and by automatically rerouting to avoid congestion.

“Spatially-aware vehicles will drive together at no cost to safety and future capacity increases will be achieved by platooning or cooperative adaptive cruise control,” says Lizi Stewart, managing director, UK transportation, at Atkins.

“Platooning will allow cars to drive with shorter headways and gaps of just 0.5 seconds is the equivalent to at least another whole lane on the motorway.”

Working with smart traffic control could further optimise road use and increase road safety.