22nd May 2019

The interaction of autonomous vehicles with their surroundings – what do the present and future look like?

With Elon Musk aiming for fully-autonomous ‘level 5’ Tesla models to be on the road as soon as next year1 – and with the UK government recently having expressed the wish to see “self-driving cars on our roads by 2021… on any road provided they are compliant with UK law”2 – autonomous vehicles (AVs) imminently becoming an everyday reality may seem inevitable to many casual media consumers.

However, with a regular stream of accidents involving AVs, from the admittedly very small number of tragic deaths to less consequential collisions such as when the ‘an Apple hit by a Leaf’3 episode occurred in California when a driven Nissan LEAF hit the rear of a self-driving Lexus RX 450h tested by Apple, it’s clear that safety concerns are a long way from being fully answered. Additionally, AVs are themselves often EVs running on electric power, the quietness of which may be embraced by urbanites but poses further safety worries. We look at the latest news, developments and views relating to the interaction of AVs with other vehicles of varying kinds, and also with pedestrians, cyclists and other animate and inanimate ‘objects’, as they would interpret them.

Recent SMMT analysis3b identified the UK, perhaps in a slightly biased way, as the ideal country for supporting autonomous vehicles, but it’s fair to argue that multiple and rather major adjustments will be required, from driving legislation and motor insurance to the Highway Code, physical road layouts and infrastructure such as traffic lights, all unarguably causing upheaval for a time, which some may feel is worth it for progress’ sake. It’s ironic that although driverless cars and other vehicles have been tested for much longer in the States, the legal frameworks in the U.S. that enable their use are less advanced than in the UK and in some Asian countries.

Who is liable if an AV is involved in an accident?

In February, ahead of the organisation’s annual conference, the Association of British Insurers (ABI), along with Thatcham Research, issued a response to the Law Commission, the body continuing to assess how driving rules being devised by the United Nations Economic Commission for Europe may be integrated with the UK’s own laws enabling AVs. ABI and Thatcham’s ten criteria for defining an autonomous vehicle quite correctly include assertions that AVs themselves much be programmed to be law-abiding, able to handle unanticipated as well as clear handovers, equipped to perform emergency stops, fitted with backup systems and provide accident data.

Reassuringly for the public and indeed fleet drivers, the ABI’s policy adviser for motor insurance, Laurenz Berger, stated: “There will come a point when the drivers of today are effectively passengers in a vehicle driving itself, and at that stage we want motorists to be reassured they won’t be held liable for an error made by a vehicle or a piece of on-board technology.”

The Automated and Electric Vehicles (AEV) Act, passed by the UK’s parliament in summer 2018, is welcome in making it clear that drivers won’t be liable if technology fails23 and that they will be covered whether they were at the steering wheel or not.

How motor insurance could evolve to facilitate autonomous vehicles

It would likely prove overly complicated to require autonomous vehicle ‘owners’, leasing customers or other users to have two separate insurance policies in place to cover both fully-autonomous driving and times when the human fancies assuming control, so a single combined policy would make more sense during the grey period in which various vehicles may still provide steering wheels and other conventional controls.

Autonomous vehicles are ultimately expected to reduce and perhaps even eradicate road traffic accidents and fatalities, heralding significant changes to the motor insurance landscape. Again, though, during the transition phase when AVs will share the road with driven vehicles and the formers’ technology continues to be refined, it’s possible that insurers will increasingly seek to recoup a percentage of their costs when OEMs’ autonomous systems contribute to accidents.

The AEV Act is regarded by some as deficient in its addressing data and insurance recoveries, with no firm framework laid out, meaning that it will be vital for OEMs and insurers to work together increasingly closely in ensuring suitable access to, use of and sharing rights in respect of AV accident data and other statistics.

Autonomous stages

Vehicle owners, which will likely include contract hire/leasing funders, ride-sharing and subscription operators and other organisations who buy or lease vehicles from their respective manufacturers, will plausibly be legally mandated to ensure that the software in AVs under their care is updated promptly, although it’s currently even less clear what would happen if a vehicle was hacked.

AVs intermingling with conventional driven vehicles will unarguably herald a period of initial complexity for insurers, legislators and others. But when the latter are eventually phased out perhaps entirely by driverless iterations, which likely won’t even include steering wheels and controls in the future, accidents could indeed be reduced to almost if not actual zero.

While traditional premiums will significantly drop following the advent of driverless cars, the insurance sector will undoubtedly evolve and adapt, the focus turning more to liability-type coverage and products, while policies will shift from personal to fleets24.

Tesla’s take on usage-based insurance (UBI)

Ever the disruptor, Tesla4 is so confident in its Autopilot technology (which is admittedly only permitted as an advanced driver assistance system (ADAS) in most jurisdictions rather than allowing its promised full autonomy) that this month will see it offer insurance to its customers whereby premiums will be lower for those who use Autopilot more regularly.

Elon Musk recently clarified that if a Tesla driver insured by the firm uses their vehicle “in a crazy way”, their insurance payments could increase. UBI is far from a new concept, our own Carrot Insurance broker for young and newly-qualified drivers continuing to lead its specialised segment several years on, but when it comes to AVs we envisage an increasing number of OEMs offering their own bundled insurance, with premiums based on big data including telematics.

Entrepreneurs address changing risk

A Californian entrepreneur and venture capitalist, frustrated after his insurer quoted him over $10,000 per annum if he bought a Tesla Model X, has created an insurance product called Avinew5 and formed agreements with ‘most’ manufacturers offering autonomous functionality such as Ford, Nissan and Tesla. Avinew will monitor drivers’ usage of autonomous systems and the data continuously produced, with the aim of writing its own car insurance policies in certain states later this year at significantly reduced premiums compared to traditional underwriters and actuaries who, to paraphrase founder Dan Peate, don’t understand the changing risks and don’t have enough data.

AVs’ interaction with the world around them

Ford is developing a sophisticated system based on self-driving vehicles using roof-mounted lights7 to communicate their movements and intentions to pedestrians in order to engender trust, with turquoise found to be the most readily identifiable colour.

Lyft has interestingly patented what it sees as the solution23 to enable AVs to communicate with humans, a preset message displaying on the windscreen by means of a projector or other method, advising people that it is, for example, safe to cross the road. Although we’ll have to see what other communication ‘languages’ for AVs materialise, work by Ford, Lyft et al is a positive step forwards.

Self-driving vehicles’ near-silent nature also requires a solution and Volkswagen recently commented to Reuters that the artificial sound assigned to an electric vehicle8 “has to be futuristic and cannot simply add the sound of a combustion engine”, alluding to SUVs and sports cars emitting a deeper sound. Mercedes, meanwhile, has taken a different approach with its fully-electric EQC SUV, designing it to be as quiet as possible for occupants while producing an artificial humming noise to warn pedestrians. This is the same tack Jaguar9 has taken in developing a unique Audible Vehicle Alert System (AVAS) for its I-PACE in conjunction with charity Guide Dogs for the Blind.

Developing sounds for EVs and AVs is laudable and wholly necessary, but such steps probably won’t prove effective in every case, an article on The Register10 rightly pointing out that “urban trotters currently take their lives into their hands by never looking up even when they do hear the growls of the good old combustion engine.”

Driven vehicles mixing with autonomous cars, vans and trucks, some of which may ultimately move in platoons, will undoubtedly pose initial challenges and dangers, so it’s reassuring that virtual reality (VR) simulators12 including Matlab Simulink and PreScan are being used to test the behaviour of semi and full-AVs, with technologies such as cooperative adaptive cruise control (CACC) materialising with the end aim of platoon lead vehicles communicating more precisely to those following in convoy behind. Lane-changes should also be made safer thanks to the groundwork laid by VR simulators.

In the UK, the Ordnance Survey has recently embarked on a project called OmniCAV whereby a detailed 3D model of the roads between Oxford and Abingdon is being created, partly using driven vehicles that scan the surroundings18. Data from traffic cameras along with near-miss analyses will also form part of the 3xD model built up providing a highly accurate simulator environment complete with VR pedestrians, cyclists and other road users, with which CAVs can be tested safely before being let out onto county’s roads.

OmniCAV’s work will in due course contribute towards a certification tool19 for connected autonomous vehicles, which insurers, OEMs, regulatory and accreditation bodies can use to ensure that self-driving vehicles are developed safely.

Weather, too, poses a challenge for OEMs and other AV developers. Snow and even heavy rain can quickly obscure all-important lane lines and other road markings, while fog, sandstorms and leaves can render complex and expensive camera and radar technology ineffective. Relatively little has been done to address how AVs will react in wintry weather, because most of their testing continues to be, perhaps wrongly, carried out in warm regions. Researchers16 are working on software to enable self-driving vehicles to better differentiate obstacles and physical weather such as large snowflakes, and advanced laser sensors are also being developed to see through weather more reliably – but with progress comparatively slow, might AVs initially be confined to weather-benign climates?

Emergency services vehicles will also need interacting with, as highlighted when a Tesla in California failed to respond to lights and sirens while being ‘driven’ using Autopilot, the police units involved having to resort to halting traffic from behind while forcing the car to bring itself to a stop17. Questions including how police will be able to pull over AVs and how such vehicles themselves will be programmed to respond to sirens, lights and other instructions are very valid and currently remain unanswered, again showing that driverless cars on everyday roads aren’t as close a reality as many may now believe.

Autonomous Uber Volvo

Potential lessons to learn from aviation

With the planes’ MCAS (Manoeuvring Characteristics Augmentation System) ‘anti-stall’ technology in the spotlight following two tragic Boeing 737 Max crashes, it would be sensible to argue that independent scrutiny and regulation will be absolutely vital when it comes to autonomous cars and commercial vehicles, which are historically much more dangerous than aircraft.

Proactivity needs demonstrating by all stakeholders, the U.S. taking a perhaps reactionary approach to regulation, compared to China and South Korea where legislation has already been introduced for the driverless segment of the automotive sector. Some commentators20 highlight that Congress has refused to pass AV legislation because of not trusting any of the parties involved, which is a lamentable position to be in. Others assert that oversight of AV development is worryingly patchy, sensors and other technology aren’t tested for reliability, and that the operators involved are largely able to act as they please through self-assessment21. We wholly agree that no corners should be cut along the road to autonomy, and that public engagement will play an important role in moving forward.

Self-driving cars making ethical decisions

Unless segregated lanes dedicated for autonomous vehicles are introduced, which would arguably put paid to the traditional perception of driving, from which many motorists derive a degree of pleasure, AVs will be required to make ethical judgements such as, in facile terms, whether to collide with a pedestrian, another vehicle or a cyclist, if one of these scenarios was simply unavoidable.

Q3 of 2018 saw a major report entitled The Moral Machine Experiment13 and co-written by moral psychologist Azim Shariff published in ‘Nature’, in part containing the results of an online game that went viral. In it, people were asked to choose whether a driverless car should, for instance, hit a homeless, elderly or overweight person in order to prioritise a younger person, athlete or executive. Such a notion could be considered as disturbing and inhumane, but it’s a reality that societies will have to face as AVs are introduced.

“If cars were always programmed to be slightly closer to the bicycle lane, they may slightly reduce the likelihood of hitting other cars, while slightly increasing the likelihood of hitting cyclists”, Shariff commented plausibly to Forbes.

A cyclist in Texas was killed in 2015 when his completely motionless stance at an intersection was misinterpreted by a Google driverless car, at least reassuringly resulting in the software’s algorithms being improved. In the UK, the Highway Code currently states that drivers should overtake cyclists with the same space they would another motor vehicle, although it’s fair to say that many motorists either knowingly or unwittingly fail to heed Rule 163 anyway. But it once again points to how legislation will need updating to account for a plethora of AV scenarios.

Debates, misconstrued comments and other views occasionally make the headlines in relation to whether some manufacturers will program their self-driving cars to prioritise the safety of their occupants ahead of other road users including the more vulnerable, although this does seem an unlikely move.

Warwick Business School along with Jaguar Land Rover has recently expressed concerns that AVs’ artificial intelligence capabilities are currently vastly lacking and will perhaps struggle to interact successfully with other road users and their wider environments. “The rate of progress on this challenge may prove a decisive limiting factor in the development of autonomous vehicles”, behavioural science professor Nick Chater told CIO15.

Thankfully, pioneers in the field are pouring considerable resources into ensuring that the cameras, LiDAR radar systems and GPS at the heart of many self-driving cars will be able to interpret humans’ gait, body symmetry, foot placement and other physical movements to generate 3D computer simulations on the fly in order to predict the behaviour of pedestrians more safely14.

An angle not typically considered

Writing for The Conversation6, Clint Witchalls states accurately that driverless cars “won’t get tired, drive drunk, look at their phone, or speed” and is right to question what will happen to connected autonomous vehicles (CAV) if a human steps in front of them, potentially bringing a city’s finely-balanced traffic network to a halt. This vision seems credible given that AVs will be developed with a safety-first approach.

His suggestion that self-driving vehicles may be programmed to occasionally run into humans in order to prevent complacency and maintain a level of caution may seem unthinkable to some, but its premise is understandable. There is indeed little point in AVs being introduced to optimise traffic flow only for people, animals, cyclists and other objects to cause congestion to return. Some voices in society are against cars in general and could well adapt and exploit AVs’ in-built caution in order to move around urban areas more easily on foot than they can today.

Andrew Ng11, well-known in the artificial intelligence (AI) field, has expressed the view that pedestrians need to reshape their thinking and behaviours to become less erratic, crossing roads only at designated places, for example. It’s fair to say in response that humans will always be unpredictable, though, making it impossible to rule out the possibility of someone crossing a road on a blind bend or in some other unsafe place. So-called ‘jaywalking’ is now illegal in the States but still happens. Psychology professor Gary Marcus, for instance, feels that Ng is “just redefining the goalposts to make the job easier” for AVs, and has added his support to the notion that autonomous vehicles can only become as successful as hoped it they are to be segregated from humans and other animate beings. We also envisage increased reliance on geo-fencing proving to be an integral contributor to cities successfully welcoming AVs.

Traditional OEMs, tech’ start-ups and other pioneering firms are perhaps rushing under their own self-created pressure to introduce new autonomous vehicles and capabilities, but it’s fair to say that there are still several sizeable hurdles to overcome before AVs equipped with remarkable machine learning and AI take to developed nations’ roads and successfully navigate the complex and unpredictable world around them.

 

Sources:

1. https://www.techradar.com/news/tesla-wants-a-fleet-of-self-driving-taxis-by-2020

2. https://www.theinquirer.net/inquirer/news/3070645/self-driving-cars-could-be-on-uk-roads-with-no-safety-driver-come-2021

3. https://www.siliconrepublic.com/machines/apple-nissan-leaf-autonomous-electric

3b. https://www.barrons.com/articles/u-k-is-most-prepared-to-support-driverless-cars-51555077352

4. https://www.autonews.com/finance-insurance/how-tesla-plans-cut-customers-insurance-costs-tap-autopilot

5. https://www.bloomberg.com/news/articles/2019-02-19/autonomous-vehicles-may-one-day-kill-car-insurance-as-we-know-it

6. https://theconversation.com/nothing-to-fear-how-humans-and-other-intelligent-animals-might-ruin-the-autonomous-vehicle-utopia-114504

7. https://www.motoringresearch.com/car-news/ford-is-using-lights-to-help-autonomous-cars-communicate-with-pedestrians/

8. https://thenewdaily.com.au/life/auto/2019/03/05/car-makers-create-electric-car-engine-sounds/

9. https://media.jaguar.com/news/2018/10/sound-jaguar-i-pace-protects-road-users

10. https://www.theregister.co.uk/2019/03/06/sound_of_electric_cars/

11. https://www.insurancejournal.com/news/national/2018/08/17/498414.htm

12. https://www.roadsbridges.com/node/25205

13. https://www.forbes.com/sites/carltonreid/2018/10/28/driverless-cars-will-kill-cyclists-warns-nature-magazine/

14. https://www.futurity.org/self-driving-cars-pedestrians-1982002/

15. https://www.cio.co.nz/article/660893/science-has-scarcely-addressed-how-autonomous-vehicles-interact-human-drivers-research/

16. https://phys.org/news/2019-04-autonomous-cars-anytime.html

17. https://www.bloomberg.com/news/features/2019-02-20/someday-your-self-driving-car-will-pull-over-for-police

18. https://www.oxfordmail.co.uk/news/17601401.oxfordshires-roads-used-in-virtual-simulator-to-test-driverless-cars/

19. https://warwick.ac.uk/newsandevents/pressreleases/new_27m_research/

20. https://qz.com/1595490/what-the-737-max-crisis-can-teach-the-autonomous-vehicles-industry/

21. https://wtop.com/national/2019/04/boeing-crashes-and-uber-collision-show-passenger-safety-relies-on-corporate-promises-not-regulators-tests/

22. https://techcrunch.com/2018/12/11/lyft-self-driving-car-communication-patent/

23. https://www.insurancebusinessmag.com/uk/news/auto-motor/insurers-v-manufacturers-whos-liable-for-autonomous-vehicle-collisions-107011.aspx

24. https://www.insurancejournal.com/magazines/mag-features/2018/12/03/510344.htm