Are semi-autonomous systems making cars safer?

Assisted driving tech should boost road safety, but it can actually make things worse if it isn’t tuned properly
Matt Saunders Autocar
18 May 2019

The question I’m asked the most is also the most simple one,” says Matthew Avery, director of insurance research at Thatcham Research. “‘What is the safest car on the road?’ I’m beginning to think the answer is a white Ford Fiesta,” he says, nodding at a pretty convincing mock-up of the very same car parked on the runway ahead of us. 

The ‘Fiesta’ in question is made of flexible, detachable plastic panels quite loosely fixed onto a moving base that looks like an oversized speed bump. It is, in fact, a robotised mobile target, with wheels hidden away underneath it and a top speed of around 15mph. And Thatcham has been using it to design, develop and prove a new batch of tests for the latest active safety and crash mitigation and avoidance systems fitted to new cars. 

“These systems use stereo cameras, radar sensors and sophisticated image processing software to recognise threats before responding to them,” says Avery. “People might think we could simply drive at a pile of empty cardboard boxes to test an AEB [autonomous emergency braking] system on a car, but they’re not so easily fooled. 

“The industry’s software engineers tell us that we have to use realistic targets in order to trigger the systems properly. So I wouldn’t be at all surprised to find out that those systems end up being particularly good at recognising white Ford Fiestas,” he adds, joking, “because that’s the kind of car our target happens to look like.” 

Our Verdict

Ford Fiesta

Ford Fiesta

In remaking Britain’s best-selling car, Ford has trodden lightly with the new Fiesta. But does the all-new supermini do enough to keep its place at the head of the table?

Find an Autocar car review

Driven this week

We’re with Avery and his team of research engineers to get a taste of exactly what kind of tests of these assisted driving technologies Thatcham has been devising, because they’re due to become a parallel part of Euro NCAP’s new car safety testing regime later this year. 

Founded in 1969 by the insurance industry, Thatcham is now a signatory member of Euro NCAP. “Twenty per cent of the total Euro NCAP safety score that a new car gets today is defined by the effectiveness of its driver assistance systems,” says Avery, “and you already get a 10% discount on your insurance if you choose a car with AEB.” 

The industry picture we’re looking at now, as Avery explains it, is one in which almost every major car manufacturer is fitting what we call ‘SAE level two’ driver assistance systems to their cars: lane keeping systems that will work to prevent you from changing lanes into the path of another car, for example, or adaptive cruise control systems that not only recognise the current speed limit but can also automatically adopt it. 

“But they’re all very different,” says Avery, “so there’s a real need to assess the effectiveness of them in a strictly objective sense [for which Thatcham has come up with meticulously repeatable tests done by robots] but also how sensitively they’re tuned, how well they’re integrated into the driving experience and how usable they are.” 

We’re about to get a firsthand idea. Having earlier run through an S-bend marked on Thatcham’s proving ground runway as if on a particularly windy dual carriageway to show how it tests lane keeping systems, we’re now motoring towards our plucky fake Fiesta at 50mph in a Volvo V60 as if we’re about to undertake it on the motorway. The Ford pulls into our lane at the last minute, as part of what Thatcham calls a ‘cut-in’ test – something most of the models in its first fleet of test cars apparently struggle to negotiate satisfactorily. Sure enough, the V60’s AEB system fails to detect the threat and the Volvo thumps through the target’s deformable plastic panels and rips them clean off. 

“The best cars we’ve tested have balanced driver support systems,” says Avery. “They don’t feel like they’re driving themselves, keeping the driver fully engaged; they don’t break in and out abruptly or ‘throw control over the fence’, as we refer to it – but they do provide a dependable, robust amount of assistance. Not too much and not too little.” 

Our man is very clear, too, on the need for that kind of system tuning in the most relevant technological context in which the car industry now finds itself: the sweep towards fully autonomous driving. 

“Our research suggests that there is already a sense among today’s drivers that their cars are ready to drive themselves – but, right now, they’re anything but,” he says. “If assisted driving technologies encourage drivers to disengage at the wheel – and one or two of them already are – we could see road safety statistics suddenly get a lot worse when you would reasonably expect them to be doing the opposite. 

“And so, for safety reasons if nothing else, we need to stop thinking of autonomous driving technology as if it’s already fitted to the cars we’re buying. I’d be in favour of changing our terminology: throwing out the SAE’s five-level classification for autonomous cars and instead putting clear blue water between the ‘assisted driving’ technologies of today and the properly automated systems we’ll only begin seeing in 2021.” 

There is no safe halfway-house solution for the self-driving car, as Avery sees it. We either have full automation, when the technology, road network and drivers are ready – or we wait until they are. And until then, we plainly need to keep drivers informed and clear about their role at the wheel.

Read more

Calling cars ‘autonomous’ is dangerous, say industry experts​

Some automatic braking systems have proven to be ‘ineffective’​

Is the public ready to share the roads with self-driving cars?​

Join the debate

Comments
12

Bazzer

18 May 2019

What, you couldn't have added a video to this article?!?!?  Far be it from me, an engineer, to advise you on journalism, but this sort of article (given its very subject matter) demands a video!  Seriously, just what is happening down there at Autocar?

Peter Cavellini

18 May 2019

 To make cars truly accident proof there’s one thing needed taken out of the equation......the human driving it!, ok, tech isn’t totally AI it has no empathy, only uses what it brain has been programmed to do, this relentless march towards just getting in a car and telling it where you want to go isn’t going to be a stress free experience not unless there are no windows to see out of, and, as any animal, you won’t totally trust it, there’d have to be an over ride if you saw something the car is ignoring, I don’t think full autonomy is going to arrive in a short time....

eseaton

18 May 2019
And how do you do that, Peter?

Cars are made by humans.

They have to interact with humans (pedestrians).

So you can't.

Peter Cavellini

18 May 2019
eseaton wrote:

And how do you do that, Peter? Cars are made by humans. They have to interact with humans (pedestrians). So you can't.

But, Cars only do what humans want them to do at the moment, when a full autonomy car is faced with a situation it can only do what it’s been programmed to do, we the human can change our mind at an instant, sometimes the programmed response isn’t appropriate but the autonomous car will do it because it can’t change its response halfway through, that’s why I think the full autonomy thing is further off than we think.

smokescreen38

18 May 2019
Should try some of those systems on the hgv's. They are useless, some I've used see a motorway bridge & think it's a stationery vehicle & all the alarms going off (Daf Xf). The version on the Mercedes Actros is even worse that doesn't see the vehicle until you've actually stopped no matter the range settings. Volvo's system is far better but it still throws a wobbler with the chevron signs on bends. I think true auto drive systems are a few years off yet.

inkpen

18 May 2019

Yes, much more needs to be done on the semi-autonomous features of cars, but nothing will ever completely eliminate car accidents, especially where humans are involved.

Presumably, the theoretically human-driven Fiesta in the didn't look properly before pulling straight into the Volvo, something that shouldn't have happened if the Fiesta was driving semi-autonomously.

Presumably, the human driving the semi-autonomous Volvo didn't have time to react to the sudden incursion of the Fiesta into its path. The Volvo's systems should ideally have reacted within microseconds to the predicted path of the Fiesta, but equally, it can't do an emergency stop every time someone drifts around in their lane.

The reactions of the Volvo's autonomous systems can be tweaked to best deal in future with this illustrated incident and a software upgrade rolled out to thousands (or millions) of semi-autonomous cars, something you can't do with humans!

If a semi-autonomous Fiesta made the same mistake of pulling out without checking, again the accident can be analysed & a software fix applied to millions of Fiestas worldwide - not possible with human Fiesta drivers.

Newer Teslas already have "black box" style recording from 8 cameras, 12 ultrasonic sensors & forward radar, so each accident can be analysed, learned from and all of the cars made ever safer. I imagine that all will follow where Tesla lead, until full autonomy becomes a remarkably safe reality.

The Apprentice

18 May 2019

I wonder how long it will before before these systems are widespread enough to have wider data? Such as a car performing an emergency stop getting rear ended because the system over reacted to a situation, or failed to stop the vehicle in time when an alert driver could have taken evasive action (mirror/lane change).

I recently tested a new CR-V which has the strongest driver aids I have experienced, I was surprised at how quickly it became comfortable to slip the adaptive cruise and lane keep on regularly, the systems were decisive enough to quickly let me develop a pretty strong trust in them, not sure if that's a good or bad thing.. for example they will only encourage the phone addicts to spend more time messing with them instead of concentrating on driving.

voyager12

18 May 2019

Waymo (bought 62,000 Chryslers Pacifica) and Uber (Volvo XC90) ought to know by now, that 1. the more road space a car takes up, the less margin to maneuver, 2. the more boxy thecar, the more difficult it is to piece the fragmented scanning, sensoring and imaging together, 3. the bigger the car, the more it tends to isolate the driver, who is supposed to function as 'backup' for the autonomous technology, from what's happening around the car. EPIC FAILS.

russ13b

18 May 2019

to stop cars crashing is to not have them moving in the first place. Normal ones crash due to driver error, autonomy will crash due to fault sensor readings, software bugs, and the software/hardware crashing in exactly the way that a computer can do. Look at the problems boeing have recently had, if they can have problems like that care companies can and will too. We still have problems making cars without needing to recall them, this is only going to make things worse. i'm not against progress or technology, but this is a big increase in cost, weight, complexity, and maintenence. Aren't we supposed to be trying to save the planet? Shouldn't there be a push for lightweight efficiency, and simplicity?

x218

18 May 2019

Of course cars are safer WITH autonomous braking. Do they work in evey situation ? no.  Is that a negative when it saves thousands from crashes? hell no. 

 

My E class brought me to a complete stop 1 year ago, avoiding a crash. It saved my life.  The question we should be asking ourselves is " How good are these systems , not IF they are good "

Pages

Add your comment

Log in or register to post comments

Find an Autocar car review

Driven this week