Researchers blind autonomous cars by tricking LIDAR

If you've ever been dazzled by some idiot's high-beam driving towards you at night, you'd probably welcome a self-driving car – except one of the key “eyes”, LIDAR, can also be blinded, or tricked into reacting to objects that aren't there. LIDAR - Light Detection and Ranging - is an important self-driving vehicle technology: …

[notowenwilson]

I always wondered, if the majority of cars end up with LIDAR systems that run on the same wavelength, what are the chances of those LIDARs interacting in the same way the researchers found? Provided your vehicle density is low enough the chances of getting interference would be pretty low, but what about in peak hour traffic where you could have multiple cars blasting away with multiple LIDARs in a confined area?

[Field Commander A9]

The lasers will be coded, just like in military laser designators.

[Anonymous Coward]

But as the article itself notes, coding the beams means incoming beams need to be processed first: problematic when you have almost no time for that stuff.

[Mage]

Lidar is a dead end.

We need to solve self driving trains, ships, aircraft and trams (in about that order) FIRST.

We might need real AI and better positioning / maps.

We might need better passive vision analysis systems.

We need fully autonomous vehicles that work without remote data or GPS etc as tall buildings, tunnels, trees and interference are issues. Privacy too.

It's evident that existing self drive tech has too much hype and needs to be better tested in secure mock up road systems away from pedestrians, cyclists and others.

Lidar isn't reliable or scaleable.

[Orv]

Re: Lidar is a dead end.

Self-driving trains have existed for decades in subway systems and elevated rail systems; the basic tech to do it using track circuits started appearing around 1880.

The trick, as with cars, is to come up with a system that doesn't involve installing and maintaining equipment along 140,000 miles of existing rail line, much of which is marginally profitable at best. Railroads also put great value on having things fail-safe as much as possible and are pretty conservative about technology.

[Fred Flintstone]

Re: Lidar is a dead end.

The trick, as with cars, is to come up with a system that doesn't involve installing and maintaining equipment along 140,000 miles of existing rail line, much of which is marginally profitable at best. Railroads also put great value on having things fail-safe as much as possible and are pretty conservative about technology.

Hah! You want conservative? The solution was invented ages ago. I'd give it a few weeks before non-self driving cars get the message to stay out of the way, though.

As for the LIDAR problems, at least it gives idiots with laser pens another target :).

[cray74]

Re: Lidar is a dead end.

We need to solve self driving trains, ships, aircraft and trams (in about that order) FIRST.

Development of self-driving systems is not a single program run by a united humanity. Even if you brought 200-ish nations together in a harmonious union to first solve self-driving trains, such a program is not going to rob other industries (like the automotive and aerospace industries) of the personnel and funding needed to develop self-driving systems for their vehicular products.

Tesla and Ford build cars and their customers want self-driving cars, so they're going to develop self-driving cars without contributing to self-driving train, plane, ship, or tram programs.

[JeffyPoooh]

Processing Time for coded transmissions

"...you have almost no time for that stuff."

Processing coded reflections is done with gate arrays where the signal basically falls straight through. The "processing time" to add coding, done correctly can be next to nothing.

The people conceptualizing autonomous vehicle systems are just getting started (I mean they're naïve).

[Anonymous Coward]

> I always wondered, if the majority of cars end up with LIDAR systems that run on the same wavelength, what are the chances of those LIDARs interacting in the same way the researchers found?

If you had read the research paper, which is linked in the article, you would not have had to wonder.

[notowenwilson]

Awesome, as the designated Reg reader with enough time on their hands to read the paper, care to share the TL;DR version?

[Anonymous Coward]

> Awesome, as the designated Reg reader with enough time on their hands to read the paper,

As opposed to enough time to read and comment on random articles on El Reg, I see. The Twatter generation at its best.

[notowenwilson]

"As opposed to enough time to read and comment on random articles on El Reg, I see. The Twatter generation at its best."

I was torn between replying with

a) a quip about the dismissive older generation that has to hide behind the anonymous coward veil in order to sling insults with impunity at random strangers on the internet

b) noting that if I hadn't been reading random articles on El Reg I never would have seen a reference to the report and hence would never have read it anyway, and

c) pointing out that you can often get better information in the comments than you can in the article itself.

As is the case with my generation we frequently struggle to make decisions so you can have all three

[Anonymous Coward]

"...read the research paper..."

Been there, done that, sent in the corrections. Idiots.

Those people aren't necessarily any smarter than anyone else. As evidence of this "controversial" assertion, I present a Tesla with the roof torn off, and another one that crashed head-long into a parked car, etc., etc...

[as2003]

It's important to do research like this, but I feel they may be overstating the seriousness of this attack.

An autonomous car is going to rely on more than the reading from a single LIDAR; It's going to be combining readings from multiple ultrasound sensors, multiple optical cameras, radar, wheel position and speed sensors, etc, etc, etc.

This attack seems less of a threat to human safety than just shining a laser pen into the eyes of a more traditional meat-based driver.

[Charles 9]

So the havoc-wreaker will simply dazzle ALL the systems at once using multiple spoofing systems. Even better, giving each one a different false reading will produce a logic bomb no matter what the judgment system used. After all, what kind of system would be able to figure out they're really traveling over black ice when one system tells them they're going 100mph, another 15mph, a third still 25mph sideways, etc.? Every sensor different and every sensor WRONG, too?

[Anonymous Coward]

I could see a use for this attack in cargo hijacking...

Especially if you could feed differing signals/conditions across 2 or more types of sensors. Stop the truck, relieve it of some precious cargo, and get out before anyone can call in the hijacking.

Probably much less useful in doing something like a kidnapping or a highway robbery of other passengers on the roadway.

[Anonymous Coward]

So what would you do if you were blinded while driving?

You'd slow to a stop and pull over to the best of your ability, using your last memory of the road ahead. A self driving car would have a more accurate memory of the road ahead and know exactly the steering input to get it there, so it could very neatly pull over out of the way. It isn't going to "slam on the brakes" like people are suggesting here, that's ridiculous and counterproductive.

There are laws against blinding drivers and pilots, there would be similar laws against blinding self driving cars. Sure, criminals might use this to make a self driving armored car pull over so they could rob it, but they better blind ALL its sensors in every direction inside and out lest it record them committing their crime, they better jam its cellular lest it call the cops the instant it is blinded providing GPS coordinates, and they better disable its ability to move before they leave lest it follow them to help the cops track them.

[Doctor Syntax]

Re: So what would you do if you were blinded while driving?

"You'd slow to a stop and pull over to the best of your ability, using your last memory of the road ahead."

Not so easy if it's on a multilane road and is now blind to traffic between itself and the side of the road. If multiple vehicles are being attacked there may already be a stationary vehicle at the side of the road waiting to be hit from behind.

[CrazyOldCatMan]

Re: So what would you do if you were blinded while driving?

Not so easy if it's on a multilane road and is now blind to traffic between itself and the side of the road.

Indeed. Which would mean that each vehicle now needs to be generating an "I'm here" signal which is used in concert with the lidar.

Except that that system will then add an extra exploitable sensor - the Bad Guys^TM will just scatter emitters for that signal over the road and thoroughly confuse everything.

And tying all the vehicles together into a senser-net won't fix it either - it'll make it more difficult to spoof stuff but it still won't make it impossible - it'll just increase the number of cracked sensor controllers in the vehicle.

[Anonymous Coward]

@Doctor Syntax

The car remembers exactly where all other cars around it were and their speeds, and where any cars pulled over were, so unless it is very heavy traffic it should know enough to successfully pull over. If it can't safely do so it will slow to a stop where it is, and other cars will have those other lanes available to divert around it.

Assuming someone hasn't jammed all communication abilities with other cars, it would also be telling other cars about its plight and desire to pull over, and they'd make room. Yeah, yeah, I know, I know, this theoretical attacks blinds ALL sensors and disrupts ALL communications, right? If it does that then all the cars will be slowing to a stop where they are, and it won't be any different than what happens today when a truck jackknifes and blocks all lanes of traffic - except that it will happen a lot less often.

[Alan Brown]

Re: So what would you do if you were blinded while driving?

"There are laws against blinding drivers and pilots, there would be similar laws against blinding self driving cars."

New Zealand has a nice catch-all charge of "endangering transport" which covers that (and messing up trains, etc) with a penalty up to 15 years imprisonment.

The UK could do with it. Apart from the issues above (I've been lased whilst driving and it bloody HURT), it'd be a very real dissuader to the problem of copper theft on the Railways.

[Dazed and Confused]

Re: So what would you do if you were blinded while driving?

You'd slow to a stop and pull over to the best of your ability, using your last memory of the road ahead. A self driving car would have a more accurate memory of the road ahead and know exactly the steering input to get it there, so it could very neatly pull over out of the way.

A few years ago I was driving on the motorway when the bonnet (hood) of my car blew open. Apart from the fact it hit me across the top of the head through the roof, I obviously lost all forward vision. But I quickly realised that I could see out of the side window and could use this to ensure that I was steering parallel to the white lines and armco, secondly my wife immediately realised that because of the way the bonnet had bent she had a better view than me if she lent right up against the side of the car. Together this allowed me to slow right down and then look over my shoulder and see when it was safe to pull across the lanes to the hard shoulder.

Self driving cars will need to be designed to cope with sensor failures and to make decision based on anything that is left to them. Lets hope they do this better than most bit of SW we experience in everyday life.

[I3N]

Practice ...

ExoMars Schiaparelli Mars lander ...

[Phil O'Sophical]

An autonomous car is going to rely on more than the reading from a single LIDAR; It's going to be combining readings from multiple ultrasound sensors, multiple optical cameras, radar, wheel position and speed sensors, etc, etc, etc.

Set it up your attack so that the camera sees a holographic child standing in the road, the LIDAR senses a stopped lorry, the ultrasound detects a sinkhole and the radar says there's nothing there. What happens? The vehicle falls back to wetware mode: "Oy, driver, wakey wakey, I'm confused. All stop.". M25 grinds to a standstill.

[Anonymous Coward]

Even with the basic systems on cars today a similar attack can be carried out and for less money. Some already have been for instance:

Drop a concrete block off a highway/motorway bridge

Drop a load of nails off a highway/motorway bridge

Drop yourself off a highway/motorway bridge

Pierce the brake lines

Blind the driver with a high power torch or laser

Slam the brakes on in your car in fast moving traffic

Drive a motorbike past a car and kick it

Move a load of traffic cones in a construction site to direct the cars towards a large hole

Roll a large tyre down a hill that intersects a main road

Etc etc

[Anonymous Coward]

Motorbike

"Drive a motorbike past a car and kick it"

The chain of events seems more to be:

- ?? something happens that causes motorcyclist to get angry

- motorcyclist kicks car

- car violently swerves, attempting to sideswipe bike

- bike brakes, car misses bike

- car driver over corrects swerves all over the road

- car hits truck, flipping it

- car hits edge of road

I'm pretty sure a self driving car would probably not attempt to run other vehicles off the road, and if it did (for the authentic Audi experience) it would be less likely to fuck up the maneuver :)

[cbars]

Re: Motorbike

That video was insane.

Looks like the car made an illegal move (hence why the filming started) - in the US you can't cross into the car pool lane. The biker kicks the car (after it comes into his position via his blind spot), which isn't legal either, but does not warrant the attempt on their life!

No one comes out well, but I think it's the same as someone pulling out in front of you on a roundabout and you beeping at them - except you're not in a protective shell and very much more likely to die. And instead of having a dashboard/steering wheel to hit, the perpetrator is within reach....

The real issue is the completely innocent driver in the white pickup. Anonymous coward, you troll.

[tedleaf]

Re: Motorbike

Ahh good old American driving behaviour and standards.

I liked the way the bike just pulls away up the road leaving the carnage behind him..

[Androgynous Cow Herd]

Re: Motorbike

A friend of mine, years ago, went to court for manslaughter in LA after a similar thing. Car made an unsafe move and darn near ran him and his wife off the road. My friend (A fireplug of a guy and a bonafide "Motorcycle enthusiast" of the heavily tattooed and belligerent variety) chased the car down on the freeway and got into it with the driver.

The end result can be read here: http://articles.latimes.com/1992-03-14/local/me-3342_1_mark-gold

Marty was acquitted of all charges but left with a pretty large lawyer bill.

[Elmer Phud]

Drive-by booting

As one who has managed this while negotiating London's traffic on a bicycle, it takes a lot of internal processing to judge the correct lean angle that counters the reaction from the kick (best aimed at thin panels).

A motorcycling mate has a denim jacket with Kevlar inserts and stiffening - he just clotheslines the mirrors instead.

[tedleaf]

Re: Drive-by booting

It works much better with proper real para boots on,you dint hurt toes like you do in sloppy bike shoes and by God does it help build yer leg muscles up having 3 pounds of leather on the ends of yer legs.

+ they give a much more satisfying and effective result when planted in some red listed motorists crutch or give you some real heft to kicking drivers door shut on their leg as they try to get out when they think their going to win a fist fight,me I think it's fair tactics,I'm on a 15 LB bike,their in a two tin car,big boots even things up a bit...

If you realy want legs like Chris hoy had,wear para boots...

[Alan Brown]

Re: Drive-by booting

" it takes a lot of internal processing to judge the correct lean angle that counters the reaction from the kick "

Follow the dutch example - an openhanded slap on the roof will have the driver veering away across 2 lanes of traffic from the noise.

Yes, the cyclists there have had a lot of practice.

[Anonymous Coward]

> Even with the basic systems on cars today a similar attack can be carried out and for less money. Some already have been for instance:

> Drop a concrete block off a highway/motorway bridge

Years ago (possibly 30+ years ago), I remember reading an SF story that involved self-driving vehicles on highways and bored people tossing Christmas trees on to the road and then watching traffic grind to a halt because they couldn't cope with the obstruction, especially if the trucks were driverless.

[Doctor Syntax]

"M25 grinds to a standstill."

In other words it'll be completely ineffective.

[Kernel]

"The lasers will be coded, just like in military laser designators."

Yes, the obvious answer here is secure encryption of the LiDAR and other sensor's signals - oh, bugger!

Never mind, only law enforcement will have access to the backdoors, so it shouldn't be too much of an issue.

[Anonymous Coward]

No encryption would be used, it would simply pick a random number at startup and encode that in the beam's modulation, so that it could tell which beam "belong to it" when reflected back.

[annodomini2]

The problem is noise, you have a moving object attempting to scan another potentially moving object.

The potentially moving object may be flat or may be of complex shape with variant reflective and refractive indexes.

Then you need to send a sequence of pulses, over a time period that your detector can observe.

Many-many-many non malicious things could disrupt these signals. So where do you draw the line to accept or refuse a signal?

There are so many edge and corner cases it's unreal.

[CrazyOldCatMan]

it would simply pick a random number at startup and encode that in the beam's modulation

And, depending on the entropy pool used, how do you stop two vehicles setting the same number? And, if it's encoded into the beam, how do you stop the bad guys capturing and replaying a static encoding?

Answer to both - you can't with any confidence.

[Charles 9]

You can beat a replay attack with a rolling code, which will also reduce the odds of a collision. But it's still not going to do much good in a twitch situation (say a tree suddenly falls across the road within the braking distance).

[Alan Brown]

"say a tree suddenly falls across the road within the braking distance"

That's a driving test fail right there.

Trees don't suddenly appear on the road, they topple - and if you're watching for hazards as you should be, you'll see it long before it lands.

Otherwise you'll be the one telling the coppers "that kid just appeared out of nowhere" as the ambulance drives slowly off with its lights and sirens OFF.

[Charles 9]

"Otherwise you'll be the one telling the coppers "that kid just appeared out of nowhere" as the ambulance drives slowly off with its lights and sirens OFF."

Dude, Not Funny!

That's more real than you realize. Little kids can be smaller than the cars they hide behind and completely concealed by their bodies. They don't know what's out there, then suddenly jump A MERE THREE FEET in front of your car.

Physically unavoidable, and an emotional train wreck to boot. You CANNOT tell the grieving parents that you can't fight physics.

And PS. Parts of tree can drop pretty suddenly and without warning. Maybe not the while trunk, but think a very large and heavy overhead branch that picks that picks the wrong moment to drop fifteen feet in front of you on a crowded road (so no room to swerve). Yes, they CAN just drop. Frequent auto insurance claim, in fact.

[Anonymous Coward]

Interesting research

It's just a shame that it had to be picked up by a non-technical journo who decides to write it up in full sensationalist mode.

For the record, the interesting bit is that it has been experimentally proven, albeit on a LIDAR system, not in a full self-driving car system, which uses data from a multiplicity of sources and has thousands of tests that check the functioning of every sensor.

The possibility of this type of "attack" (or more likely, failure from environmental effects) has been well-known by self-driving car developers for years.

It is worth mentioning, btw, that current production systems use optic and/or radar "vision", with LIDAR being very much an emergent technology in this application. At least one manufacturer (Tesla) currently has no plans to use LIDAR in their cars.

[Doctor Syntax]

Re: Interesting research

"t is worth mentioning, btw, that current production systems use optic and/or radar "vision", with LIDAR being very much an emergent technology in this application. At least one manufacturer (Tesla) currently has no plans to use LIDAR in their cars."

And radar is surely going to be subject to the same attacks. The "dar" in their names is a clue - it means "direction and ranging" in both cases. Optical sensors are also going to be subject to dazzling, at least in low light levels.

"At least one manufacturer (Tesla) currently has no plans to use LIDAR in their cars."

And Tesla has already been shown to miss a large object in front of it under adverse seeing conditions.

[Amos1]

Re: Interesting research

Yes, but one not only outside not only its design parameters (since corrected) but where the driver put way too much trust in his own limited experiences (incorrect sample size for the task). The one smart thing Tesla did was to grab (and apparently stream in real-time) the telemetry so the facts could not be seriously disputed. That's the data you also need to tamper with reliably to cover up the crime.

[Doctor Syntax]

Re: Interesting research

"outside ... its design parameters (since corrected)"

If I were to trust my life to an automatous vehicle (and with trials permitted on public roads I might have no choice) I'd want the design parameters to cover what actually can happen on the roads, however unlikely. A vehicle from one carriage way turning across another at a road junction doesn't sound like something that ought to have been outside design parameters in the first place. Correcting design parameters after obvious omissions have come up against reality isn't the best way to proceed. And just wait until one of these ventures down a Devon lane with passing places.

[Nick Ryan]

Re: Interesting research

And Tesla has already been shown to miss a large object in front of it under adverse seeing conditions.

This unfortunate accident was caused by two things:

Firstly the driver not understanding that the software wasn't really an autopilot and more of a driver assist. This is Tesla's marketing department's fault, since rectified, but from what I understand this particular driver should have been well aware of this as he was close to Tesla and a keen advocate.

Secondly US trucks do not typically have guards down the side presenting an open space that often cars can fit into too well - there are many many reported instances of human controlled cars falling foul of trucks as a result. There's a reason that such guards are mandatory in most of the rest of the world. As a result the car, quite correctly, saw an open space in the road because there was one - the fact that it wouldn't fit entirely in the vertical space is an issue, but cars are not expected to be driven at speed at exceedingly low bridges and for practical reasons the height of the scan (lidar/radar,etc) is limited, although hopefully enlarged now.

[tedleaf]

Re: Interesting research

Yes and teslas system has ready been shown to be flawed,not seeing a small object or car I can almost understand,but the object it missed was 60ft,8 ft wide and 13 ft tall,that's a pretty big object to miss because it was the "wrong colour" and not precisely where tesla thought it would be..

Mind you ,if that trailer had side bars,then it might have just seen those instead as their usually dark or dirty..

[Voland's right hand]

By illuminating the LIDAR with a strong light of the same wavelength as that the LIDAR uses, we can actually erase the existing objects in the sensed output of the LIDAR.”

So how does this work reliably when every car has one and they are all using sensors at the same wavelength. Even if different manufacturers use different ones, ending up in a stretch with all others using the same wavelength as you is simply a matter of probability.

[Anonymous Coward]

They will probably use different frequencies that are in the group of light frequencies used to define laser light, just like different radio channels exist on the FM radio frequencies. They might also be able to alter the wavelength in a way that the receiver knows what light frequency to expect back, a bit like shining blue light then red light then green light and expecting reflections back in that order.

Still I wonder what will happen if the sensors are swamped with light across all the frequencies in effect making them blind? Whilst it might not be a problem on say a motorway with lots of vehicles in convoy communicating between themselves, think F35 swarm sensors relayed back to the different pilots helmets for a single unified battlefield visible to all, but a single vehicle on a road with little regular traffic could be tricked into stopping or reacting adversely.

[John Smith 19]

Creating mirages for fun and profit.

Who knew?

I'll note there are ways to encode the pulse patterns produced by an active laser to be unique. They are called Gold codes and are uses in the GPS system.

However you'll need a pretty good RNG to ensure that in dense traffic yours is unique to your vehicle.

[Nick Ryan]

Re: Creating mirages for fun and profit.

The coding doesn't have to be universally and persistently unique because the coding can cycle and change as fast as required. The range is limited, as in the number of sources that are reasonable, or even excessibly, expected to be in front of the vehicle therefore a changing coding will pretty much ensure accurate responses unless the RNG is exceptionally poor.

Of course, all this doesn't matter when it comes to sensor flooding as not a lot can help this unless different techniques or frequencies are used as well or instead.

[jmarked]

More development needed for this technology.

[Anonymous Coward]

Nothing new here...

That hasn't been seen in military radar target tracking for decades. I love the way "old" technology (in a different form) is always seen as "new" within the automotive sector and they very, very rarely draw on decades of experience within other sectors.

[Alister]

Re: Nothing new here...

I love the way "old" technology (in a different form) is always seen as "new" within the automotive sector

So... reinventing the wheel, then?

[Baldrickk]

Re: Nothing new here...

Isn't that their job? They sell more by making it 'new'

[Alan Brown]

Re: Nothing new here...

" they very, very rarely draw on decades of experience within other sectors."

Not just the automotive industry.

When I sat down and explained PGP's dual-public/private key system to an elderly customer in the 1990s, his response was "That's almost exactly like the systems I was working on in the 1950s" - and yes, he was working in an outfit that officially didn't exist at the time.

[Orv]

Re: Nothing new here...

Unfortunately, when it comes to things like radar and lidar anti-jamming tech, the state of the art is often stuff civilians aren't allowed to access, and therefore must reinvent on their own.

[Anonymous Coward]

One day the goverment will update this crap

"As per the data from UK Department for Transport, 55m is the braking distance for a car driving at 60mph."

This was relevant when we were driving Austin Allegros and Morris Minors, it's like those 50 MPH bends on motorways*. It's time they were updated.

https://www.youtube.com/watch?v=lWmEbbPlQ_c

* I exclude many US cars as they still use 1940's suspension, drive trains and weigh about 15 tonnes.

[CrazyOldCatMan]

Re: One day the goverment will update this crap

This was relevant when we were driving Austin Allegros and Morris Minors, it's like those 50 MPH bends on motorways*.

Some of us (sometimes.. when I'm granted the keys by my wife since it's her car) still drives Morris Minors.

Admittedly, it's got disc brake upgrades[1] and a brake servo fitted[2], but it's still 95% Morris Minor.

[1] At my insistance when I discovered that she was happily pumping the brakes first thing in the morning to get them to actually work. Discovered when I drove it and my foot went almost to the floor when I first put the brakes on. "Oh yes" she said, "didn't I tell you? You need to pump the brakes a few times until they work.". I pointed out that this almost certainly invalidated her insurance. )..

[2] And front+rear heated screen, a fog lamp, an alternator instead of a dynamo and -tive earth electrics

[tedleaf]

Re: One day the goverment will update this crap

12 volt electrics I hope,ahh,the dim yellow glow from 6volt headlights,if they fitted them to cars today,that would rapidly thin out the numbers of joy riders,bad drivers and old farts who cannot see without the use of 500 watts of lamps to light things up half a mile ahead so their awful reaction times have some kind of chance..

[Alister]

Re: One day the goverment will update this crap

This was relevant when we were driving Austin Allegros and Morris Minors, it's like those 50 MPH bends on motorways*. It's time they were updated.

55metres at 60mph is the distance calculated to be safe for the broad range of vehicles on the road.

Yes, you could probably reduce it for modern cars with vented disks and low-profile tyres, in perfect mechanical order, but there's still an awful lot of vehicles on the roads which will not be able to stop in that distance, so a safe margin based on an average is better.

I would be prepared to bet that well over 50% of private cars on Britain's roads at this moment have one or more defects which will have a detrimental effect on their stopping distance, whether that be worn tyres, under-inflated tyres, worn or loose suspension bushes, faulty shock-absorbers, worn brake pads, etc, etc.

Very few car owners today do any maintenance on their cars, and most only fix things when they fail the MOT.

[tedleaf]

Re: One day the goverment will update this crap

And all of that relies on them first actually seeing something and then reacting..

Me,I've tried stopping from 60mph in 55 metres,it's a myth,you might just do it,with prior warning on a perfect surface in a light vehicle with very soft grippy wide tyres.

Me I would put the chances of most drivers/cars stopping before they hit the object at about 5%.

[Orv]

Re: One day the goverment will update this crap

It's also been shown that most drivers don't use the car's full braking ability, either, even in a panic stop. That's why anti-collision systems that "help" you a bit on the brake pedal are becoming a thing.

[Orv]

Re: One day the goverment will update this crap

This was relevant when we were driving Austin Allegros and Morris Minors, it's like those 50 MPH bends on motorways*. It's time they were updated.

In the US, they've actually been revising those numbers upwards in the last several years, to bring them more into line with the speeds drivers actually take curves at in modern cars.

I exclude many US cars as they still use 1940's suspension, drive trains and weigh about 15 tonnes.

That was true up through the 1980s, but no longer. At some point people started to expect tighter body control, and the old wallowing leaf spring setups disappeared. The general switch to front wheel drive also helped; while it didn't do weight bias any favors, it's a lot easier to make a good rear suspension if you don't have to deal with the unsprung weight of a heavy axle and differential housing.

None of this applies to trucks, which are still built the old way for Reasons. But the curve warning signs aren't meant to apply to heavy trucks. Truck drivers have to apply their own judgement. Partly this is because it depends a lot on the load being carried -- a loaded semi will roll over before its tires lose grip.

[pchurchi]

Seems to me that this is about as likely as somebody shining a laser at the driver of a car today. Possible, plausible, but certainly not common...

[Charles 9]

They're shooting dazzlers at pilots now, why not drivers?

[Elmer Phud]

It's Party Time!!!

The Spoofing method could be much cheaper by use of readily available 'glitter cannons' it seems to me.

(Not, of course, that I would in any way advocate such actions being undertaken.)

[tedleaf]

Er ?

Where did they get that 55 metres at 60 mph from and what is it meant stand for ?

Is that meant to be a reaction time or a stopping distance ?

There are very few cars/drivers that can stop from 60mph in 55metres,possibly with prior warning,on a very good surface and huge tyres on light vehicle,but most cars/drivers would fail if attempting it.

Before I get slagged down,if you have the confidence,try it yourself,find a short private road,stack some soft building blocks across it,measure and Mark a line 55 metres from yer stack,then accelerate to 60moh and see what happens.

At 60mph,most folks reaction times are so bad that most would hit an object at 55 metres before they even got much movement in their foot.

[John Brown (no body)]

Re: Er ?

"Where did they get that 55 metres at 60 mph from and what is it meant stand for ?"

According to the article, they specifically mention that as the stopping distance ONLY, ie EXCLUDING thinking and reaction time, probably based on the assumption that for all intents and purposes, a AI car will react "instantly".

[tedleaf]

Re: Er ?

Cheers me dear,my ropey old phone and awful 3G signal wouldn't let me get to article.

[Swarthy]

Re: Er ?

From TFA: "the braking distance is the distance required solely for braking". 55m is the "standard" distance to decelerate from 60mph to 0mph. It does not account for reaction times; the 55m is after the time seeing, processing, and reacting.

[ChrisC]

Re: Er ?

You asked: "Where did they get that 55 metres at 60 mph from and what is it meant stand for ?"

Article said: "As per the data from UK Department for Transport, 55m is the braking distance for a car driving at 60mph."

You asked: "Is that meant to be a reaction time or a stopping distance ?"

Article said: "Because the braking distance is the distance required solely for braking,"

See also the braking distance chart at https://assets.publishing.service.gov.uk/media/559afb11ed915d1595000017/the-highway-code-typical-stopping-distances.pdf

[tedleaf]

Re: Er ?

So it's the distance from apication of brakes point on some imaginary perfect surface.

Anyone want to volunteer to walk out on a motorway 55 metres in front of a car and see what happens in the real world ?

[Charles 9]

Re: Er ?

The point is, ANY kind of processing (say to winnow out false signals) is going to add reaction time, and in this case, added reaction time adds stopping distance. Now, according to my calculator, 60mph = 26.8224m/s. So a second lag time means you basically need half again as much distance. Even a realistic tenth of a second delay raises your minimum stopping distance by nearly 3 meters (about 9 feet). That's why the end of the article noted concerns around processing time.

[Orv]

Re: Er ?

Yup. And if the vehicle is a truck with air brakes, you can add another half-second for the brakes to apply.

[Alan Brown]

Single point of failure

Literally.

Now prove the same attack against a vehicle with multiple lidars fitted.

[Orv]

In their next paper, they will paint a tunnel on the side of a cliff.

[Herby]

Wile E. Coyote Lives.

Got to have that tunnel painted on the side of a cliff!

[trollmonster666]

And what about deer? Their eyes are very reflective.

[Orv]

Stopping for a deer in the road is correct behavior, so I don't think that's a problem. Also, they must be able to account for that or every roadside reflector post would dazzle them.

[Calimero]

That is called research?!

30 years ago it would be called testing - and a good tester would know where the system breaks. Later on it would be called V&V (referring to the SW used to analyze data and make decisions).

Driving autonomously is an AI-complete problem and only after it will be accepted as [nearly] impossible, a [close to bullet-proof] solution will "emerge", and will called the rail road. And it will be that, too.