Volvo accidentally smashes new car in safety demo | Hacker News Reader

Volvo accidentally smashes new car in safety demo

wired.co.uk

42 pointsshrikant42 comments

Mercedes had a crash during a demonstration of Distronic a few years back as well: http://www.mbworld.org/forums/w211-amg/128286-distronic-demo...

I interned at Mercedes's advanced vehicle group in college. The culture was very safety conscious and the bias in developing driver aids was "Minimize the chance that your system will cause or increase the severity of collisions as compared to if your system wasn't there."

Which nicely shows what is pretty much a fundamental rule of thumb - adding complexity to a system has to be carefully managed to ensure that the end result is going to perform better on average than the simpler version would have done.

I'm reminded somehow of caching overly aggressively and the resultant code spending more time in the cache logic than it saves through not having to do the calculations.

While a non-programmer might think "geez, it should have stopped, it was going to hit that truck!", I can only think how complex the system must be to detect an oncoming collision and brake only when necessary. I'm sure they could tune it to be more sensitive to oncoming objects, but I'd rather defer to the drive in those circumstances..

The story I've read was that it wasn't a problem with their the collision code as such, but that the car battery died completely shortly before the demo and the computer systems didn't start up right when the battery got recharged.

Draining the battery could be argued as "human error", but it is a pretty common error.

There is no "human error" in the "computer systems didn't start up right" half of the excuse. That is a design error.

Having said that, I've worked with avionics. Avionics power systems are absolutely horrible with respect to voltage regulation and glitches. The power source is switched between a ground cart, APU, and multiple engine generators at various times and causes glitches each time. Creating hardware and software that is able to start reliably when hit by multiple glitches with random intervals and random duration plus power brownouts and surges is extremely difficult. We used a programmable power supply that was able to do random glitches (it was affectionately known as the "widow maker" because the engineers' wives would become a virtual widows while the engineers worked out the bugs ;-). When the systems engineers would assure us we would have clean power, guaranteed, we would just roll our eyes and roll over the widow maker... (The battery bus actually is reliable and glitchless, but only flight critical hardware gets to hook up to that.)

Just to complicate life, power glitch testing (and brake system safety testing) is an instance of the halting problem. http://en.wikipedia.org/wiki/Halting_problem

IMO, there are more fundamental problems.

Run any computer system long enough and you will get random bit errors in both memory and computation. Avionics systems mitigate this though redundant systems but cars are not built to that level of safety. So yes, assisted braking systems will fail. The real question is how often and to what result.

I am led to understand that ABS systems often do have multiple independent microcontrollers.

I know of very few cars with multiple rotation speed sensors on the wheels (and the system has to know the rotation speed of each wheel if the system is to function properly). So each sensor (and all the associated cabling, wire harnesses, etc) is a single point of failure.

ABS systems are far from fail-proof...

Not sure how the glitch testing is an instance of the halting problem, could you expand on that a bit?

Testing by experimentation (random glitch spacing and random glitch duration) is inherently an instance of the halting problem. If you don't have any failures, you know all the combinations you have tested work, but you can never be sure there isn't a combination still out there that you didn't hit that would cause a failure.

Actually, watching this is really good, it shows you how much energy is present in a relatively low speed collision.

One down for the demonstration effect.

There's no shortage of crash test videos out there, though. It is, however, a good demonstration of the safety of a modern Volvo that there was no damage beyond the A pillar.

Compare to this rather disturbing crash test of a Chinese car:

http://www.youtube.com/watch?v=PHZqcKj7jNM

That hurts.

Slight difference though, this is an 'offset' crash, the volvo is straight, makes you wonder how well the V60 would have fared in an offset crash.

Wonder no more:

http://www.youtube.com/watch?v=qSYxj1wFULM

Actually that's the last model. These are done at 40mph so it's more severe than the truck crash above, but there's still no damage to the passenger compartment.

Here's the new model; I couldn't find an offset crash test but thought I'd post this anyway because it's fascinating -- it shows all sorts of crazy angles you don't get to see, like a view from below:

http://www.youtube.com/watch?v=kTUqyaSHwzA

You can see how the exhaust system is designed to come loose from its moorings rather than buckle during a crash -- I assume this is to stop people from being burned by hot exhaust gases if the engine keeps running after the crash?

All cars in recent years are designed to crumple well under offset frontal crashes. It's sort of an industry standard.

Even big old Volvos don't fair well in an offset with a tiny new Renault:

http://www.youtube.com/watch?v=qBDyeWofcLY

Compare that to an F150 http://www.youtube.com/watch?v=lB0araA0T_k&feature=relat...

Not at all what you'd expect. or is it?

Hm, when living in Canada I had one exactly like that, it really bugs me how fragile that thing is. I always figured my biggest risk was a roll-over, now I'm thinking the biggest risk might have been hitting a tree.

Incredible the deformation.

Afaik deformation is good, as it absorbs energy instead of delivering it to the people inside

Deformation of the front of the car is exactly what you want. Ideally everything in front of your feet should be crushed like crazy. But what you see in the Brilliance and the F-150 is that the passenger compartment itself starts deforming, increasing the chances that you're going to wind up with a jagged piece of door metal in your skull, or crushed between the roof and the floor. Nasty.

After viewing the F-150 video I was tempted to say that standards must be different for trucks -- they're built differently and have a helluva lot more momentum to dissipate coming from the back. But then I looked at crash tests for the Chevy Silverado and Nissan Titan, and it looks like that F-150 is just a piece of junk because those other two hold up a lot better.

Let's keep in mind that we're comparing a 2001 car with much more recent ones. Here's a video of the 2008 F-150, where it holds up dramatically better.

http://www.youtube.com/watch?v=1LkAzt_0qIg&NR=1

Progress is awesome.

Yes, that is quite a difference, it's like it's a completely new design from a passenger protection point of view.

Indeed: http://www.youtube.com/watch?v=KWaLPqrZI1E

Found in a related video:

http://www.youtube.com/watch?v=3ReZ0C_UF-8&feature=relat...

This is why an F-150 is still a pretty good place to be in a real-world accident where you're more likely to hit another car rather than a thick concrete wall. Conservation of momentum says that when a big heavy thing meets a much smaller thing the big heavy thing comes off looking much better.

That's actually not exactly what's happening. The force on each of the cars is identical, but the major difference I see is that the F150 is actually much taller than the escort. The grill/engine of the truck hit the car above its center of gravity.

All of the impact energy went into the truck's engine block, but into the passenger pillar in the car. That's why the car came off looking so much worse.

As long as there is some structural integrity to the passenger compartment, yes.

In the ford video the driver is the deformation zone.

Does it say anywhere what year that F150 is?

After the impact, you can almost make out the year and model on the door. It does look a lot like it says "2001."

What would have happened if the collision avoidance system malfunctioned and slammed on the brakes while you were driving fast on the highway, causing an accident?

I am kind of scared to give software control of critical aspects of my car. Do they test this stuff as well as the space shuttle software?

I have some similar sentiment, but at the same time, I wonder if it's a psychological issue? In theory, I should be willing to accept even flawed software, so long as the likelihood and severity of any flaws are less bad than my own. And humans are not actually all that good at avoiding car accidents, as the rather appalling crash statistics point out, so the bar isn't even all that high for "well, at least it's better than humans" software.

Yes, but this system seems to be worse than my driving. For instance, I wouldn't have hit the back of that goddamn truck.

How do you come to that conclusion? With the information we have it seems to me that this system can’t make your driving any worse. It’s a emergency system, it’s supposed to only kick in when you already failed. It too failed in this case, but that only means that with or without the system, sometimes you are dead either way.

This system doesn’t even want to be autonomous which means that if the worst thing is does is flat out fail (and not, say, misfire) it’s no better but more importantly also no worse than a car without the system.

I theory you are right, but I fear that drivers don't pay so much attention to their driving, since "after all the computer will do it for me".

There has been a fascinating linked posted a couple of times to a story about a guy who worked on the London road system: each time he would clear out congestion issues, more cars would go on the road - which meant that the average speed didn't change. He argued (or the article did, can't seem to recall whom) that people had a specific level of risk they were prepared to accept: make it safer than that, and people drove less safely.

> He argued (or the article did, can't seem to recall whom) that people had a specific level of risk they were prepared to accept: make it safer than that, and people drove less safely.

Tom Vanderbilt also writes about this in his book "Traffic". I recommend it to anyone interested in the psychology of driving and road design.

I worry that if it's buggy enough to get false negatives (not braking when it should) then it might also get false positives (slamming on the brakes when it shouldn't).

Maybe one day I'll have a car with a system like this, but I'd want to see it get quite a few years' real-world testing before I'd trust it to slam on my brakes whenever it felt like it.

That's illogical. We could design a system that works exactly like brakes do now — never engaging on their own — and it would get an enormous number of false negatives. By your logic, it probably gets a lot of false positives too.

In general, false negatives aren't a sign that the system is buggy, but that a heuristic is tuned to avoid false positives.

Also, this wasn't a false negative — the system wasn't turned on in the first place.

it doesn't need to be better than the space shuttle, it just needs to be better than /me/ and really, I'm not that great.

I think this incremental approach to the self driving car (starting with ABS, then moving up to radar braking, etc...) is a good thing.

At least Volvo aren't claiming they are the safest car in the world in Sweden anymore, after our courts ruled it was non-verifiable: http://www.thelocal.se/25492/20100312 :)

I can't tell if it's just the angle of the video, but it doesn't look like the airbags deployed. Maybe they were intentionally disabled, but yikes, seems like a car you should stay away from until they work the bugs out.

If I understood what I read correctly, the car came out of the gate at 30 MPH. It hit the truck a bit slower than that, it seems. Airbags might be set not to deploy under a certain speed because there might be more risks from the airbags themselves than from the crash below a certain velocity.

Alternatively this may be a prototype with airbags disabled for some reason. Frinstance if you're going to remote-control a car you'd probably wind up fitting something over the steering wheel, so you'd probably disable the airbags for that.

I thought most modern cars had a sensor that detected if someone was sitting in the seat, and only fired the airbag if there was actually a person there. I know my last rental car (I drive an ancient jalopy) had a little 'passenger airbag disabled' icon that went away when a passenger sat in the passenger seat.

Direct link to video http://www.youtube.com/watch?v=QJ6z3IArINI

Reply on news.ycombinator.com