Cars That Avoid Crashes by Talking to Each Other | Hacker News Reader

Cars That Avoid Crashes by Talking to Each Other

nytimes.com

44 pointssew30 comments

My parents spent the better part of the last year contracting on this project. They traveled with car company engineers, DOT people, and test cars equipped with the system around the country to various race tracks to test the system with normal drivers.

On the race tracks they would put normal people in these V2V-equipped cars with (I believe) a professional driver/racer riding shotgun. Then, they would run people through scenarios that would set off the system - i.e. they would send two cars barreling towards a t-bone collision, and the testers would be instructed to not stop until the buzzer/light went off - at which point they were to slam on the brakes as hard as they could. My dad said that this test was the scariest and he chickened out while in the driver's seat. The system is great for blind corners where you cannot see the oncoming car, but on an open race track it is scary to see the other car on a collision course with you.

After the race track test, my parents would debrief and interview the participants.

Here are some scenarios I remember them testing:

* The peek-a-boo: Three cars are driving down the road. The first car slams on the brakes. The middle car veers to avoid it. You are in the third car and suddenly there is a stopped car that you are barreling towards. The system would have alerted you as soon as the first car panic-braked.

* Oncoming car - you are on a country road, about to pass a car. You start to pull into the oncoming lane to overtake the car, but the system senses an oncoming car that you did not see and sends an alert to you.

Also, I recall a problem with testers who were former policemen. Apparently they are taught to drive with two feet - one on the brake and one on the gas, for quickest reaction time. I think this fooled with the system because the alert was disabled as soon as you put your foot on the brake, and having a foot on the brake at all times caused problems.

One final tidbit: If you are ever looking to rent a race track in the USA, apparently the Disney/Richard Petty track has the best catering for the least money, compared to all the other tracks!

Regarding the two-foot driving police: I thought the purpose of using one foot to brake and accelerate was all about removing the mental process that occurs when having to choose which foot to use. I was always taught that removing that decision was the reason for using one foot. The point of heel-toe pedaling is that your foot can reach both pedals simply by pivoting, which theoretically makes both pedals equally quick to access. I learned on stick, which clearly dictates that style. But, I'd be interested in hearing what others think.

Regarding the stories from your parents: pretty cool info!

Warnings, yes please. Allowing my brakes to be controlled wirelessly by anyone clever enough to spoof the right data, no thank you!

For the curious, there is a system built-in to every modern aircraft called TCAS ( http://en.wikipedia.org/wiki/Traffic_collision_avoidance_sys... ), Traffic Collision Avoidance System, that does something similar for airplanes.

TCAS is definitely useful but it's not be-all end-all in collision avoidance, and unfortunately I think that translates well to automobiles as well. There's always the possibility of failure with the devices or perhaps sending erroneous(perhaps malicious) data that could screw with the system in other people's cares.

That said, I do like seeing these kinds of innovations!

Yes, a very tragic example that happened quite close to where I live was a mid air collision between a freighter and a regular airliner ( http://en.wikipedia.org/wiki/2002_%C3%9Cberlingen_mid-air_co... ), with 71 fatalities.

Combine this approach with Google's driverless cars and allow pedestrians to participate via their cellphones and the future looks brighter, safer and hopefully quieter

I worry about the potential for reliance on this system - 30 years from now (the time the article stated it would take for mass adoption), will we have head-on collissions around corners because one of the cars was malfunctioning and neither driver bothered to look "because the computer said it was OK?"

> I worry about the potential for reliance on this system

People also worried that drivers using ABS would be less competent than drivers not using ABS. Before that, we worried that drivers wearing seat belts would be less careful than drivers not wearing seat belts.

Risk homeostasis is a real problem (witness the number of SUVs lining the highway ditch during a snowstorm) but the sheer benefits to engineering-for-safety more than offset the risk of encouraging cavalier behaviour.

Re. ABS, one study seemed to indicate that when drivers have ABS they tend to adjust their risk taking upward

"Among a total of 747 accidents incurred by the company's taxis during that period, the involvement rate of the ABS vehicles was not lower, but slightly higher, although not significantly so in a statistical sense. These vehicles were somewhat under-represented in the sub-category of accidents in which the cab driver was judged to be culpable, but clearly over-represented in accidents in which the driver was not at fault. Accident severity was independent of the presence or absence of ABS."

http://psyc.queensu.ca/target/chapter07.html

The rate of people driving around corners looking properly is probably already staggering. I think, even given a relatively high failure rate, the statistics will still improve.

Mind you, like everything with autonomous vehicles, legislation needs to be laid down now in order to clarify who is responsible for what. I would hate to see a situation where this technology is withheld from the market because of the possibility of the company being sued for every penny.

Software can be refined, people can't .

Probably, but compared to losing 32,000 people per year, I'd bet the rate of computer fault would substantially lower.

Particularly when you consider the power of the systems we'll have in 30 years. That doesn't ensure quality (bugs, poor algorithms, etc), but it does make it plausible there will be numerous very effective fail-safes built in.

The x factor in that calculation to me, is the potential for people to treat computer fault deaths particularly harsh. If human fault kills 32,000 - I'd just about wager people would be more willing to excuse it than, say, if computer fault killed 1,000 instead (That is, there's some equation of human fault to computer fault deaths that is necessary before people will find it acceptable, and unfortunately I bet people aren't going to be very forgiving of computer fault deaths even if the ratio is 10, 20, or even 30 to 1. Let's say you eliminated human fault driving deaths, and all that was left is 5,000 computer fault deaths - I think people would be outraged at that number, quickly forgetting the old 32k number).

I think you're right about the double standard, but there is another factor - distribution of culpability.

Current situation: 16,000 drivers, each responsible for about 2 deaths.

Situation with the computers in control: 1 CEO of a manufacturing company, could be prosecuted for causing 1,000 deaths. Who would want to be in that position?

And as for being forgiving of human error, it's easier to be forgiving of an individual driver who causes an accident (as long as he was not reckless, e.g. drunk) because it could be you next time - anyone can make that kind of error. But the CEO of Samsung/Google/whoever could not be you: You know you will never have that job, and also you may be envious of their wealth, which will make you (and the judge) less sympathetic.

Can you really trust what other cars are telling yours? This sounds like an interesting security problem.

In Orlando FL, (one of the car-crash capitals of the world), the devices would be issuing warnings every 15 minutes (causing warning fatigue). I have to deal with these situations:

1. I am cruising on a road 45 mph. A perpendicular street intersecting the road has a stop sign. They are accelerating toward their stop sign. If they don't brake, there will be a collision. I often brake a little and prepare to brake hard if they don't brake at the last possible millisecond. A collision system would warn me in that situation.

2. People behind you accelerate up to you to 5 feet behind your bumper to "make a statement" that they didn't approve you pulling in front of them. Ethics aside, the system would think the person behind is out of control, causing a warning.

3. On the interstate going 75mph, cars pull in front of you with about 5 feet to spare. Cars coming into near contact with each other. A crash detection system would issue a warning.

So it means these crash detection systems need a "Tolerance" dial, that the idiot drivers are going to set as close to zero as possible because of warning fatigue. Also, this system will embolden the idiot drivers to drive even more dangerously. A system like this must account for warning fatigue in places where people drive like in Grand Theft Auto and would be confused if a system told them that what they are doing is dangerous.

> A system like this must account for warning fatigue in places where people drive like in Grand Theft Auto and would be confused if a system told them that what they are doing is dangerous.

No, it doesn't. In that case, people will just disable it or ignore the warnings, and continue their existing unsafe practices. People who drive like that aren't going to improve because of a box beeping at them.

I don't think that widespread reckless driving is a problem that can be solved with technology, short of full automation (driverless cars).

I'm not saying this isn't useful - but for those drivers, feedback that what they're doing is dangerous isn't going to change their behavior. Deploy it anyway, for the rest of us. It at least won't make reckless drivers any worse.

it would be interesting to have those warnings go directly to the car black box, even when disabled, it might help in case of collision. Or have it send the "log" every month to the DMV and remove points from the driver's license, and revoke it when it reaches zero, but this sounds too Orwellian to many people.

There's a device called Snapshot from Progressive that seems to do this. In fact, they set your car insurance rates based on the logs. You plug it in right under your dash.

The problem is that it's based on YOUR sudden braking

So some idiot cuts you up and you brake to give them a safety zone - you are penalized, if you blindly drive at top speed causing total chaos the box gives you a gold star

Reminds me of the Fifth Element: "You have 5 points left on your license"

That's because automated law enforcement is an Orwellian idea.

I'll happily take that kind of Orwell on my streets.

We have automatic speed-cameras and automatic toll systems already.

Bus and truck drivers are required to have a blackbox onboard (mainly to control that drivers don't exceed their allowed driving-periods).

I wouldn't mind a mandatory blackbox to record driving patterns. Feel free to go full Orwell and compute my insurance rate from my driving patterns, because that would most certainly fix quite a few reckless drivers.

As long as the process is transparent (i.e. everyone knows about it and the algorithms are open source) I don't see a problem with that at all. Streets are a highly regulated environment anyway (for good reasons) and since we're approaching the era of driverless cars this is just logical progression.

Well, Orwell isn't to be read as a prescriptive list of things you can't do.

The problem with automated law enforcement is quite simple: It operates in the gap between that which is dangerous enough that society decides to punish offenders, yet not dangerous enough that society needs to intervene to make the offender stop offending.

That coupled with the perverse incentives of fines being a source of revenue, and you have a recipe for law enforcement gone wrong.

Also, not explicitly linked to automated LE, but emphasised by it, is the graduation away from discretion in driving. Many drivers tend to forget that the fastest you can ever drive, anywhere, is as fast as is responsible. Even on German autobahns with no speed limit, you can get a steep ticket for going too fast - this is at the discretion of a police officer and ultimately the courts. Automated LE wipes out that discretion in favour of determinism.

> As long as the process is transparent (i.e. everyone knows about it and the algorithms are open source) I don't see a problem with that at all.

Well, you have near perfect empirical evidence to expect that won't be the case.

Or in a BMW it would just be wired to the ignition. - you are driving too fast / you are too close / you didn't stop at the red light - on a continuous loop.

Fascinating to consider that within a few decades that annual 32,000 fatality rate could be reduced by some radical amount (95%?) with software. That's an incredible amount of lives waiting to be saved by some good programming; 300k per decade is a staggering number of people.

I'm not aware of too many other private sector problems that you can go to work on that can have such an impact (with programming), outside of biotechnology perhaps.

In several European countries, fatality rate was reduced by a dramatic amount by being stricter on speed limits, alcohol, reckless driving, etc and building (much) better cars. See the statistics:

http://www.securite-routiere.org/Fiches/statistiques/statint...

In fact, USA are about the only developed country which didn't make any significant progress in the past 20 years.

only developed country Sorry, but the numbers don't support your argument.

Grèce 1502 1511 1699 1737 1790 1995 2008 2076 1993 2068 2000 2182 2116 2037 1880 1654 1605 1619 1658

USA 46390 47087 45582 44599 44508 39250 40150 40716 41817 42065 42013 41501 41717 41945 42116 42815 42884 42636 43443

http://www.google.com/publicdata/explore?ds=d5bncppjof8f9_&#...

http://www.google.com/publicdata/explore?ds=kf7tgg1uo9ude_&#...

So, both countries has reasonable population growth and the rate of fatalities dropped by a similar amount relative to their populations. What's different is the absolute number peaked in Grece in 1990's and dipped in the US in the 1990's.

To expand on that:

The second table on wazoox's link give the number of fatalities scaled to the population (in death per million) for European countries. It doesn't include the USA for some reason, but you can estimate it from Retric's info.

  - 1990: 186.3
  - 2005: 147.2

So, improvement there has been.

It still is much worse than most of Europe except Greece which just qualified in the last decade as a developed country for road infrastructure.

From my personal experience, the average American driver seems to have a frighteningly low driving ability coupled with a surprising self-confidence; and I've seen more road rage in 6 weeks in the US than in 40 years in Europe (apparently tendency to reckless driving is proportional to distance from the Ocean, too; Californians seem reasonable enough). There are probably lots of progress to be made in teaching and punishing there.

Reply on news.ycombinator.com