Hacker News Reader

Comment by ghshephard | Hacker News Reader

The interesting thing is that, based on my experience with the cruise control/driver assist on the 5 Series, it's far better at responding quickly to changing road situations, like when somebody in front of you just jumps hard on the breaks. The car will take it all the way down to a stop in just a few seconds - quite often before the driver would have even realized something was happening, and way before they could have reacted to prevent an accident.

I'm guessing once the technology is widespread, we'll see almost total elimination of rear-end accidents on the freeway - the technology really is quite amazing.

View on HN

Ideal human reaction time is around 200 milliseconds[1]. At 60mph that's more than 1 car length.

How many milliseconds does it take the 5 Series to notice what's going on and start applying the brakes?

[1] If you are a) early 20's, b) male, c) responding to a audio cue, and d) doing exactly one thing on the cue (as opposed to making even a very simple decision based on the cue), you can get down to 150ms.

On the freeway, the time it takes to react to seeing the next car's brake lights isn't the main problem. The bigger problem is how long it takes to judge that car's deceleration.

Are they just tapping the brakes because the road is bending---some people always do---or are they slamming on the brakes because there's a traffic incident that you can't yet see? You're 200-millisecond reaction needs to be completely different in these two cases, but you just can't immediately tell the difference.

Both for people and for automatic cars, judging deceleration is a harder problem than just reacting to a red light going on, but the sensors on automatic cars should give them a big advantage.

> Ideal human reaction time is around 200 milliseconds.

Sure, if they're not looking down at the radio or at the accident on the other side of the freeway.

And then there's the time to actually make your movements. Getting your foot onto the brake pedal and pushing on it takes appreciable time, both for the foot to make the movement and for the nerve impulses to get there to make it move.

I recall from looking into this stuff a bit that a good reaction time from "oh shit" to "brakes are applied" is around 750ms. Accounting for people who aren't fully on top of their game (distracted, tired, injured, old, etc.), road designers assume more like 1.5-2s delay for applying brakes. And even that will be optimistic in some cases, as many car crashes demonstrate.

There's that, but also the realization that you need to brake "as hard as possible"-- we're not trained to do that often so there's additional delay before that happens.

Hence why he said "ideal".

And my point is that talking about the ideal human response time is absurd in the context of driving.

Why? It provides an lower boundary on the human response time. It's significant because the ideal response already represents _at least a whole car length_ of travel distance. We don't even need to delve into what a realistic response time is: it's already so bad that we should be looking at alternatives like our machine-overlord-self-driving-cars.

The protection is useful because for a highway rear-end accident to occur, a chain of at least two events has to occur. The car ahead of you has to significantly decelerate, and in the same window of time, the rear-ending driver has to be distracted for longer than the time it takes to close the gap and stop. Most of the time both of those events don't occur at the same time, but just look at traffic accident reports to see that in areas with heavy traffic - the probability that both events occur over the population of cars in traffic is 100%+ (because it happens multiple times every day).

Adding some sort of auto brake feature adds another layer of safety (presuming that the ratio of 'saves' to added risk is very high with this tech.)

With autobrake and safe following distance, even approaching 25% of cars using the systems would drastically reduce the amount of multi-car pileups.

I think a recent one in the news was 20+ cars. If just one of those cars had autobrake and safe following distance this could have been halved or quartered and wouldn't have blocked a highway for a few hours.

You will never be as fast as software, hardware and sensors.

This was the point I was trying to say, but I think I didn't communicate it well.

I would like to know just how fast these cars can react. I'm betting they could be actually slowing the car before a human is even aware something is wrong, but I would like to have some numbers to cite.

Also - keep in mind that this type of Driver Assist is (in the scheme of things) a very straightforward problem compared to others. It's pretty easy to calculate what type of braking is required to avoid running into something in front of you based on the relative velocities.

Not sure about the Model S, but Google's Prius and Lexus autonomous vehicles use a radar sensor on the front of the vehicle that can see past a semi tractor trailer in front of it, providing sensor input a human would never have.

Ideal humans do not cause accidents, normal humans do and normal human reaction time while driving is not around 200 milliseconds.

Better than what? The average human driver?

That seems to be the logical comparison.

Yeah but can the computer be situationally aware to see the kid in the 2003 3 series bmw ahead of him on his cell phone and preemptively switch lanes when his head bobs down to send another text to the furthest lane away from him before he ever has a chance to pose a risk to you in the first place?

We can increase the efficiency of reactionary driving as much as we want with computers, but for me, you just can't beat defensive driving.

Just curious, How well do these systems handle potential rear threats as well? How would it handle tailgaters? Some tailgaters are more aggressive than others. Some just need a friendly reminder that you do not appreciate the tailgating by a nice easy slow down, or a slight tapping of the breaks. Others -- such as the guy in F250 super duty with 30 inch tires and a lift kit, you might just want to go ahead and move over when you see them coming.

But the gauge for success shouldn't be whether it handles every situation better than humans, it should be whether the use of these tools leads to a lower rate of accidents/injuries/deaths. Just pointing out that there might specific situations where humans would do better, while downplaying all the situations where an automated system would do better, isn't all that useful. I mean, it works on an emotional level, but we should try to be objective by looking at the numbers.

Saying everyone should just drive defensively isn't a solution. The evidence for that is the fact that that's what we currently try to tell everyone to do, and it's still a problem that people drive poorly.

Right.

Saying everyone should just drive defensively is most definitely not a solution.

I'm not really sure what you're implying...

We need better enforcement? We try to do that by writing laws and issuing tickets. And, again, poor driving is still a problem.

How about we FORCE people to drive defensively? That sure sounds like what automated driving is trying to do.

> Yeah but can the computer be situationally aware to see the kid in the 2003 3 series bmw ahead of him on his cell phone and preemptively switch lanes when his head bobs down to send another text to the furthest lane away from him before he ever has a chance to pose a risk to you in the first place?

Probably about as much as the average driver. Hell, a lot of drivers on the road don't even notice a turn signal and get pissed that they're stuck behind a turning car in the lane.

Around here, I'd say the average driver probably thinks that an active turn signal is just faulty wiring.

For the foreseeable future, there will always be scenarios in which a human could perform better than an automated system. What I'm talking about here though, is the 6,000-7,000 accidents/year that could have been prevented by an automated system that is never distracted, and has <10 millisecond response timing.

What I worry, is that these systems will (of course) have bugs, and they are mechanical systems, so they will also physically fail - and, they will cause some deaths (cars are dangerous - 36,000 fatalities/year) - - but if for every one of those bug/physical failure deaths, you have 10 where the automated-system avoided an accident - that's a pretty big net savings in human lives.

And, software/hardware systems are only going to get better - that's certainly not the case for human drivers. Indeed, with cell phones, regardless of whether they are hands-free, there is research to show we are becoming less capable drivers than we were 30 years ago.

For the average driver who's on their own cellphone instead of doing "defensive driving", the system will do better than them. Professional drivers will still drive manually. Most people aren't professional drivers. (A reverse computer-to-car-analogy: most people don't know enough to know what programs are safe to run. Better something like SmartScreen/GateKeeper for most people. Professional computer users can "drive manually." But don't force everyone to do it, just because you want the option. They'll suck at it.)

I can seem to reply to derefr, but to echo his comment, this will be a safety measure in the same way that Apple's TouchID is a security measure.

We know that a fingerprint is a bad password.It can't be changed, you leave copies everywhere, authentication is based on an approximation, etc. Strong, frequently changed passwords would be much better. But is it significantly better than the security on most people's phones: a 4 digit passcode, or none at all.

If it can effectively respond to the dangerous moves of the kid on his phone, does it need to be as forward-looking and defensive as that?

Tradeoffs are different between computers and humans. Defensive driving is about trying to stay within an envelope where you can safely respond to whatever happens. A computer system will have a different envelope, so it won't need to do the same stuff.

To make a terrible analogy, a human pilot flying by eye needs to eventually make a decision to land at an alternate airport when his primary is fogged in. Is a computer smart enough to make that judgment call? Well, if the computer is equipped with a zero-zero landing system and can safely land in the fog, who cares?

Your question about rear threats is intriguing. The proper response is a lot less clear there, I think. Do you speed up? Slow down? Change lanes? Squirt them with washer fluid? Drop some caltrops from your rear bumper?

I'll be sure to set (hack, reprogram?) my future self-driving car to dispense liberal spraying of washer fluid after finally weaving around and getting in front of the asshat going 54mph in the left most lane on 101.

Reply on news.ycombinator.com