For $10,000, your car can drive itself – Cruise (YC W14) | Hacker News Reader

For $10,000, your car can drive itself – Cruise (YC W14)

theverge.com

252 pointsajaymehta159 comments

This reminds me of a Kickstarter campaign I saw a few months ago. The campaign's goal was to build low-cost bouncing balls that cleared mine fields. You could roll the ball into a dangerous, abandoned area and it would cause dormant mines to explode.

The problem with that campaign is that it instilled a false sense of confidence. The technology was "an in-between": certainly better than nothing, but not as effective or safe as traditional mine-clearing methods.

Cruise is in a similar position. They have a chance to improve people's lives and prevent serious injuries. At the same time, they're toeing a delicate line between improving drivers' safety and causing them to become complacent.

There is a fatal virus in the US that claims 3,000 people every month. And it's a nasty one. It's the #1 cause of death for teenagers. That virus is called a car accident, and the fact we're complacent with that is what scares me.

We understand the delicate balance you describe and it's our top priority to make sure our product actually does improve drivers' safety. That's why we're using independent third party testing by the same companies who test products for the major automakers, even though there is no law or regulation requiring us to do so.

> even though there is no law or regulation requiring us to do so.

Are you honestly saying there's no laws around safety requirements in vehicles?

Also, there's the fear of lawsuits.

We do not make cars, and very few of the NHTSA regulations or FMVSS apply to this kind of product (at least for now). My point is that we recognize our own testing, regardless of its thoroughness, is not enough.

And in what way, exactly, is that a virus? Epidemic, at a stretch.

(I'm not typically this pedantic... I just think viruses have particularly interesting information dynamics and it's really weird to me when applied as a metaphor outside of where that would make sense.)

I think "disease" would be a decent description. "Virus" carries way too many implications about things like communicability and origination that clearly don't apply.

I'd say it's more like cancer: lots of stuff influences it, and you can make yourself a lot safer, but it's ultimately fairly random and anybody can fall victim to it. You can't catch it from other people, you can't get vaccinated against it, surviving it doesn't give you any immunity for the next time, better surgical and other medical techniques are making it more survivable.... One big difference, of course, is the age factor.

"You can't catch it from other people,"

You might catch memes (drunk driving, reckless driving, street racing) that significantly increase the likelihood that you suffer from it. Still not quite the same thing, of course.

"you can't get vaccinated against it"

The opposite memes from the above would seem to serve a role not too far from inoculation.

"surviving it doesn't give you any immunity for the next time"

Except to the degree that it changes your behavior.

All of that said, I agree. "Disease" is a much better term.

Agreed. Was simply trying to demonstrate that we're far too comfortable with the consequences of unsafe driving. In any other circumstance we'd all be up in arms.

I appreciate your comment and what you're trying to do. I guess I still can't picture how Cruise will achieve the balance between convenience and awareness.

Let's say you're a teenager using Cruise to keep your car within the white lines on a 5-hour, highway road trip. The device is capable of steering the car this way on your behalf.

In that situation, do you stare at the white lines for 5 hours without touching the steering wheel? Or do you text and talk to your friends? What compels you to be just as vigilant about steering and being aware of your surroundings?

I agree, but the guy I drove next to reading a magazine this morning wasn't doing a terribly good job staying in the white lines this morning either.

So you'd rather his car automatically keep him in between the lines, so that he feels even more safe and complacent about reading his magazine, and rear-ends you later when you end up in front of him and have to brake suddenly?

I don't think I would. I think that's the point of the naysayers in this sub-thread. Doing X% of driving for bad drivers, might just make them even worse at the remaining (100-X)% of driving.

But the sensors and actuators that are keeping him in his lane would surely fire the brakes far faster and harder than he would while reading his magazine. If all the tech does is keep you in your lane, it's certainly dangerous, but I seriously doubt that's what they are building.

Fair enough; I believe they do include this, or maybe it comes standard with the car. Still, should we be "enabling" bad driving? Are these systems actually capable of letting people safely read while they "drive"? Does a driver have a right to do that? Is it prudent? Who is morally/legally responsible when something goes wrong and damage is done?

I guess as always, we'll figure it out as we go, but I'm not the only one who's a bit nervous.

Playing with phones, shaving, applying makeup, dozing off, etc., are all common. I ride a motorcycle. I'm dreaming of the day when technology makes those thing immaterial to the safety profile of a ton of metal moving at 70mph.

But the sensors and actuators that are keeping him in his lane would surely fire the brakes far faster and harder than he would while reading his magazine.

In fact, since human reaction time is on the order of 200ms, the car could likely see what's going on, double-check, and then hit the brakes before he noticed what was going on, even if he were paying rapt attention and expecting you to slam on the brakes.

What is the reaction time of these systems? There are papers on the Internet but they are behind paywalls (that I could probably get through if I were an automotive engineer).

> and rear-ends you later when you end up in front of him and have to brake suddenly?

All major car brands have their version of automatic braking already. http://en.wikipedia.org/wiki/Collision_avoidance_system

I'd say even if you're trying to be a good driver in this situation, your eyes are going to "glaze over" after 1 hour of staring at lines and occasional other cars without using the wheel or pedals. If this is true (and I'll grant it's a fairly big "if"), then this product would effectively impair the driver even more! Are we really that confident that the technology is good enough to lower crash/injury rates, even with zoned-out drivers? Perhaps it's trading one type of crash, for other crashes that are less harmful overall, but do we know?

I'd heard of this kind of fatigue before, and just now pulled up an article abstract[0] in search results. It says "analytical results indicated that 80 min was the safe limit for monotonous highway driving." Highway driving would be even more monotonous with this product. I'l grant that this may not be very solid scientific citation and reasoning here, but it seems like nearly common sense, that a driver can't remain alert for hours on end with a computer doing all his work.

[0] https://www.sciencedirect.com/science/article/pii/S003193840...

I have a three-hour drive I make twice a week. I can second the motion of "glazed-over" eyes.

But let me tell you: When the "collision avoidance alarm" fired mistakenly once, I was VERY AWAKE for the next half hour.

Assuming the sensors are good, and that they include failsafes that will let you know right away when they AREN'T working, it absolutely would be a net positive to have something else watch during the "boring" times and then let out a loud "Red Alert!" buzz as soon as something unexpected came along.

So no, I don't think the driver would be "impaired even more," at least not if the sensors worked well enough to throw an alarm if something unexpected happened. A set of sensors on top of a car should have a better view of the surroundings than a human, and wouldn't suffer from that kind of fatigue at all.

More, the human is not going to be hypnotized by the road if they aren't actively driving. As a passenger my eyes frequently check out the scenery; as a driver you're forced to watch the road nearly 100% of the time.

P.S. in my drive, I tend to pull over at the halfway point, so after about 90 minutes. And last night that did feel like about 10-15 minutes too long.

What about an AI that asked you questions (especially about your surroundings and the environment ahead of you) and generally engaged you to stay awake?

Even an audio-only version of Duolingo that taught and tested you so you felt productive rather than as though you were wasting time with a prattling AI?

Nice ideas, actually. Yeah, I'd consider that.

You have to be sure that whatever it is doesn't engage any "spacial" thinking, however. I remember reading about studies that demonstrated that giving people visual descriptions or 3d puzzles was more distracting than other kinds of thinking while driving.

While I sometimes just want to relax, there are times when I'm feeling I COULD be productive if there were something I could do that didn't involve my hands or eyes. :) One thing I've done is to talk into a voice recorder. But there is only so long I can talk about an idea in a useful manner.

Yes, interjecting with questions about the road, traffic in the area, terrain ahead and so on could be helpful if it wasn't too repetitive. I think it would have to be personable and engaging ("Her"-style) to be really effective.

It seems like eye tracking and "hands on the steering wheel" sensors would solve this.

> Highway driving would be even more monotonous with this product

I suspect it would actually be the opposite. If you can take your eyes off the road for say 15-20s every 15 minutes or so, the drive is much less monotonous, and likely safer.

Yeah, others have had some good counterpoints on this too. Maybe another argument with the same point, is that you can't be ready to drive at any moment if you're gawking at the scenery, etc.? (Or is the system actually going to be able to drive certain routes with a "driver" who can allow themselves to not think about driving at all?)

It seems like at some point there would be a qualitative change in the mental processes, where you're effectively not alert to driving, even if you think "I'm only looking away every few minutes." At least if you have been driving for years, where "driving" = "re-evaluating road and vehicle conditions every 3-5 seconds and reacting appropriately." Maybe it should be thought of as learning a new skill, passive driving.

> I'l grant that this may not be very solid scientific citation and reasoning here, but it seems like nearly common sense, that a driver can't remain alert for hours on end with a computer doing all his work.

There may be lessons to learn from pilots here. There are countless reports of pilots falling asleep in the cockpit while the autopilot is in control of the plane.

Although it still seems safer in spite of that

I think the biggest immediate payback from this technology is for trucks. A trucking company with a large fleet can more easily afford the cost if it cuts full consumption and reduces accidents.

The way to improve drivers safety is by insisting on better schooling. Compared to the kind of testing I had to go through to get a driving license in NL the American (and the Canadian, for that matter) license test is way too simple.

If you lower the bar to entry then you get more accidents, technology alone will not help with that unless you cut the driver out of the loop completely.

I'd upvote this 10 times if I could. It's amazing how much Americans rely on driving while being abysmally bad at it. The amount of driving errors, almost accidents and accidents you can see during half an hour driving on 101 on the peninsula is more than what you'd see in an entire day on the Autobahn.

It's more like 1.5x (per distance driven) when comparing Germany and United States:

http://en.wikipedia.org/wiki/List_of_countries_by_traffic-re...

Interestingly, average cars age is 8.7 years in Germany and 11.4 years in USA (in 2013) which is 1.3x:

http://europe.autonews.com/article/20130703/ANE/307029987/ge... http://www.edmunds.com/car-news/average-age-of-cars-in-us-ju...

Given what I've seen here vs. in European countries with good driving test standards, I'm surprised it's not much higher. Maybe the distances in the US are larger? Or perhaps our roads are wider and so have more wiggle room, which compensates for the stunningly bad driver behavior I see every time I drive.

Also, lower top speed limits and higher penalties for speeding.

Which seems like a potentially reasonable policy tradeoff: make it easier for more people to get their licenses, then make it harder for them to kill each other. (NB I would prefer more stringent licensing standards).

I didn't compare Germany and the USA, I compare the German Autobahn to a specific stretch of 101 in California (for the Autobahn I was thinking of the A9 here but save for the north east most Autobahns are pretty similar).

Comparing on a miles driven basis for the entire country doesn't show the full picture because the miles driven per road type are likely vastly different, traffic density varies extremely in the US but not so much in Germany, and there are 50 different sets of traffic laws and educational requirements across the US.

Unfortunately that doesn't work.

I live in Slovenia, where it takes at a minimum 30 hrs of driving with instructor (but usually around 40) until he lets you take the test.

Additionally young drivers need(i think its mandatory now, used to only be recomended) to go to school for safe driving (slippery road etc.).

And jet most foreigners from USA, who come here say we are one of the worst drivers they have seen. (guess they haven't been to south of Italy or Turkey)

We used to have high accident rate. The only thing that helped is expensive ticekts. Its easy to get 500+ EUR ticekt if you are not careful.

I agree that there is a remarkable difference between the quality of driving in Slovenia, Romania, Bulgaria, Greece and say United Kingdom, Germany, France, Spain, Scandinavian countries and so on.

But I suspect that your average Romanian or Slovenian driver would have a hard time passing a Dutch license test.

Quality of the education is just as important as quantity and should be paired with suitably stringent examinations.

Expensive tickets are just a way for rich assholes to do whatever they want to because they can afford it. (There are interesting schemes where the tickets get progressively more expensive as you earn more, those are more effective).

I disagree in part. Tougher tests have to be combined with mandatory retests to renew a license every 5-10 years, it's not just the young who drive badly.

So are you saying we should suspend our skepticism, and think of the children?

I'm kidding but also serious. This seems like it'd be incredibly difficult to test, and also like that testing would make an incredibly interesting series of articles/blog posts.

Well done for having a go at this. Given the global rate of car deaths at over 1m/yr you can definitely make an argument for developing cheap, slightly hackish systems now in rather than just the Google approach of developing very good but unavailable $100k systems. If this kind of approach can bring development forward a year or two that's a lot of lives saved.

Kyle, toootally dig what you're up to with this. Completely badass. If anyone can handle it, it's you.

Nothing but the best wishes!

-harry

>There is a fatal virus in the US that claims 3,000 people every month. And it's a nasty one. It's the #1 cause of death for teenagers. That virus is called a car accident, and the fact we're complacent with that is what scares me.

I don't think you understand the word "virus".

So you want me to pay you $10,000 to loose my freedom to drive my own car? Not a chance in hell. Look, I am sorry that 3,000 people die everyone month, but I would rather live in a world where 3,000 die in car accidents and people have the freedom to drive then live in world where computers drive everyone around.

The project I believe `hawkharris is talking about: https://www.kickstarter.com/projects/massoudhassani/mine-kaf...

My first thought is that it potentially could have a warning system stating whether the system guaged its ability to drive as "safe", "be careful", or "hey I can't do this, you should drive." A noise or color system could be used to alert the driver. Perhaps on a freeway with little traffic that the Cruise has seen plenty of times, it could be fully safe. But on some backroad that it hasn't seen before, or during bad weather conditions, it would go into fully manual mode.

Spitballing here, but let's imagine a common impaired driving situation- being very sleepy and needing to get somewhere on a time-schedule.

I don't know if you've ever been in that situation, but it's possible to fall all the way asleep while driving. It's such a common thing that they carve rumble strips into the sides of highways.

Being so sleepy is a tricky thing. You can fall asleep even with the radio going full blast. I can't imagine any alarm system that would stop someone being lulled to sleep by the engine and movement of a car doing 90% of the driving.

Comments here are fairly negative. Pointing to downsides, insufficient upsides and such. Gimme fully self driving or go home. Here's my perspective:

I think that self driving cars are coming and that. Even cars that drive on built-for-human roads are just a bridging technology. Eventually roads will be built for robocars. I think it's a big important technology. All transport revolutions are.

To get on that road we need to start getting this tech into consumer land, a beachhead. I don't know if this specific product is it, but it could be. It's got a nice gradual path from cruise control plus to auto-drive for 80% of the ride.

There might be other beachheads. Long haul trucking is often mentioned. This seems like a very similar problem. They both need to deal with highway driving only. Maybe robocar friendly road networks will be extended into built up areas.

I'm happy to see a company take what's working now and put it in a consumer product. We'll see if people like it. If they do, it seems like a good a place to start developing this stuff outside of research labs.

edit: clarity

Exactly! People fail to factor how yc companies evolve - how fast, and how far. I made this mistake twice in my life. These days I just stfu.

3 years back, I met with the poundpay team. They knew zero about handling fraud, and they said so. They said they'll figure it out over time. The cofounder said he was "reading a bunch of machine learning papers over the weekend" on combating fraud. I found that amateurish & scary.

2 years back, I met with the interviewstreet founder. The founder did not have any technical chops to speak of, and was actually quite misinformed about scalability challenges, algorithms & programming languages. I couldn't see how this guy could take on topcoder.

Today, 3 years hence, Balanced has processed over a half billion in payments, & Hackerrank has signed up a half million developers & both are well on their way towards a billion dollar market cap. Both are great companies to work for - in terms of actual tech & the potential for ipo riches.

What Cruise is right now is a useless datapoint. Think about what it'll be 2 years from now. Use your imagination. There are tons of low hanging features they could iterate on & completely dominate the space. Enough said.

The underlying message I'm getting is that HN applicants tend to find the "actual" problem they can tackle to make their fortunes while initially trying to solve a different problem.

Well, that's exactly what all the "lean startup" types say :) , the theory goes:

- 1) release an MVP - 2) get actual feedback - 3) pivot - 4) repeat 1, 2 and 3 until you find the actual pain point to solve, or run out of cash - 5) profit :)

I was thinking of the "schlepping" thing, which is pg's observation that the actual pain point is often your own.

> To get on that road we need to start getting this tech into consumer land, a beachhead.

But it's been there for years! All of the high-end car manufacturers offer this technology, in their current cars, well-integrated, well-tested, and for less money than what this company charges!

Remember, they're not offering self-driving, they're only offering lane assist and adaptive cruise control.

But the vehicle manufacturers are also working on self-driving, it's not just Google! BMW and Volvo have both committed to having self-driving cars on the roads by 2017.

I'm not sure what the price point is, but I'd pay non-negligible dollars to add this kind of tech to my '99 Accord. I don't even want the full $10k package with "self-driving," just lane assist, warnings, adaptive cruise control, etc. I think many of us would be onboard if someone can work the value proposition so that buying such a kit/package is a better idea than buying a new-ish car.

I understand why this sort of progress is interesting... but I feel like in this form, this is of minimal value. If I have to pay attention anyway (because of the risk of failure or situations it can't understand), then it isn't that helpful. I mean, it isn't like I can just zone out and read a book: I still have to pay just as much information.

Actually, to be more concrete, I definitely see the utility of this as an extra line of safety (a fallback if/when the driver does lose focus). But as something that is "between cruise control and self driving cars"... nope, not convinced. I wish it were pitched more conservatively.

It's useful in the same way cruise control is.

managing one's speed so as not to get pulled over for speeding?

I probably wouldn't drill holes into my brand new car's roof and spend $10k just to get something like cruise control.

Also FWIW, the Audi in question has adaptive cruise control as an option. That seems to be a much better bang for your buck.

Lane Assist is also an Audi option (but maybe only in A6+), so I really don't understand the value of their proposition.

This "frightens" me a lot more than the Google car does. I trust that automated cars will be safe "enough." I don't think I would necessarily trust an automated car that can find itself in a situation in which a driver needs to intervene. I barely trust other drivers on the road (maybe its the NYC driver in me...). Trusting those drivers' reaction times in a situation where the car needs to disable automation would require a lot of built up good will/trust.

That being said, I think this could be a great idea for more closed environments. Honestly, I'm not really sure where they are. Maybe huge airport parking lots (as "drop offs" at specific locations), or in large corporate parks or something of that nature.

Aside, but can we drop the scare quotes? If something frightens you, just say it. Or choose a better word.

That's probably fair. I, funny enough, used quotes, because I couldn't think of a less serious word than frightens. Concerns is maybe a better word choice.

Troubles. Bothers. Unnerves.

It seems like vehicles will have to go through an "unsafe valley"[0] where there's a huge gulf between "does a couple of things really well" and "go to sleep and you'll arrive safely at your destination".

[0] Not the same, but it's a bit analogous to http://en.wikipedia.org/wiki/Uncanny_valley

No car on the market offers a driving assist at this level (both lane tracking and distance keeping). I personally drive hundreds of freeway miles a week and would pay much more than 10k to avoid doing that.

It has been super impressive to watch the Cruise team of just 4 built a car that can drive down the freeway in just 7 months.

As far as I understand their technology this is not true. My understanding is that the system keeps a distance to a vehicle in front and corrects the steering should the vehicle leave the lane.

Even Audi themselves offers active lane keeping assist and adaptive cruise control systems. I'm not sure if this is available in the A4, but they offer it in their more expensive models. Mercedes-Benz also has both systems available in the S-Class.

I couldn't find any detailed information on their site as to what they cover, so please correct me if I'm wrong, but I don't see their additional value right now.

I had a drive in one of Cruise's prototypes a few days ago. This technology works.

About a year ago, my family had a car crash that shook up my 3-year-old daughter so much that she still talks about how scary it was. This technology is important.

Congrats to Kyle and the Cruise crew. I'm very impressed, and you guys should be very proud.

My 2010 Lexus had "lane keep assist" and adaptive cruise control (which used radar). It would bring you from 60mph to stop if needed without you ever needing to press the brake.

Adaptive Cruise Control + Lane Assist is in fact quite common. You can get it on the big Audis, JLR have it as standard and even Ford do it (I worked on it for the Kuga, but it will trickle down eventually)

> No car on the market offers a driving assist at this level (both lane tracking and distance keeping).

Yes they do. Mercedes, BMW, Audi and Volvo (and probably many others) offer this technology in their cars, and adding that option is much cheaper than $10k, comes fully integrated, and appears to offer the same benefits.

Meanwhile, this Cruise thing only works on Audi S4/A4, sits on the top of the car, costs more than the manufacturer charges for it, and doesn't appear to be better than that version.

It's really impressive to offer an after-market adaptive cruise-control system, kudos for that, but who is the customer here? Who would buy this thing?

If you have an Audi A4/S4 and $10k, why not trade to one that has the option installed?

While some automakers have lane keep assist on the market, in most cases it just vibrates your steering wheel or, at best, lets your hand wander away from the wheel for up to 15s and only on mostly straight roads.

Cruise is meant to operate hands-free for nearly the entire highway segment of your trip and over a speed range of 0-80 mph. No product on the market does anything close to this.

Related technologies that sit somewhere between cruise control and self-driving cars in terms of automation (I only have personal experience with older versions of the BMW and MB systems, and the situation has clearly improved in the last few years):

- BMW active cruise control will automatically adjust if you get too close to the car in front of you

- Infiniti cars will stop themselves if they detect obstructions in the direction of motion (rear and forward)

- Ford cars can perform the motions of parallel parking

- Mercedes-Benz cars warn you when you drift into the next lane

Other car manufacturers have related technologies.

I think most of those are a little more capable than that, it's mostly a question of which aspects of automation their playing up.

EX: A high end Acura with adaptive cruse control will also automatically break if it detects an unavoidable collision.

Still, I think we are better off focusing on assistive driving systems like blind spot warning, stability assistance, etc than slightly better forms of cruse control.

The newer Mercedes-Benz system will actually steer the car to keep it following the car in front, while in slow enough traffic (< 37 MPH).

Mercedes-Benz has a very nice product, but it's not hands-free. There's a huge difference.

That's the point here. Drivers will be tempted into a false feeling that the car is driving itself, just because it accelerates, decelerates and steers. MB has the same functionality, and I actually worry that it would cause even more distracted driving.

I think people probably said that about cruise control when it was introduced, and that turned out okay.

Define "okay." Cruise control has gotten "stuck" at 50mph[0], and electronic throttle control has left the throttle open disregarding user input[1].

So yeah, it's not some apocalyptic, universally-agreed failure that happens every day, but people have actually been killed[2] because the machine did something the driver wouldn't have done if they were in full control. (Maybe it's "compensated" for by deaths that were prevented by having cruise control -- I suppose we can't know for sure.)

[0] http://www.dailymail.co.uk/news/article-1236020/Horror-ride-...

[1] http://embeddedgurus.com/state-space/2014/02/are-we-shooting...

[2] http://www.eetimes.com/document.asp?doc_id=1319903

I believe that's only for legal reasons. When the laws get sorted out MB looks to be in a great position.

My boss has one of the new Infiniti's and it also automatically stays within your lane if the car starts drifting on its own.

> - Infiniti cars will stop themselves if they detect obstructions in the direction of motion (rear and forward)

I wonder if that can be disabled? Otherwise, an Infiniti would be a poor choice of vehicle for escaping a zombie apocalypse, or more realistically for making some types of emergency maneuvers (e.g. "I need to get away from a threat, and to do so I need to go over the median or through a small barrier"). Sure, it's the one-in-a-million case, but nonetheless it'd be ridiculous if there was no way to override it even for emergencies.

If I were designing such a system, I imagine that I'd only apply it in a certain throttle range that could be described as "cruise" or so, and disable it near full throttle, maybe with a bit of additional input that sees if you're currently adding throttle and takes that as "I don't want to stop right now, really". I have no clue if that's how they really build it....

If so, I bet the guy who got to code the 'ramming speed' override had fun doing it.

Going backwards, you'd always want to stop (little Timmy ran behind the car in the driveway). Going forward, outside of unrealistic hypotheticals, I think you always want to stop. How many pedestrians get hit walking across the street because somebody turning a corner didn't see them?

In fact this technology exists and is quite common in higher end vehicles. The terms are Adaptive Cruise Control (keep same distance with the vehicle in front) and Lane Assist (keep the car within the white lines, as long as the curve is <Xdeg). All JLR vehicles have it as option, as do the larger Audis. Even Ford offer it as an option on the Kuga (which is their highend SUV in Europe).

I worked on the systems for both JLR and Ford, the former even have an option where cameras read signposts to adapt the speed (falling back on GPS info if none were visible).

When I lived in San Diego, my route for my bike commute to work included Interstate 5[0], because bikes can legally take the interstate shoulder in California where there's no practical alternative route. Y'all handling that edge case?

[0] Here: https://www.google.com/maps/@32.886728,-117.225045,3a,25.3y,... (note the bottom of the Bike Route sign says 'use I-5 shoulder')

It is illegal to drive on the shoulder, so they need to handle this case for a lot of reasons (another being that substantial debris from collisions usually ends up on the shoulder). I do wonder how much space this thing will give cyclists when it passes them if they are near the white line, or if it understands to stay behind sometimes.

Making the computer react to a cyclist ahead who is indicating a lane change or turn with his arm will be extra tricky.

This video shows how the Google self-driving cars react to cyclists, as well as obstacles on the road and such: https://www.youtube.com/watch?v=dk3oc1Hr62g

No idea how Cruise deals with this, but it does appear to be possible anyway.

Mad props to the Cruise guys for taking this on. This kind of life and death programming is not something I'm willing to do. Just too freakin scary.

Self-driving cars will not be able to catch on without legislation limiting damages to manufacturers.

Proponents of self-driving cars predict ~90% reduction in fatalities. What this means is that over 3000 people per year will be killed by self-driving cars. This is way better than what we have now, but in fatal car crashes, often the driver at fault is killed, and juries tend to assess much lower damages against dead people than against large corporations.

Combine that with the fact that there typically aren't any damages at all awarded for single-occupant single-car collisions where the driver is at fault, and it seems entirely possible that the total damages awarded for traffic fatalities could stay at the current level, or even go up, leaving the manufacturers of self-driving cars to foot the bill.

Now, I generally think that laws to cap damages are not good policy, as it does make it harder to discourage negligence or even malfeasance.

Although the limited self driving systems we have already on Audis and the like are catching on already. I guess the driver has overall control so it's treated legally like a cruise control or similar. I think the main barrier at the moment to the adoption of the current limited systems is cost. If it became a $500 option I think basically all new cars would have it.

What this means is that over 3000 people per year will be killed by self-driving cars

I appreciate this point, but there are two big assumptions hidden in there.

1. That number turns 30,000 accidents into 3,000 accidents, but that's assuming all cars are self-driving.

2. It assumes that in those 3,000 accidents, the self-driving cars are "at fault." But those last 10% of accidents are things in which driver error is not a factor. If a cinderblock falls off a bridge onto my self-driving car, it would take an incredibly aggressive jury[1] to blame that on the car.

What is the cause of those last 10% of accidents? Is it things like manufacturing faults, in which case the manufacturer is already absorbing the risk? (And a computer-driven car could handle a blowout better than a human could.)

Oh, I found an NHTSA paper[2] that gives that "90% due to human error" figure, and puts 4 to 13% on "vehicle factors (brake failure, tire problems, etc.)." So I think my point in the preceding paragraph holds: a big company is already on the line, the automatic car could do better health checks to prevent them, and it could have a better way of dealing with mechanical failure.

[1] Not that such a thing is impossible.

[2] http://webcache.googleusercontent.com/search?q=cache:0B5GzBW...

I think you missed the poster's overall point: how people perceive accidents matters. This is independent of what the actual causes are, and if there are less of them. When automation is involved, people may perceive that someone did something wrong, even though that automatic thing outperformed what a human could achieve.

I think manufacturers are already facing damages for those remaining 3000 fatalities. I assert "deep pockets" is the overriding factor.

I'd really like to get a sampling of those 3000 non-driver-error fatalities.

I think the term of art is "vehicle factors." Using that I found http://www.ncbi.nlm.nih.gov/pubmed/17916886 which suggests that manufacturers could address "the majority" of them, and I don't think I'm being crazy by suggesting self-driving vehicles already include a lot of them, since they aren't exactly Geo Metros.

Self driving cars will get into accidents that are not currently "vehicle factors" since they will be superior drivers, not perfect drivers. Just imagine the headline: "A computer killed my son."

I've long said that people would rather a human kill 1000 people than a computer kill one.

I wonder what the long term goal is? Getting acquired by some auto manufacturer?

I don't see many people that own modern enough cars (having the required electronics/mechanics already installed so all that's needed is a sensor and some data processing to send out control signals) investing into a retrofit device instead of just buying the next model which comes with the same thing out of the box.

But I have to say that I'm impressed that they have made it this far, and almost certainly without any support from Audi.

We use our own actuators to control your steering wheel, gas pedal, and brake pedal. It's designed to work on any car, regardless of whether it has any drive-by-wire capabilities.

My blood runs cold.

What, you think you can't overpower a little servo?

I'm more worried about having to outrun it.

https://www.google.com/search?q=old+person+drives+through+fa...

An actuator getting stuck on heavy acceleration? I'm also a skeptic, but you should be more clear about what your concern is, since the proponents will say some kind of automatic radar braking system could have prevented this.

On many cars, even if a deliberately hostile entity takes control of your gas pedal, hitting the brakes cuts power to the engine. And even if it doesn't, brakes can usually (not always, but usually) win a fight with the accelerator.

Of course not.

I'm afraid of the interaction between laypeople, a retrofitted cruise control that was designed on a kickstarter budget and several tonnes of hurtling steel.

Very interesting, how do you control the steering wheel? Most solutions I've seen for that were either really clunky (and unsafe, as they obstructed the airbag etc.) or used/replaced the power steering. I just assumed you would have used the existing interfaces, I know for a fact that at least the current A6 has all the necessary hardware on board, and am pretty certain that the A4 has the same for park assist.

Have you thought about selling your system directly to manufacturers for integration? I'm certain that there are chances if you're able to deliver a working component that can simply be integrated into existing models. Most of the current assistance systems aren't developed in-house but come from one of the big third party suppliers.

We 3d scan the cavities in the driver's side footwell, design brackets in CAD, then manufacture them mostly out of 3d printed polycarbonate and aluminum. They bolt directly onto the existing components and are tucked away behind the trim. It's one of our design requirements that the actuators are completely hidden and imperceivable when not in use.

We don't touch any of the existing drive-by-wire controls, even if they're present. It's too dangerous to send signals down a CAN bus without a full understanding of the safety implications, and that kind of proprietary information isn't available to us.

That is a curious choice. Why not use existing sensors on the vehicle?

I don't want my car to steer itself, at least not just yet, but I do want to add a radar-based aftermarket active cruise control system to it that can maintain a preset following distance to the car ahead by controlling the accelerator and brakes.

This was actually available as a factory option in my car, but I didn't order it because I thought (incorrectly) that it didn't work in stop-and-go traffic. I've been kicking myself for that ever since.

Perhaps Cruise could consider developing an ACC system for the benefit of customers who would welcome some automation but who aren't quite ready to surrender the steering wheel yet.

These systems have been available for several years now, so I don't really see how many people would buy a retrofit option. People don't even upgrade their car stereo systems anymore.

That's because the manufacturers tie so much stuff into them. It's been 3 years since I looked into it, but it was my understanding that if I upgraded my stereo I'd lose the nice display just below the windshield line. I think I'd also lose either the steering wheel music controls, or the bluetooth controls for my phone.

IIRC, you have to have an ACC (Adaptive/Active Cruise Control) enabled vehicle in order to control the speed over the standard OBD-II port. This is because the vehicle has to have the capability to control the speed already, in order for Cruise to interface with it and adjust the speed. In other words, any car that Cruise can work on, should already have ACC built-in, or at least have optional ACC in that model.

Since Cruise works on any Audi A4/S4, I'm guessing Audi uses the same base control component for all 4 series, so that even for a vehicle that does not have ACC or Auto Parking enabled, Cruise can still interface with the vehicle and adjust steering wheel and speed. This suggests that upgrading your car to support ACC might be feasible?

Elsewhere in the thread, Cruise's co-founder indicates that they're using the old-school mechanical automation method: actuators on the brake, wheel, and throttle.

Most cars with active cruise control don't comply with any standard - while the commands to accelerate, decelerate, or steer may be available over a CANbus which is exposed on the OBD-II port, there's no standard like there is for OBD-II diagnostic information and the protocols are often obfuscated or encrypted (either for "security" or manufacturer protectionism).

FYI: http://reserve.getcruise.com/join-the-driverless-revolution

The font here is illegible. Some weird anti-aliases very-thin white-on-black. Highlighting doesn't help.

I have this too in Chrome 35 on Windows 7.

Here's how it looks: http://imgur.com/SSMV8j6

Or, you could just buy the already existing product. Is there anything they want to add beyond these?

http://www.mercedes-benz-intelligent-drive.com/com/en/1_driv...

For Mercedes: https://www.youtube.com/watch?v=JSXUApikcOk

For Audi: https://www.youtube.com/watch?v=T8sJwuZyVAY

Similarly from other high-end car producers.

Cruise goes on an Audi A4, which is roughly half the cost of an A7 ... so you could get an A4+Cruise and still have money left over

The A4 absolutely offers "active lane assist" (steering) as well as "adaptive cruise control with stop and go" (automatic throttle and braking, all the way down to stopping in traffic).

http://www.audiusanews.com/newsrelease.do?id=2747

I don't think it's quite as autonomous as Cruise - as noted by others in the thread, Cruise seems to be trying to reach a somewhat scary "good enough that you could read a book but please pay attention because it doesn't work just right yet" plateau, while the Audi system is simply supposed to be a gentle helper of sorts.

But the Audi system also works on any highway rather than just pre-mapped areas, and is (for better or for worse) integrated with OEM brake and steering actuators and the car's throttle-by-wire system rather than relying on retrofitted physical systems.

This is exactly the sort of thing I want to do next. Good luck guys!

It's perfect. A real world problem with world changing impact. It's hard. Its orders of magnitude harder than anything else startups are doing around here! The opportunity is almost infinite, and there are potentially billions of dollars to be made. I wish I was working on it!

I really hope these guys go all the way – although I'm sure there will be a ton of offers along the road. I wouldn't be surprised if Google X hadn't already made an offer, but if it comes down to it, I hope they go with Tesla over Google X. I got an inside look at their autopilot project a few months ago, and I'm really excited about it!

Honestly though, there are probably a lot of other big players who could be interested – Amazon especially. Again though, I really hope they take this all the way to a mass market consumer product, preferably with multiple iterations, while expanding the capabilities toward what Elon described as the 99% Autonomous autopilot.

This feels like an addition to the future self driving cars more than anything else. Sort of like a adaptive cruise control (ACC) or Active Lane Keeping Assist in Mercedes? I see trust & cost being a major hurdle here. Like trusting a self driving car to take you home.

I believe we need to totally remove the driver to really stop the number of deaths on the roads. As indicated below by others, there are plenty of bad drives in the day.

When I came to the valley few years ago I was shocked to see the number of drivers failing to use indicators (blinkers). As a pedestrian this was alarming. How do you cross the road not knowing intent of so many cars on the road. As a driver it was tough to gauge the intent of drivers around you. The human behind the steering wheel is the problem and that is where most of the disruption should be focused on.

Once we replace the human driver from the machine we will see less deaths on the road, faster commutes and more efficient use of resources.

I've been thinking about the applications of auto-driving cars a lot over the last year or so, and I think this kind of consumer use is not likely to be that successful.

It seems more likely to me that this kind of highway driving system would be installed and work on tractor-trailers. It's a good compromise; drivers keep their jobs but get less stress, and when these are legalized with less issues, the next stage where driverless trucks are introduced can come into play easier legislatively.

Legislation will likely be the biggest issue with these things in the United States. Taxi drivers already oppose Uber and Lyft; what do you think they'd do with an invention that removes their profession entirely? And their legislative pull pales in comparison to the teamsters'.

So, this kind of staging is a good idea for introduction to driverless vehicles. It just won't work for an Audi.

I'm very curious about the procedures and paradigms followed to develop the safety critical software inside the system, as well as the engineering pedigree of the developers. I don't mean this in a rude, finger pointing way, just proposing that some background and a slightly deeper technical description of the precautions would make a huge difference in confidence level.

For example, are the redundancy systems space and time separated? What kind of methods were used to ensure that? Are there any overall hardware, software, communications development standards used and audited? Does anyone on the team have demonstrated experience in these areas?

If this system were to be audited, what organization would be responsible?

Thanks guys, great work so far and excellent vision!

What about exterior cameras combined with interior projection to provide an AR experience? Show optimum trailing distance, highlight hazards, display useful information in the field of view. Is that illegal to do, perhaps?

Reading through all the safety concerns reminds me of this:

http://spectrum.ieee.org/computing/software/automated-to-dea...

The problem I see is, if you still have to be fully attentive, supervising the computer and ready to take over, but not actually do anything... driving will be mind-numbingly boring. (That's assuming people are even capable of maintaining that attention, which many other comments address.) I mean, driving is usually already kind of boring, but at least the task keeps some part of your mind occupied. If the computer can take over completely - Google style - then you can do something else entirely. But this sounds like some kind of intermediate purgatory.

It's also worth noting that Google hires test drivers specifically for ability to maintain full attention during mind-numbing boredom.

Mmmm according to the article Cruise is not a driver replacement. It can steer, brake, and avoid objects. I'm sure it can be quite safe on a highway, but what are its limits? Can can and cannot do?

It's clearly an intermediate product for people who want to get closer to driverless vehicles. I doubt this will make strict business sense for anyone right now.

It seems more like an acquisition play as all the major car makers are going to end up in this space in the near future.

There's going to be a big business in self driving car retrofits -- convert non-self driving to self driving. But it's going to need to be fully automated for people to bother.

Has there ever been big business in retrofitting cars? More likely a small number of hobbyists will retrofit and for the rest it'll gradually roll out from manufacturers. Cruise might be hoping for a partnership with Audi or something.

What happens if a leaf or some other obstruction blocks the cameras?

The human takes over until it can be cleared I assume.

This system is not a self driving system -- it's basically a smarter cruise control. You still have to pay attention and be at the wheel. At least for now.

That just sounds dangerous. A system that takes care of a lot for you, until it doesn't. And then you have to step in at a moments notice, even though you might not be paying enough attention to react fast enough. I think there has been cases of airline pilots' senses being dulled by all the automation, and then them stumbling to fix things when they have to step in and do its job manually.

It seems to me that self-driving cars has to be a all-or-nothing deal.

It doesn't have to slam on the breaks if a leaf gets in the way. It can continue driving its last-known-good-path while gradually slowing down, signalling to the driver that it's lost its data connection and that it's time to take over.

I do wonder what it takes to block all the sensors at once, though. It has both cameras as well as radar systems. How independent are those?

Honestly, I tend to agree (despite asking for a similar feature in another comment.)

It's easy enough to build an automated vehicle, but, wait, what? It has to coexist with human drivers? Mmmm, I'll get back to you on that.

OTOH, smart cruise control and obstacle avoidance are actually shipping features on current model year luxury cars.

http://en.wikipedia.org/wiki/Autonomous_cruise_control_syste...

http://en.wikipedia.org/wiki/Collision_avoidance_system

Oh, and other systems to notice if you start getting drowsy:

http://en.wikipedia.org/wiki/Driver_drowsiness_detection

This set of technology is already on the road today, from nearly every major carmaker. Just look at the luxury car in each line, and find "electronic power steering", "adaptive cruise control" and "active lane keeping". The 2013 Lincoln MKZ is one example, and there were tons of writeups calling it "self-driving, sometimes" when it was announced.

That's the combination that lets the car drive itself on well-marked roads like highways: EPS/ALK allows it to turn the wheel to keep you from crossing divider lines, and ACC maintains a minimum distance from any cars in front of you. Many of them also have automatic crash avoidance that will stop the car if the car in front of you suddenly slows and you don't hit the brakes yourself.

As for ensuring the driver can react fast enough if any of the systems become unavailable, AFAIK all the cars that have this set of features also have systems that will beep at you annoyingly if you take your hands off the wheel. It's one of the unwritten purposes of those "driver drowsiness detection" systems these same luxury cars now come with.

> As for ensuring the driver can react fast enough if any of the systems become unavailable, AFAIK all the cars that have this set of features also have systems that will beep at you annoyingly if you take your hands off the wheel. It's one of the unwritten purposes of those "driver drowsiness detection" systems these same luxury cars now come with.

That's great, 'cause that's one of my concerns with this kind of thing. Still, I wonder if it just might be better to leave everything to be handled manually, since then you have more things to keep you busy and thus awake. I imagine that I could end up like I sometimes do when watching a boring movie late at night - it's very hard to stay awake because I'm not invested in what I'm watching and I am not doing anything but watching it. Of course, it might just be best to pull over and sleep for 15 minutes at that point.

Can this be removed easily? It looks like it would be a beacon for vandalism unless you spent your time in secure parking.

Shame that thong looks ugly on the roof if that car. Who's liable if an accident happens when Cruise is turned on?

Being that this is marketed as a sort of cruise control++, I would suspect the same party who is liable when an accident happens while cruise control is engaged.

Legal costs would be very high, in any case. Look at Toyota's gas-pedal settlement.

Please go public soon so I can buy shares.

so what do they expect will happen the first time someone gets in a serious accident while using one of those?

I assume it's an MVP. They're probably hard at work on all of the core challenges of interpreting the road data in real-time and controlling the car. It's technology that I expect a bunch of big industry players want / will need. Lets see where it goes.

I interpreted this as a hilariously myopic SV view of the issue. "Move fast and break things" is not exactly a winning motto for the development of automotive tech, unless you're into the demolition derby scene;)

The article states they plan to sell directly to consumers, not "big industry players", so it is a real issue. It also states that this is being released under the aegis of new CA law that goes into effect in the fall. I would imagine this law makes at least some provisions for liability, but the reality will probably be played out in the court systems for the rest of most of our lifetimes.

This is an issue for all self-driving car tech.

A little Googling uncovers the fact that airbags killed some 158 people in a recent year -- in low-speed accidents where the occupants would have been unharmed without the airbags. But airbags continued to grow in popularity (though now they're mandatory in the US, IIRC); automobile manufacturers just consider a certain amount of money paid out in liability a cost of doing business.

To help protect the manufacturers, they should make sure to have a good black-box recording of all the sensor data in the case of an accident. If there was a bug, they need to pay out the damages and fix the bug. If the accident was caused by someone else behaving dangerously, though, they should be protected from liability.

Bad press, however, is likely no matter what. The recent "Teslas can catch on fire!" media storm shows that; as I'm sure most HN readers know, a smaller percentage of Teslas caught fire than gasoline cars, and no Tesla owners were harmed, making the Tesla fires a non-story. Or it should have been a non-story, anyway.

Not 158 in one year, 158 "to date" as of the year 2000 (source below): http://www.airbag-law.com/index.html "As of April 1, 2000, the National Highway Transportation Safety Administration's Special Crash Investigation (SCI) has confirmed a total of 158 fatalities induced by airbag deployment. Of that total, 92 were children, 60 were drivers, and 6 were adult passengers."

Wikipedia puts it at 175 fatalities between 1990 and 2000 (source: http://web.archive.org/web/20080226234316/http://www.nsc.org...). https://en.wikipedia.org/wiki/Airbag#Airbag_fatality_statist...

Also, most of the fatalities occurred from being sitting too close to the airbag or not wearing a safety belt. From the wikipedia source again: "These fatalities have predominantly occurred when the children and adults were positioned precariously close to the compartment where the air bag was housed. Most of the children killed were not secured by safety belts and were thrown forward during pre-crash breaking, placing their heads just inches from the air bag when it deployed. Therefore, the best defense during an air bag deployment is to be wearing a safety belt."

Thanks for the correction. It's too late for me to edit my comment, or I'd change it.

Regardless, my point isn't that airbags are a problem. They clearly save lives on balance.

The point is that car manufacturers are happy to add features to cars that buyers want even when those features might kill people. On the other end, manufacturers have removed critical safety features to save a dollar or less on production costs, and those decisions also have killed people. [1]

In this case, though, it's more like airbags, where, on balance, more lives are saved than lost as a result of the feature. At least in theory.

[1] https://en.wikipedia.org/wiki/Ford_Pinto#Fuel_tank_defect

Its 158 total since 2000 or so. Most of those deaths were children. That was before weight sensors were put in to disable the passenger side airbag and before parents were encourage to always put children in the back seat.

>automobile manufacturers just consider a certain amount of money paid out in liability a cost of doing business.

Actually for airbags its federal law and I believe liability issues there are handled by the federal government. I believe the government pre-empts "common law" liability here and can more or less wash its hands of it.

There's this great scene in Silicon Valley where one the characters gulps nervously before entering a self-driving car. I think there's going to be an element of fear here that's undeniable. I can't imagine riding in one at expressway speeds. Frankly, I think considering all the issues like weather, accidents, etc the self-driving car might be our generation's jetpack. Sure, sounds great on paper, but in reality its horrifying.

>Actually for airbags its federal law

And I said:

>>(though now they're mandatory in the US, IIRC)

You also say:

>liability issues there are handled by the federal government

Considering law firms exist that specialize in suing car companies over airbag accidents [1], I'm going to say [Citation Needed].

>Frankly, I think considering all the issues like weather, accidents, etc the self-driving car might be our generation's jetpack. Sure, sounds great on paper, but in reality its horrifying.

Cars with some of these features already exist, and are already on the road. Specifically auto-breaking collision avoidance, as well as auto-follow-distance cruise control (not to mention auto-parking).

These features are not only not horrifying, but they're starting to be standard on high-end cars. People will get used to them, and when the next innovation becomes commonplace, they will get used to it as well.

It's not only unlike jet-packs, it's almost inevitable at this point.

[1] http://airbag-law.com/

I thought this was going to be some sort of one-time monthly fee for an actual human driver.

> monthly fee for an actual human driver.

I can find someone to drive me around for a month for way less than 10K

Self driving cars have to deal with a human centric analog world. Why not add a layer of information to help them ? smart paint to help align, smart milestones and road signs broadcasting local driving constraints. Your car will read these and plan accordingly, instead of relying on heavy computer vision to be interpreted ?

Chicken-and-egg. It's expensive to add a bunch of stuff to the road for computer cars to read, and if there are no computer cars to read them, what's the point?

Computer cars will lower the bar on what's valuable to put into the road. I can see the point in 30 years where computer driven cars are even simpler than today, because it made so much sense to add the automatic road markers, which made automating cars easier, which make all cars automatic, which led to roads being updated only electronically, which is now a simpler problem than today (driving on roads with only other computer-driven cars actively broadcasting beacons of their location and position, and road markers/signs/lights being trivial for computers to read).

For $10k I'd get either e46 m3, s2k, rx8, or mx5 or a cobra.

My mx5's worth about $2000 so for $12k you could have a self driving one. Though there's not much roof space to bolt the thing on in an mx5.

seems like a fairly good bargain, life is far more precious than money. Spending on it will be worthwhile if it ensures safety

tl;dr

Q.A.

The homepage video play button doesn't work in Firefox (latest stable release)...

[imagining RP-1 driving me off a cliff the moment I venture onto an unsupported road]

Unfortunately a slight oversight. I thought mp4 was supported. Fixing now. Thanks!

And....for $1.50 you can put a brick on the gas pedal.

Now, this is missing many of the features of autonomous cars like the ones Google is developing and the driver will need to remain at the wheel and ready to take over in case of emergency...

(ok, let the down votes commence. I couldn't resist. It's great people are working on things, and I hope something comes of it but this seems like a case of all or nothing to me).

I think its the other way. Taking the plunge to 100% self driving is going to take a very long time. Google's systems seems to require elaborate 3d mapping of the entire road. In the poor states like where I live and where many people live on dirt roads that streetview has never seen (and some that don't even show up on Google maps), it is going to be a long time before fully automated driving is a proper 50 state solution.

This on the other hand is something that could show up in cars in 5 years, and provide meaningful benefits far before Google is ready.

I suppose maybe incremental progress is the way autonomous driving will happen, so maybe you are right. And, all progress is progress and it is good.

At the same time, I simply don't see much benefit and substantial risk in the system as presented. It works on one kind of car in one state. It only works under certain conditions. And it doesn't really do that much. Oh... and it seems a bit risky. I do understand the need to raise capital though, so if they sell some of these and it funds future development, great! I won't be buying one but hopefully someone does.

Incidentally, I saw a device similar to the "brick" for sale in a truck-stop once. It was an adjustable pole that pushed down on the gas pedal and locked into place so the truck driver didn't have to use his foot (big trucks don't have cruise control I guess?). It seemed a little iffy to me to and I can't believe it would be legal.

Reply on news.ycombinator.com