Hacker News new | more | comments | ask | show | jobs | submit login
Driverless Cars Tap the Brakes After Years of Hype (wsj.com)
45 points by mudil 34 days ago | hide | past | web | favorite | 68 comments

I wouldn't trust an autonomous vehicle at all unless it wasn't networked; in fact I've made sure that my car isn't networked at all due to the fact that autonomous or not they can be hacked by cutting power to everything except power steering, the radio, and the car outlets. Networking and control of the car being linked is asking for trouble, not to mention that somehow they fucked up the non-autonomous cars so that it's possible to hack those and drive them with a laptop too.

Link for the lazy: https://www.independent.co.uk/life-style/gadgets-and-tech/ne...

Having driven very new cars, rentals mostly, there are few compelling features that I can't just add to an older, not smart car.

And that is exactly what I plan on doing for some time yet. Frankly, the bright nav screens, menus and other gunk prove quite distracting. I really hate most of them. A few can be dimmed, or turned off, but none seem to remember that state, always eager to display some warning or other.

Seems to me, if it always needs a warning, I really don't need it in the car.

I feel we are way over hype on self-driving vehicles. The tech is advancing nicely enough, but the problem scope still seems very large relative to what we've solved for.

I guess I missed the predictions where people thought we'd have fully autonomous vehicles by 2019. It seems to me that we're on track for having something in a 5-10 year timeframe. Yeah, it won't be perfect, but it doesn't need to be. In the meantime we'll see more and more driver assist technologies which will slowly transition us to fully autonomous driving.

I don't have specific references, but yes, for several years I have been reading about how we are on the cusp of autonomous driving. For long enough that I got bored of discussing it maybe a year ago.

In support, this article says "Tesla Chief Executive Elon Musk in 2016 promised to demonstrate a vehicle traveling in fully autonomous mode from Los Angeles to New York by the end of 2017."

Is it possible you don't remember the predictions because you weren't reading about autonomous driving a couple years ago? It seems like trends and popular obsessions come and go very quickly these days. If you blink, you can miss it, like the rise and fall of bitcoin.

Or it could be that you were looking to techno-optimists for your predictions while others were looking to those that actually were involved in the tech.

I was a skeptic all along, because it seemed like the "techno-optimists" failed to grasp the difficulty of writing software in general. I'm sure the people that engaged in most online discussions that I remember were not experts.

However, I don't see why you point to people actually working on self-driving cars as if they would have been pessimists - why work on such things if they didn't believe it was feasible?

He is a techno-optimist, but Elon Musk is not involved in the tech?

Like Musk?

And Waymo promising full autonomous cars by last year, even if they limited the scope in space and weather they missed it

Yea in 2015, I distinctly remember people were predicting a 5-10 year "aggressive" timeline [1] with most AI peeps citing 10-20 years (I'm discounting Muskian bravado here messing up my averages).

We are about on track on that timeline I'd say. As usual, the truth is a lot more boring than most people predicted. But WSJ gotta have something to write about I guess.

[1] https://www.forbes.com/sites/federicoguerrini/2015/01/31/tec...

We are almost certainly not on track for autonomous vehicles.

The simple test for autonomous vehicles is “can they drive in Indian traffic.”

We are very likely, on track for a correction in the form of “ we never meant level 5 autonomy.” We will see some form of assisted driving, and nothing beyond.

It’s concerning that people aren’t thinking of attackers (or the govt) taking over your car first on HN.

This is the happy start of AI, Just like the happy start of the web.

Those were the days before the RIAA, before phishing, five eyes, and the unending loss of innocence that it has been since then.

>It’s concerning that people aren’t thinking of attackers (or the govt) taking over your car first on HN.


I am quite concerned at how capable newer cars are, sans an equal level of security and safety consideration applied.

There are going to be some ugly things happen, then we will see a correction.

For now? No way I plan on owning anything networked, drive by wire.

And so many things are coming long before we get great self-driving vehicles. In fact, I think they will be requirements to ease the transition to them, due to the reduced complexity when humans are not part of the equation.

"We have geofenced your car, due to erratic, incompatible, [insert irritating justification here], driving patterns."

"Your car has been speed limited for your safety..."

Yeah. Gonna be fun.

And the argument FOR such controls writes itself!

Do you know how many people die to drunk driving?

Rash driving?

Accidents in general?

Imagine if we could stop all that? Those millions of lives all over the world, saved.

Who could stand against such a moral argument and defend the right for people to act rashly while piloting tonnes of moving metal every day?

If I thought for a minute those things would not be abused, count me in.

"GM says it will launch a robot taxi service in 2019" - 11/30/17


"GM Hitting Bumps on $5 Billion Road to Driverless Cars by 2019" - 10/25/2018


The hype has definitely been there, and on HN too. I can't find it now, but there was a great Twitter thread around New Year of someone calling in a bet they'd made with an SV figure about self driving Lyfts by 2019.

As an observer of this development, this is not correct.I recollect press releases deliberately ignoring the big elephants in the room. As an outsider relying completely on press, I think they lied by omission.

When self driving was first demonstrated to work, constraints were never highlighted. For example I did not see any discussion on the topics like black ice, needing a sufficiently accurate area map, proper road marking etc. in advance were never in new. Humans might struggle but can still drive in those conditions.

On the issue of driver assistance, I believe more of might give drives a sense of false safety and hence cause more accidents.

This! So much this!

I remember a interview by Elon musk at a conference called CODE where he espoused that driverless cars are solved because all you need to do is get a car to stay in Lane and brake safely at low speeds. And everybody bought it.

Well, where is the cross country summon, coast to coast autonomous drive, driverless taxi networks. It astounds me that one can blatantly lie repeatedly and not be called out.

I remember in 2017 being pretty convinced that we'd have something commercially operating within 5 years and my wife being skeptical. As recently as a few months ago, I still felt pretty good about that prediction. WayMo's non-launch launch has made me a lot less bullish, since I really had assumed they were almost there, at least in terms of Phoenix. And 2022 is looking a whole lot closer these days.

Bits and pieces like trucking or taxis in relatively easy cities have been expected very near-term by a lot of companies for a while. I don't think anyone has said manually driven cars will be entirely off the road at a given point in time. Even the bits, though, should have a hugely disruptive impact on their markets.

As soon as a few months ago even mentioning here that you didn’t think autonomous cars were close to being ready got downvotes. People don’t understand how complicated driving a car in all conditions is and how much the human brain is doing already.

When you have car companies throwing about the name AutoPilot you can't blame consumers for thinking it's coming soon.

There are tons of technical problems but the fundamental issue is our expectations.

Driverless cars can’t just respond as well as a human would have - they have to be better.

They have to “know” the black ice was there. Or the pothole was there. Or the construction was there. And then they have to respond “reasonably”, which is tough because computers are fundamentally “reasonless” machines at the moment.

If our expectations were driving during the day on a few defined streets in nice weather, perhaps going a bit slower and with the understanding that things may get a little jerky, I think we would be ok.

In other words...we need to give self-driving cars permission to be like we once were: not amazing at driving.

We need to give the cars a learners permit.

Yep, you are talking about experience; robots only need to learn once and it spreads to everyone instantly. That can’t be said about humans, everyone has to learn it at their own pace.

Humans aren’t that good at driving either and most humans won’t know what to do with black ice either at first. I certainly didn’t, my driving school didn’t even talk about winter driving either. I had to learn on my own and it wasn’t “great”. Nothing like spinning out on ice to know never to do that again.

I’d trust robots right now more than drunk or high drivers though.

Tesla has a massive fleet gathering data on every mile driven by humans and they are using that data to build their neural networks. The AI is getting better every second.

That’s why they are deploying driver assitent technology now and human co-pilots for any cars with autonomous driving capability.

Unlike with humans, the learning is transferable with driverless cars.

The drivers provide instruction in a form that will be retained by the software?

Yes, via recording and labeling what the human does when it’s not engaged and during any take overs when it is engaged.

People often say self driving cars have to be better than humans but they often neglect to ask "what % of the population does it have to be better than?".

Today's self driving cars are already better than some drivers. You don't have to look farther than r/IdiotsInCars. I would say that if self driving cars are better than 50% of the population, that would be good enough. Removing most of the accident prone individuals would be greatly beneficial already.

I’m sure it will be 5 to 10 years away 5 to 10 years from now, just as they were saying 5 to 10 years ago.

I do think it will happen someday but this is an insanely complex problem with enormous danger and risk. I am most impressed with the AI hucksters who convinced investors that the tech was close and then commanded 7 figure salaries to work on it. An impressive feat!

I have never believed that we will get fully autonomus cars suddenly. But each year cars will be able to do more on their own. It will be a gradual transition. Where the driver's will let it park on it's own, drive on the freeways/highway on it's own. To allowing it to go from the highway to home or work on it's own on a regularly used routes. I feel people that dismiss autonomuos cars completely and people expecting them to be perfect in the next few years are both wrong.

Driverless cars are a much harder problem than driverless trucks.

Once the latter become day-to-day reality on highways, I'll become cautiously optimistic about autonomous cars in cities.

I've read somewhere that driverless trucks are pretty much as hard as driverless anything, because there's always that last mile in a city. It's impossible for the ride to be 100% automated, and anything less is pointless at the moment.

I think we'll need some extremely standardized "pre-last-mile" highway exits/transfer points across the whole country to have the beginnings of a new solution. And I doubt the Teamsters will be onboard with that.

> ...there's always that last mile in a city

But what if that's the only mile driven by a human? Trucks drive autonomously on highways in good conditions, then a human takes over for the last mile?

One issue is current sensor range. A truck on the highway has a way longer stopping distance than a car and requires sensors that are not quite there yet: sure radar could see obstacle that far but the lane detection necessary to distinguish a car parked on the line from a car on the shoulder can't yet see that far

As long as they both share the same road and drive in the same traffic it doesn't make sense as to how can one be a harder problem than the other.

In fact I never want a driverless 80,000lb 18 wheeler anywhere near me when I'm driving. A 5000lb driverless car is scary enough.

There will be, and have been, people on both sides of the SDC debate, bulls and bears, for years going back and forth. There are good talking-points for bulls in that the technology improves yearly and there are local roll-outs, there are great talking points for bears in delayed national-scale roll-outs, and taking on the overoptimistic estimates of years past.

What astounded me talking to people working in self-driving car start-ups and spin-offs from incumbents, is that how little they care about simulation. The last accident from Uber confirmed that yet again.

The fact is absolutely difficult to comprehend, when Waymo has allegely done a huge amount of simulation.

they should focus on mastering mundane routine driving and have a backup to switch to remote controlled operation for risky or unusual situations.

I am not convinced you understand this problem space well. The very notion of defining two such states is extremely elusive.

Might want to look into Toyota Guardian before dismissing the approach.

As it's exactly what they are planning to roll out in the short to medium term.

what's elusive about it? there's some pretty obvious situations where you expect driverless cars to have trouble: bad weather, unusual traffic or pedestrian patterns, emergency vehicles, rarely travelled roadways...

Outside of a few limited areas, remote control driving isn't a viable option due to the unreliability of cellular data networks.

making a safety-grade data network would be an interesting challenge. I do think it is a possible thing to do over limited areas, and I think that if we had defined areas that had safety-rated networks, that's a thing that self-driving cars could plan around... e.g. if you request a trip outside of those areas, the self driving car would take you as far as it could and then park, allowing you to prepare to drive manually.

Almost all technical problems of self-driving cars are solved by current algorithms once we get 100x faster HW, i.e. processing at 1000 fps instead of 10, making things like tracking, collision avoidance, vision, etc. trivial; the only problem left is path planning (down to inch resolution with dynamic vehicle models), which is coincidentally NP-hard and there is very little hope to get it to human level in reasonable time (Waymo is still orders of magnitude worse than an average human).

>Almost all technical problems of self-driving cars are solved by current algorithms once we get 100x faster HW,

That sounds off to me, because my understanding is that the computer is a small portion of the cost and the mass of the self-driving car. If you could solve the problem by just filling up the back seats with computers, someone would have prototyped it, and probably even brought it to market as a way to drive large vehicles.

(I mean, unless you were on some non-parallelizable process, but my understanding is that most image processing tasks are massively parallel.)

Driverless cars are IMHO a classic example of technology looking for a problem to solve. Improving road safety, the biggest reason for them (again IMHO) can be achieved with existing technology by implementing smart roads with sensors and cameras as well as vehicle datalinks which can predict and warn when a car is about to blow through an intersection in front of you or whatever, and instantly alert authorities when a driver speeds or is driving erratically (you might call this dystopian, but I would argue you have more freedom here than with an autonomous car you literally cant drive over the speed limit). The second major reason is automation, which I agree is nice, but if you analyze the cost of trucking [0] you see that drivers wages and benefits account for only 20%-40% of the operating costs in the trucking industry. If you use the typical startup cliche that a 10x improvement is necessary to really change a market, the economics of driverless vehicles just dont add up. It seems like a great idea to rich east-coasters who wish they could work on their daily commute, but outside of certain bubbles I don't think they make any sense. IMHO.

0 - http://atri-online.org/wp-content/uploads/2017/10/ATRI-Opera...

Oh c'mon, let's not be silly. One thing is to say it's unpractical to do now, or maybe ever, but the problem is very real. Having the ability for a vehicle to drive itself reliably on normal roads would have a transformational effect on society, first of all the end of car ownership as we have it today.

> first of all the end of car ownership as we have it today.

I see the idea thrown around a lot that self-driving cars will make us all use taxis because, absent human drivers, the taxis will be far cheaper than owning a car -- but I have yet to see strong analysis on this claim. There's lots of issues here:

* inconvenience: Unless the network is extremely dense, this means waiting for a taxi vs. driving myself. Denser networks, however, all things equal, mean lower utilization.

* storage: All sorts of people need to store things in their car. Contractors with tools, parents of toddlers with car seats, etc. Taxis prove to be too inconvenient, leading this population to prefer owning (hurting network density)

* traffic patterns: Commute patterns are heavily one-way, meaning in practice, after dropping someone off, the car will need to return without a person. Say you have a core commute time of 2.5 hours and take people on 30 minute journeys; you've managed to turn 3 cars into 1 [1], but at the cost of increasing miles by 60%. It's somewhat questionable if such a conversion has positive ROI.

* existing contrary evidence: You see plenty of private car ownership in areas with cheap labor (Chinese cities come to mind). Even cheap taxis aren't sufficiently solving transit needs.

[1] Yes, you could also pick up multiple people.. but now you introduce inconvenience again!

But of you tackle the problem first, and not the technology, it makes more sense to just build automated road systems. Dumb pod-like cars which are driven by roadside controllers. If we decide to go down this route it can be achieved with existing technology, today. This is the ultimate scenario most people imagine with self-driving cars anyway, a future where hardly any of the cars on the road are driven by humans. We should just go straight there.

That would be nice, but it would also require infrastructure investment on a massive scale before it could even begin. Self-driving cars as currently conceived function on existing infrastructure with zero changes.

But cars are already broken: see LA, SF, NYC. Why double down rather than invest in infrastructure?

I personally agree, but most people like cars as they are now, even in SF and LA; only in NYC is public transit really good enough that there is a critical mass of non-car owners voting to keep up the public transit.

I live in silicon valley, and the traffic isn't SF or LA bad yet, but it's pretty bad, and it's also clear that the voters around here aren't prepared to vote in usefu public transit; (one could make the argument that VTA is proof that voters are prepared to pay for public transit but aren't willing to allow useful public transit)

The upshot is that most people outside of NYC don't think automotive transportation is broken. Most people prefer automotive transportation. I personally am with you; driving when you could be on transit is nuts, but I'm clearly in the minority.

Because being easily deployable is a very important part of a technology.

Only to short term business interests. Long term the costs of cars to society are absolutely abysmal compared to bulk transit.

It's a real estate issue, not a technology issue. Dumb pod cars would have to travel on separate roads, and there's no space to build a whole second road system.

The idea that I am forced to drive the car myself doesn’t even crack the top 1000 problems worth solving. Transformational is not always good, and is often a terrible argument for prioritizing it.

No, this looks like SV companies drinking their own coolaid and ignoring actual world problems they can’t get rich off of.

The problem isn't so much the act of driving, though society would likely see some benefits from that time being used more productively (even if for leisure).

It's that the car is tied to you.

Everywhere you go, it needs a place to stay while you're doing anything outside your car. Parking spots average $5,000 per surface space, $20,000 per space in an above-ground parking structure, and $32,0000 underground. Every moment your car sits in one, it depreciates without performing useful work.

This is the argument that always gets made, but I just don’t buy it. The fact of the matter is that you will always need a ton more cars during rush hour, and those cars have to be stationed somewhere during non peak hours. So no matter what, you’ve got cars needing to be parked in high density areas.

That's a potentially very interesting perspective to me.

My point of view would be that Americans are spending on average about an hour every day driving. That's an hour that could very well be spent doing any number of more interesting and/or productive tasks. How is that not a high priority problem?

Because it's hard to recover those hours entirely unless you just plain don't get into a car.

Rides are often too short to context switch into doing anything useful.

Even then there are all sorts of situations today where people are being transported in a car not requiring focus -- but aren't effectively recovering this time:

* Riding in a Lyft: I suppose in practice someone could whip out a laptop, but you rarely see it happen. Rides too short, induces carsickness, etc.

* Riding in a vehicle that is L2 autonomous. While not legal or particularly safe, in practice, someone driving an Autopiloting Tesla can read and respond to emails. (i.e. similar situation as being in a Lyft)

> Rides are often too short to context switch into doing anything useful.

People on public transit usually use their newspaper / book / phone. Sometimes they miss their stop because they are too involved doing that. Have you used SF<->Bay transit on Google buses or something like that? From my experience it's pretty comfortable and productive to use laptops during ride to check tons of morning emails and it saves a lot of on-campus time.

> where people are being transported in a car not requiring focus -- but aren't effectively recovering this time:

Your examples require some kind of attention to driver or focus on road situation:

> Riding in a Lyft: I suppose in practice someone could whip out a laptop, but you rarely see it happen.

Riding alone vs riding with human driver differs a lot, especially for introvert people (e.g. talking in company of friends is much more comfortable then talking in company of strangers). It's hard to ignore presence of other human during ride.

> Riding in a vehicle that is L2 autonomous. While not legal or particularly safe, in practice, someone driving an Autopiloting Tesla can read and respond to emails. (i.e. similar situation as being in a Lyft)

You have to use your peripheral vision to control situation on the road. You can't fully concentrate on your work and be productive.

I dunno, I’d say people probably worry more about health care than their optimal productivity.

Driverless cars are IMHO a classic example of technology looking for a problem to solve.

My car’s utilization rate is about 8%. The remaining 92% of the time it’s sitting idle and since I work in a metropolitan area I even get to pay $190 per month to have it sit idle in a parking garage while I work.

My car is an enormous expense. It’s my second largest expense outside of my house. And it sits idle 92% of the time. I pay $900 per year to insure it, I pay to register it, get it inspected, pay for maintenance, depreciation, etc.

Autonomous cars run by service companies like Lyft and Uber will dramatically reduce costs for anyone driving their own cae today. Eventually overall car utilization will flip from 8% utilization to something like 90% utilization making transportation drastically more affordable.

And like you said, you get to work (or sleep or FaceTime with your kids) on your commute.

I dream of the time when I go on a road trip with a family and don’t have to drive. No, train, etc is not an option for too many reasons to list here

You can effectively do this today with high-end SAE L1.5 systems like Cadillac Super Cruise and Volvo Pilot Assist.

We really are on the cusp of L2 highway autonomy, and will get there in 2020 as originally envisioned. This WSJ article is insufficiently researched.

You can already hire a human driver.

That table stops at 2016 with driver comp at 43%, and the share was still growing at 3%/yr, and 43% savings is already a lot, so I'm not sure why you're writing labor savings off.

There's also efficiency gains to be had since computers don't need sleep, could have reduced insurance premiums, etc.

I also expect that with lighter vehicles, the cost of labor is an even higher percentage of total costs.

"Only" 20-40% of the operating cost saved would be enough to completely obliterate the human trucking industry. Amazon didn't have to beat the prices of other stores by anywhere near that much to become 10x or 100x bigger, they only had to be a little bit (but consistently) cheaper and more convenient.

  Driverless cars are IMHO a classic example of technology looking for a problem to solve
The "problem" they are trying to "solve" is driver wages.

Applications are open for YC Summer 2019

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact