Uber’s Vision of Self-Driving Cars Begins to Blur | Hacker News Reader

Uber’s Vision of Self-Driving Cars Begins to Blur

nytimes.com

128 pointsplasticchris156 comments

I have no idea what's going on internally at Uber ATG, but they weren't doing well before the killing of Elaine Herzberg. There was a pattern amongst Autonomous driving outfits funded by well-heeled and impatient companies jumping in to the game late, going balls-to-the-wall, and mostly just wasting a lot of cash and resources on a gigantic clusterfuck. We can include Apple, Baidu and Uber in this category. Baidu has since done a reboot. I have no idea where Apple is at in terms of tangible progress, and then there's Uber.

The companies doing well are Waymo, Cruise, and Zoox, and one thing common to all three of these is that they started small a few talented people and scaled up in stages when it appropriate.

A testament to the whole 'just because 1 woman can make a baby in 9 months doesn't mean 9 women can make a baby in 1 month' analogy.

Very true.

The over-exuberant business version of this is money-men types who see companies as a collection of investments and cash flows. This leads them to think of business problems in "resource allocation" terms. If company X's is bad at UI, customer service or whatnot than this will be fixed by "investing" more in it.

Here they see a big juicy prize: first to market with self driving taxis. They understand that it's risky but they still assume a very strong correlation between the amount of money going in, and the probability of their big juicy prize coming out.

The whole approach is a bad idea in new technology. Instead of thinking in expiremental, creative-discovery terms, they have a very precise destination and they try to brute force a way to that destination with money.

Yes. I suppose in the scope of industrial process there are basically two types of action: shoveling (a scalable technique that is understood and can be trained) and invention (creating new scalable techniques). If it's just shoveling, capital investment may very well yield higher output.

Invention on the other hand happens on a constrained pipeline.

The shoveling that looks like invention is when an existing invention is adapted (and sometimes called innovation). I presume it's hard for outsiders to say when a subject on which expertise is not yet commoditized is ready for shoveling or not.

I think this is what happened here - since everyone were hyping the AI revolution, they presumed the technique was ready for industrial adaption and only lacked in dedicated subject experts and existing processes which can be adapted. They presumed there was some core invention that could be adapted and monetized.

Given how poorly understood subject AI is, and how lucrative a self driving car would probably be, some risk taking in this area is certainly understandable. I have no idea if the investments were in line with the risks and rewards or not.

It worked pretty well for the Manhattan Project and the Apollo missions. I think that you're romanticizing technological progress a bit if you don't believe that more money (often) leads to more progress.

Both of your examples where government-run "monopolies" - that is, if you were a physics or rocketry expert, you didn't have much of a choice of which organization to join in order to build an A-bomb or rocket.

And it's likely you got kicked out of the competition's country for being Jewish. America got super lucky.

I dunno if I'm romaticizing. A massive mission/goal driven exercise can be quite romatic. Tesla did that, and I'd call that more romantic than (for eg) FB's path.

Good points though. Tons of resources, people working in secret, and a total absence of "market feedback" have yielded big, impressive results before.

I think their point is that you can't just buy progress on demand.

And I think the counterpoint offered was that the Mamhattan Project and Apollo programs did just that. Note that in both cases you had government-deep pockets and global sociopolitical ramifications as team motivators-a potent combination.

No, they set up a multi-year governmental project. It's rather different from throwing money at a c-suite executive and telling them to hire developers.

those where quite different projects, having to surmount lot of known unknowns. self driving is still fraught with unknowns unknowns. once the problem gets mapped out money will enable company to power trough.

Smells like hindsight bias.

not really. both those indicated projects had all the required component worked out already, plus the physics at their basis where understood. there was a working reactor and the governing phenomena was well understood. it was mostly a refinement and tuning process to get it right. same with apollo, at that point nasa had been launching people in space for a decade, working out the docking procedures and capsule operations.

contrast with deep neural network, there are no models to predict overfitting or undertraining beyond what training cross validation say. hardly the same situation. there's no model that can predict a network confidence and network output as a classifier is still very rough even before it gets converted into car commands (i.e. see Uber report)

those cited project had a theoretical framework on which they were built upon. neural network are still rooted in 'tune them if they don't work' stage.

It only works when you're talking massive, government-backed support for basic scientific and technological research. The tech industry likes to lean on this narrative, but the reality is they just exploit the gains of basic research and are probably reaching the limit of what can be drawn from the last time this happened (i.e. when governments created the internet). This is why you see tech shifting into so much rent-seeking nonsense and bizarre financialization strategies that mirror Wall St/the banking industry. They're scraping the bottom of the government-financed research well and don't have much actual ability to create fundamental technological innovation themselves.

The good news is all we have to do is shift back to massive government investments in basic research to course correct and compete with entities like China(who seem to have a clearer picture of how Capitalism actually works in practice).

Part of the reason why companies aren't simple functions of cash in -> results out is because the ability to produce results is 100% dependent on hiring the right people (who may not even exist for your problem domain). Cash only helps in so much as possibly being enough money to hire those people, but it says nothing about whether management will effectively exploit the human resources it hires, i.e. whether management will shut up and get out of the way or whether the "right people" will be hobbled by infighting, corporate politics, and "policy".

Until investors can come up with a way of measuring those confounding human factors, just plowing money into companies and praying for a miracle will continue to be a debased and ineffective investment strategy.

Of the names you call out for doing well though, it's worth pointing out that while Zoox may have so far avoided "wasting a lot of cash and resources on a gigantic clusterfuck", it's apparently their plan to explore said clusterfuck over the long-term. Zoox's stated goal, once/assuming they get the self-driving software right, is to redesign and then build a completely new concept of what a car is, from the wheels up, and then own those cars, operating an Uber-style self-driving cab service.

This is probably the most capital-intensive business model you could possibly come up with. [edit] Which isn't to say they'll fail! I know nothing about their progress on whatever internal research is going on. But in terms of including as many possible points of failure and guaranteeing that in the future you'll need an absolutely gargantuan funding round to succeed, this is tough to beat.

It's a miracle Zoox has gotten as far as they have. Their vision is crazy, but even if their prototype robotaxi concept crashes and burns, their autonomous OS is still very good, and I believe worth their (speculative) $3.2 billion valuation. Others have spent more and accomplished less.

Don't forget the companies who are ahead of this, but do not openly talk about it. These are the companies from the automotive business. There is Daimler who started that in the 1990s and incrementally releasing new features of this development for the general public as part of the drive assistants. Other OEMs are doing the same. A huge Tier1 supplier here is also MobilEye who started with BMW and now counts Volvo and VW among their partners/customers with a short intervention at Tesla.

Mobileye was and still is a leader in ADAS, but when it comes to L4, Amnon Shasua has been going around talking a big game for years, but seriously, where's the beef? They're all sizzle and no steak.

https://arstechnica.com/cars/2018/05/intels-mobileye-wants-t...

I posted this a while ago

    https://news.ycombinator.com/item?id=14317214#14317443

> 1) Drop self driving cars completely. YOu aren't getting there first, second, or anywhere close to third, just partner with a car company and call it a day.

> 2) Settle Google's lawsuit, hopefully 1 will help

> 3) Hire a new CEO, Sheryl Sandberg is almost certainly not available but someone who can show that change will and is happening internally.

They've completed 2 of the 3 things I thought they needed to go public. Dropping their own self driving car program was the third.

They need to have their story figured out when they go public as investors will want to know if they are all in or all out on developing their own self driving cars.

I can't see them going public with their current strategy for self driving cars, its a huge money pit with not much to show for it currently.

Well their huge valuation hinges upon them eventually going into self-driving cars. It'd otherwise be hard to justify a $120billion IPO without rapid growth targets. They've already reached saturation points in most of the important metros and cities.

Do they need to develop the technology themselves? So long as their cars don’t have drivers, they’re fine, no? Or is the issue that nobody would share with them?

Uber probably views it as important that they develop the self-driving tech because otherwise (if someone else develops it, licenses it to OEMs, and then Uber uses those cars), there's no lock-in to the Uber ecosystem. Either those OEMs will partner with Uber for a fee that will capture most of the profit Uber's valuation currently expects, the OEMs simply spin up their own Uber-style vehicle on demand program (far cheaper than Uber spinning up a car company), or people who can afford their own vehicles (likely a huge chunk of Uber's pool of users currently) just use their own now-autonomous vehicle a la carte, summoning it when they need a ride.

In a pretty ironic twist, Uber et al have already started rumblings about how the third option needs to be made illegal, because there's no regulatory oversight on vehicle maintenance and safety. Having spent most of their life ignoring existing regulations, they soon may need them to survive.

Couldn’t the lock-in be the Network of riders?

Uber needs self-driving tech to be cheap and available.

In the long run this seems likely, but in the short term Waymo seems to be gearing up to start their own app, and if they have a cost advantage over Uber, Uber will be burning even more cash subsidizing rides waiting for self-driving tech to become a commodity.

At that point, why invest in Uber?

The problem is they need a story like self-driving cars to maintain the (inflated) valuation long enough for existing investors to get out. Taxi rides alone are not nearly enough.

I think they are aware that they don't have a chance at it long-term. A partnership isn't as sexy, but might be enough to eek out a $50B valuation.

The problem is they need a story like self-driving cars to maintain the (inflated) valuation long enough for existing investors to get out. Taxi rides alone are not nearly enough.

Exactly. Uber still loses money on every ride.[1] They can't fix the losses without raising prices, which would lose their competitive edge. They've squeezed the drivers as hard as they can, and the drivers are pushing back successfully at last.[2] So Uber needs something they can hype. They already tried "China", and failed. "Self-driving" was next. That's not working out. Uber is getting into bike/scooter rental and food delivery, which are already crowded, low-margin industries.

Self-driving didn't need to work, it just had to be hyped for investors.

[1] http://fortune.com/2018/08/15/travis-kalanick-uber-still-pos... [2] http://fortune.com/2018/08/08/new-york-freeze-ride-sharing-v...

>Exactly. Uber still loses money on every ride.

Please stop repeating this, or at least find a source that supports it. There's a big difference between "losing money in the aggregate" and losing money on each ride; the latter is only true if the marginal costs of providing each ride are greater than the revenues from it, and the source in [1] only reports the former kind of loss. It's very unlikely that each ride is actually costing them the ~$4 they take from it; that would be some pretty inefficient IT.

find a source that supports it.

[1] https://news.crunchbase.com/news/understanding-uber-loses-mo...

[2] https://www.cbsnews.com/news/uber-valued-at-62-billion-still...

Those just repeat the original source, which (uncontroversially) says they lose money in the aggregate. Total operating costs minus total revenues is positive.

None of them claim, as you did, and as I asked for, that each ride loses them money. The closest they Come is saying that some rides can lose money because of the formula.

Do you understand the difference between “lose money on each ride” and “lose money overall”? All of your links only speak to the latter, not the former, and the former is what I was objecting to.

I think autonomy is existential for Uber, but I also think they have a decade or so before it will start having a deleterious impact on their bottom line.

My thesis is that they shouldn't shudder ATG entirely, but should scale it back to a core group, and then consider dropping the dead weight and scaling it back up again slowly as certain performance milestones are met, meanwhile keeping a patient eye out for promising startups worthy of mid-stage investments sizeable enough to grant them an significant ownership stake.

*shutter

>The problem is they need a story like self-driving cars to maintain the (inflated) valuation long enough for existing investors to get out. Taxi rides alone are not nearly enough.

This is pretty much it. Making an Uber clone is absolutely trivial, and they are being outcompeted in every single international market which has one. They also have zero customer loyalty, people will happily switch to another app the moment there is competition.

It seems like it would be difficult to clone their terrible GPS usage and mapping. Most drivers use Google maps in my experience.

And now with Uber express pool it is pretty much guaranteed it will tell you the wrong location. If it says northeast corner, just go to the south west corner.

> Making an Uber clone is absolutely trivial

Maybe so but cloning Uber is absolutely not trivial.

I love your naive hero worship. Suddenly you get Sheryl Sandberg on and all your problems are solved. Hint: Sandberg is not magician and not even that competent all rounder or polymath. She was recruited in FB because she precisely knew how ad business works and brought huge network of big players in this space with her. In a way its backstabbing to your previous employer but well who cares. Another thing to think about: Its almost exactly 10 years since Sandberg is at FB and its still one trick pony. In fact, much worse than one trick pony that Google is.

> she precisely knew how ad business works and brought huge network of big players in this space with her. In a way its backstabbing to your previous employer but well who cares.

How is this backstabbing? Nobody is owned by their employer for life. People change jobs and bring customers with them to their new employer in all types of industries.

I would never trust an Uber backed autonomous car system. They should have been shut down after it came out that they were cutting corners on safety and tech to get the program as far along as they could before it caught up with them. I can't believe Uber still exists at all with the negativity and corruption associated with the brand.

> I would never trust an Uber backed autonomous car system.

I agree, and it makes me wonder how this is going to shake out. In the automotive industry it's very common to have the tier 1s provide LOTS of tech to the OEMs, but it's typically isn't branded. Interiors, infotainment, brakes, etc. It feels like self-driving subsystems will be the same thing.

So, I might not trust Uber-branded systems, but I might trust Apple or Microsoft (Just to use some common brands as placeholders). Would we trust those systems more than the OEM-branded systems?

Uber seem to have two really key problems:

The first is that self-driving is super difficult, they don't really have the expertise, and a lot of their progress seems to have come from being able disregard proper safety procedures. So to move forward with it they'd need to fess up to investors that it'll take much longer than anticipated and it'll be much more expensive. That'll damage the company value significantly, and it's not really relevant to what the core of the business is doing right now. I actually find it fascinating - Uber has built an app that disrupts the traditional taxi marketplace. Separate to that they've got a division working on a produce to disrupt Uber's current market place.

The second problem is that if they choose not to do autonomous driving their entire business proposition needs re-establishing. Can they actually make money doing what they're currently doing? Or are they doomed to sink huge venture capital sums into acquiring market shares, only to fail to reach a dominant enough position to actually raise prices and make bank. And part two to that question: Can they achieve that profitability and a good enough return to be worthwhile for investors before someone who does succeed in disrupting the taxi business with self-driving cars.

From the article:

>> Uber first made its interest in self-driving cars public when it hired about 40 researchers and scientists from the National Robotics Engineering Center at Carnegie Mellon University in 2015.

It doesn't sound like "they don't really have the expertise" (per your comment).

Perhaps autonomous driving is even harder than press releases from other, more cautious companies, have led us to believe?

One problem I can see is that they're trying to implement autonomous cars too fast, and secondarily they're trying to replace Uber X with them. If they treated auto-cars the same way they treated Uber Black, and make it a more luxury choice with a higher price, they could slowly implement this into their business over time, one car at a time. I personally would pay more for an auto-car. But the big crux is your first point, and the solution in my mind is for them to just be way way slower and ensure full safety precautions. I don't think they have to dismantle the division nor do I think autonomous cars are out of the realm of possibility.

Why would one pay a higher price on a driverless ride?

I mean if the data says they're safer on average, I'd pay a bit more? And it'd do away with loud blasting of the driver's favorite genre in the car, turning off the AC and rolling down the windows when I'd prefer it on, aggressive acceleration, etc. that you see with some UberX drivers.

If it was much safer, I would.

They've allowed (or even actively encouraged) this narrative to develop around how they'll make money as soon as they have self-driving cars. This requires not just the availability of self-driving on some small scale in specific places or along part of a driving route but broad availability in urban areas where a lot of people live (i.e. where self-driving is especially difficult).

Admitting the tech is a ways out means either 1.) Admitting that they have no idea how they're going to make money or 2.) Admitting that they're going to have to significantly raise prices to both improve per-mile profitability and cover the inevitable associated volume dropoff.

The only real explanation I have for why Uber hasn't accepted the inevitable and jacked up rates is that so many people are feeding at the trough no one wants to be the one to admit that the emperor has no clothes and there's no magic fix for making money at the current rate structure.

>The first is that self-driving is super difficult, they don't really have the expertise

What about all of the talent they hired from CMU?

Self driving cars are not going to happen in established cities for a very, very long time.

The only thing that could make it happen is a giant like Google/Alphabet investing enough cash to lobby governments to repurpose existing transportation infra to only belong to self-driving cars. But that's also the doorway to a dystopian future where megacorps run entire countries.

By the time you're building self-driving car-exclusive tunnels and the like, you really are getting very little benefit over a train system to justify doing it, especially when you take emissions into account.

I agree. Anything with low-occupancy vehicles feels like a dead end to me. Of all things, car ownership is the thing we're aggressively enabling/optimizing in 2018? Really?

I'm usually not a person who's sympathetic to complaints about advances in technology eliminating jobs. Not for a real overall long term gain to society. But autonomous cars putting swaths of people out of work while at the same time taking us in the wrong direction on traffic, emissions, etc. doesn't feel great. Those drivers losing work slowly over time as cities evolve truly better and more sustainable transportation options feels more right.

Aside from the opportunity cost of not furthering public transportation, autonomous cars will drive around aimlessly anytime the cost of fuel is less than the cost of parking. We already have Uber creating waste by idling/driving around between passengers.

car ownership is the thing we're aggressively enabling/optimizing in 2018? Really?

On the contrary: making taxis cheaper (which self-driving can do, by eliminating the biggest cost - the driver) makes it much easier to live without owning a car.

If there was no Uber and such, I'd probably have to own a car for the exceptions not covered by public transport. And after paying for the fixed costs anyway, the marginal cost per trip is low, making me more likely to use it over PT.

autonomous cars will drive around aimlessly anytime the cost of fuel is less than the cost of parking

Per the above, cheaper taxis → fewer owned cars → more free parking spaces.

Who happens to own the car, the driver or a service provider, is inconsequential in assessing the big picture consequences of more cars on the road vs. more sustainable options.

But it's not more cars, that's my point.

Say public transport covers 95% of my needs. If I have cheap taxis, I'll use public transport most of the time, and cars only 5%.

If I have to buy a car because I can't afford those taxi rides, I might use it for 30% or 40% or more of the trips, since it just costs me a bit of gas (or electricity) - the fixed costs are sunk.

I mean, the incentives are also kind of backwards -- it's usually free to drive and you have to pay to ride the train. If we're really looking at climate cataclysm it might be worth thinking about changing that dynamic.

It's "free" to drive in the sense that you don't have to take money out of your wallet to get into the driver seat of your car, yes. But that doesn't account for:

  - The monthly payment on the car (if applicable)
  - Liability / collision insurance
  - Gas
  - Tolls
  - Parking
  - Repairs
  - Washes

With this in mind, public transportation is obviously way cheaper than driving – and exclusively using Uber may very well be in some cities, too.

But yes, I'd agree with you that many people don't seem to account for those hidden costs when justifying driving over taking public transit in a city that adequately supports it. Hopefully one day soon that will change.

Assuming you have a car (which is more or less necessary in most places you might live in this country), for any given trip it's very close to free to drive while the train is more expensive. And if you're thinking about parking and riding, between the higher cost of those tickets and the parking fee it's actually pretty expensive to do and driving all the way into the city is definitely cheaper.

Right. We already have a great solution to many of the autonomous vehicle problems. Trains and busses.

As a European living in the states. Trains and buses are not a substitute for a car, not even in Europe.

In general, no, but generalizing can miss important local markets. I live in London and trains and busses do me fine. I honestly can't remember the last time I got into a car here; certainly years rather than months.

Big-city issues like parking and congestion charging can tip the scales quite a bit, with much better public transport than the rest of the country to compensate.

Trains also do not have to completely supplant cars to make a difference. If people take the train to work and use their cars to run errands around town it's still an improvement over driving everywhere.

How many Americans drive to and from an office in a city center at the same predictable hours every single day? We're certainly underutilizing trains.

Even shared bus services can made into something desirable with the right marketing and investment. The large tech companies provide luxuary coach services with WiFi to their workers and the people riding them are seemingly very happy with the services.

I mean, let's be honest here: a lot of people don't want to ride the bus because they associate it with poor people and don't want to share a vehicle with them. There's a lot of legacies of America's social problems getting tangled up with this question.

That's only helpful if they mostly share the same route; if it's a star topology, trains won't help.

That's why usually there are multiple lines that converge on a central station.

+ circle lines that connect the spokes.

I lived in Europe for a long time too. Trains are not a complete solution but they do solve a lot of the problems and are underused in the US.

Long distance train rides are incredibly expensive in the US.

I mean even moderate-distance train rides aren't cheap. Look at the price of a single ticket from the outer zones of the commuter rail systems into the city

Trains and buses don't run on demand from point to point.

They do a pretty good job of getting you very close in London and NYC. Longer distance trains get you between cities much faster than in a car. They don't solve everything but they are definitely underutilized as a solution in the US.

We could put effort into making them better but it's not as sexy.

Metro service is the key. When frequencies get high enough the timetable exists only for the benefit of the operations personnel, customers don't care about it because their experience is like with an elevator. You don't ask for a timetable for the elevator, you just go to the place where elevators arrive, and wait, and very shortly there will be an elevator.

Very high densities are both the problem and the solution. They're the problem because under very high density private car ownership is infeasible, and the solution because public transport becomes fast and affordable. If you let it.

Exactly. A lot of the issues with public transportation are because we underfund the systems and they end up seeming worse than they could be. The complaints that users have are not with the concept of trains or busses but with the failures, which can be minimized with adequate investment.

Frankly I think commuter rail for suburban riders is great and I think there should be more of it too. Who the hell likes driving to work during rush hour?

Trains and buses don't solve the last mile problem

Cars displaced those for a reason. Especially busses.

I have commuted to work by car and by train and I'd much rather do the latter. If it were so self-evident that mass transit were worse you'd expect to see people everywhere drive just as much as Americans

I can imagine a part of the Interstate mesh getting repurposed as SDV-only. In a city...not so fast.

Long range trucking is really the killer app. Fleets of N trucks can take an entire lane of road in sequence (almost like train wagons) and talk to each other and have M<N supervisors.

But long-range trucking seems just as amenable to replacement by freight trains if we're building dedicated infrastructure.

1. Building railroads is an order of magnitude harder than building roads (gradients, turn radii, clearance, yards).

2. Not even building, just repurposing. "This road/lane/whatever now SDV only."

3. Trains only run on rails. Trucks that could be run as SDVs on dedicated infrastructure and as human-driven on shared infrastructure tackle the last-mile problem far more efficiently.

4. Trains are built on the "smart infrastructure" paradigm (go straight at the speed which the signals tell you, until the signals tell you to stop); trucks on the opposite "smart vehicles" paradigm (road exists, everything else is your responsibility). This makes a truck rollout far more scalable (as in "just add this road to whitelist").

Trains don't run on demand from point to point. All your Amazon next day deliveries depend on trucks and airplanes.

Trains work for some freight delivery patterns, but not nearly all. Stuff that can go by train for the most part already does.

I don't know why people are still casting aspersions on self-driving cars when Waymo is this close to market.

https://www.bloomberg.com/news/features/2018-07-31/inside-th...

They're already testing with customers! Customers who love the service! They're planning for a roll-out later this year!

https://www.engadget.com/2018/05/08/waymo-snow-navigation/

They can handle rain and snow!

Waymo uses remote drivers for their service. It is neat that technology has progressed to allow drivers to sit an office rather than the vehicle, but that's not exactly what people imagine when they hear driverless.

They use remote assistance for tricky emergency situations. That's a pretty big distinction. The remote assistance can be called in to help the car if it gets into a completely unknown state, but otherwise the cars are self driving.

https://www.wired.com/story/phantom-teleops/

Waymo also runs on well mapped, predictable, grid like city patterns. Most of the world is not like Phoenix suburbs or Mountain View.

The second article I linked to also mentioned work on unmapped roads:

https://www.engadget.com/2018/05/07/mit-maplite-self-driving...

Waymo has solved problem after problem you think they haven't solved yet, for some reason. They've intentionally gone after unpredictable situations and dealt with them.

Look at any video where they explain how it works, and you'll notice how many situations they can manage: https://www.youtube.com/watch?v=LSX3qdy0dFg

And even if Waymo only works on well-mapped cities? That's still far from "can only work if they repurpose existing transportation infrastructure to only belong to self-driving cars" which is the comment I was replying to.

I think the ambition will be adjusted and it will happen. The realization that will come is that solving the last percent (or five, or ten) of driving is so expensive that it's just not pursued. So we'll have these 99% autonomous cars that will be confused and handed off to a human driver, inside the car or remote. This will give most of the benefit at a tiny fraction of the cost. The race won't be to reach fully autonomous driving, but to reach 99.9% when the competition is 99.8%, because that is half the workforce needed to drive all those confused cars.

If we take "self driving cars" to mean "cars handling every siytuation", I completely agree that reaching 100% without either restricting the area or modifying the driving environment won't be possible within several decades. But my guess is we never get there because no one will make that investment for such little gain. Completely autonomous driving (handling the things that happen only once in a drivers lifetime) will require so much of human intelligence that if you have that kind of AI there are probably better things to do with it than drive cars around.

Cars that drive themselves 99% of the time but 1% of the time give up and hand off to human drivers are going to kill the drivers or somebody else about 1% of the time.

That's a faulty assumption; if the 1% isn't “give up because otherwise we are certain to kill people” but instead “give up because this is a circumstance we aren't confident that the system will handle better than a human”, for instance.

It's a very reasonable assumption. When self driving cars give up, they give you a second or less to assess the situation around you and avert disaster. This is now well understood to be less safe than just 100% manual driving since your attention is then on the road at all times instead of inevitably wandering when the car is doing just fine 99% of the time. Don't forget that Wayomo killed off their SAE-2 test program after finding drivers were falling asleep at the wheel of these 99.0% autonomous cars.

And to build a self driving system that can give you a reasonable time to actually assess the situation and respond to it? That's the same system that's 99.9999% reliable.

This is the crux of why self driving cars will simply not work in the foreseeable future unless sequestered to their own tightly controlled road networks.

> When self driving cars give up, they give you a second or less to assess the situation around you and avert disaster.

I don't think the problem will be any situation with short reaction times at all. Nothing at freeway speeds, or situations like the Uber accident. That I think is where autonomous cars will shine, because sensors never get tired and reaction times are great.

The weird things that will happen which I count to the "not going to solve any time soon" category will be when the car comes to a completely snowed over roadworks, in the middle of the night, with the diversion signs completely hidden in snow. Construction workers barely visible in the snowstorm. Are those guys roadworkers or pedestrians? Are they working? Can I pass here? Will I get oncoming traffic because they narrowed it to one lane?

When things this weird happens at highway speeds or anywhere else where reaction is important - humans probably fail too. And at that point it's not really a question of technology but one of trust. Can we allow autonomous car to kill tons of people every year, with the sole excuse that humans would have killed all those people too, and then some? I'm not convinced of that either - I'm only arguing that from a technological standpoint, it should be possible to reach the 99% cars within a rather short timeframe. Those cars may be left on the scrapheap of history because of legal or ethical reasons, however.

The scenario that you described seems to me less intractable than the “woman in an electric wheelchair chasing a duck with a broom” that Waymo handled correctly.

One problem with this theory is that the Uber car sensors misidentified the woman who was killed and they were so erratic that they felt like turning them off made sense.

You're not considering why the AI turns it over the human. It's rarely about some crazy imminent emergency where if the human doesn't respond instantly and with super-human reflexes, it's all over.

In reality it's mostly just the AI expected one thing, and observed another - so something's not working right and it seeks a disengagement. California requires companies to quantify disengagements and most go a step further and specify the reason for the disengagement. I think the reason for this is precisely because of your intuition -- thinking that disengagement means imminent danger. Even for companies with relatively large numbers of disengagements, there were generally 0 that involved any danger whatsoever.

The problem is, as air flight has found, you're removing training hours from the humans so theyll be more and more dangerous also. And how do you safely hand off to a driver who is 99% of the time not doing anything. That's the thing humans are worst at, stayomg attent to long boring periods

The safety driver will be sitting in a call-center-like office and taking remote control when the car requests it.

That sounds like a terrible job with a lot of potential for things to go wrong.

Why? Sounds similar to the maritime pilots, whose job is to maneuver all kinds of ships just for small stretches (usually near a specific port).

These would be (remote) car pilots, probably specialized in specific areas.

High-pressure and probably a fair amount of death

Well, I'm assuming we're in L4, so they don't have to pick up the controls of a car going high-speed or such. I can't imagine any company would sign up to have its employees regularly have the clients' lives in danger, it would be unsustainable legally and PR-wise.

It is a bit of an exaggeration, but the fact remains that handing off to a bored driver who likely isn't paying any attention and expecting them to take over a car at speed is not reasonable or safe.

You're imagining something akin to the autoland disconnect scenario for a jet liner, and you're correct that, in the absence of expert operators who are prepared to take over immediately this is often fatal. Which is why the jet liner has not one but two people specifically standing by to take over and try to land safely without the autoland. [ This is still scary because usually the autoland is trying to put the plane down in very poor visibility, and it is disconnecting because you're below decision height but something so bad went wrong that it's no longer able to land the plane - a human pilot may not be able to make whatever that bad thing was survivable either ]

The auto-land disconnect scenario isn't applicable to cars. It happens because jet liners are _flying_ and suddenly ceasing to fly in a jet liner is both very bad and perhaps unavoidable in the absence of enough information to operate the plane within parameters.

In contrast when a car becomes uncertain about what to do it's not flying so _stopping_ is almost always a good choice. It's not ideal, it may block traffic and be a nuisance, it might even cause a small accident of some sort - but it's very likely to end with everybody walking away, not with a burning wreck and dozens of dead.

talking about cars that drive themselves only 99% of the time is a straw man. Waymo's disengagement rate in the year leading up to nov 2017 was once per 5,596 miles [1]. i'm not sure how to translate that to a percentage of time, but let's say a disengagement is a tenth of a mile - that means waymo's cars are self-driving 99.99998% of the time. they're the closest to market, but that was almost two years ago now, and in the 65 reported disengagements they didn't kill anybody.

Waymo measures their disengagements differently than the competition.

https://blog.piekniewski.info/2017/05/11/a-car-safety-myths-...

> Now it is important to note that the definition of a "disengagement event" may vary between companies. Most companies report every case in which a human grabs the wheel for any reason. Waymo (*) only reports the events, in which if not for the human intervention the car would actually cause a dangerous situation [read more here]. The way they do it, is for every physical disengagement they gather all the sensor data and next simulate multiple scenarios. If these scenarios lead to a dangerous situation, such event is being reported. According to Waymo in 2016 nine events would have lead to the car hitting an obstacle or another road user, approximately 1/10 of all disengagements they've reported (124). Hence there is such a gap between Waymo and the rest of the pack.

I also think it's hard to generalize from Waymo employees being attentive and general road users achieving the same thing.

I don't think the goal should be seen as perfect driving, and that is likely impossible in any case. You can take the most extreme scenario that AI or human can possibly successfully overcome, and then make it just slightly more challenging to induce failure. You'll always be able to 'break' a driver's ability to correctly respond whether they're human or AI. So the goal, to start, should simply be to overcome humans on average. If a human will end up in a fatal crash every x 'average' miles, then an AI is a success if gets into fatal crashes at an average rate less than x.

I really hope that the future won't be even more car centric but that's how things may go.

You have to ignore Waymo to still have this opinion. I doubt they'd buy 80k cars if they didn't think they could use them soon.

Or you’d have to not have walked around downtown SF lately. It’s a rare day where I don’t see multiple self-driving cars.

Delta City (aka New Detroit) would be the ideal for testing.

When I think of self-driving cars I always think of this taxi scene (https://www.youtube.com/watch?v=eWgrvNHjKkY) from Total Recall. Paul Verhoeven is a genius.

Isn’t that the one with the body scanners (at the airport)?

Uber is way way far out of their league here.

Imagine being a software and services company trying to build one of the most mechanically and electronically complex items that humans have ever built with a team that has only built prototypes and not having the war chest of continuing positive cash flow to fund development.

Not even looking at the strategy of it all, this program sounds like it needs to be dumped in favor of partnerships.

> with a team that has only built prototypes

This is normal for research isn't it? If someone before you has already built a working version of something, then it isn't research any more. And Uber hires extraordinarily qualified AI researchers doesn't it?

Reasarchers aren't normally expected to build a production system to sell as a product. They might be very smart people but for something like this you need a team with at least some members who have real experience delivering automotive hardware + safety critical systens + automation.

Research tends to ignore the boring but important details that make a product.

I think what they're saying is that usually teams don't go that far out of their industrial strength. In a way, Uber here is going from being cell phone app developers to making automotive hardware, which is a big shift. Shipping hardware past the prototype phase is __HARD__ in ways that shipping a software app/building a marketplace isn't (not saying the latter aren't hard, but the two tasks take very different types of organizations).

Uber here is going from being cell phone app developers to making automotive hardware

But why would they do that, instead of hiring other developers? Surely Levandowski would have an idea of who to hire.

It’s amazing how fast and radically the narrative shifts.

An year or so ago self driving cars would be here anytime now.

Now it’s almost consensus that it’s not for going to happen in the next decade.

I can't imagine there are many experienced software developers that maintained bullish perspectives on self-driving cars.

If you hear a consensus now, I suspect it's because the din of noise coming from the bulls has finally quieted down and you're able to hear the moderates and bears a little better.

The barriers preventing self-driving cars from taking off right now aren't software related though, they're fundamentally related to the limitations of neural networks that people didn't think would be an issue 1 or 2 years back. Our understanding of the capabilities and limitations is constantly changing, which is why many (including myself) thought full self-driving cars would be available in the next several years.

Are there non-software neural networks being produced yet? I was still under the impression that all the hardware neural networks testing the plasticity of circuits were still either theoretical or at least early in development.

That's correct, but what I mean is that this isnt a traditional software problem, it is a problem related to the workings of NNs themselves, whether they are embedded in software or in pure hardware.

so more of a theoretical problem rather than an implementation problem?

Maybe your memory is not perfect. In November last year Waymo announced that they will launch this year a commercial service. They are already testing on the roads with members of the public without safety driver. Before their announcement the consensus was at best in 2020. Can you please show me how today the consensus can be for the 2030s when we have already cars without safety drivers on the road?

Aren't they all in locations with perfect weather and perfect wireless connectivity due to low construction density? I think it's differing expectations as to what IRL meant.

They seem to have solved the snow problem, at least having a viable solution for it.

https://youtu.be/ogfYd705cRs?t=5936

I think it was mainly lead by companies trying to get funding. People who understood the technology were not saying that. Good lesson to remember for next time there's a cool new technology.

I think there will be a realization soon that the AI hype in general is not going to solve all the problems CEOs and companies are trying to raise money for.

I don't think that's the case at all. At Google I/O they announced that Waymo has been doing fully autonomous driving in a limited fashion with real people in Phoenix for ~6 months.

Counterpoint, it's happening now with Waymo.

Waymo's initial commercial deployment is imminent but is now a good 6 months behind their initial announcement last November. Waymo is still using safety drivers most of the time, and the commercial deployment will likely be confined to members of their EZ rider guinea pigs. It's very exciting seeing how far they've come, but for every new region they hope to roll out in there's going to be a ton of work. AI can interpolate but it cannot generalize.

We're past peak hype and sliding into the trough of disillusionment.

Why is 6 months behind if they said that they will launch a commercial service this year?

Last November Krafcik said "In a few months". A few months later in March it was changed to "Later this year." Rumours circulated that they had planned for this month, August, which were corroborated by a big Bloomberg puff piece and a ballyhooed partnership announcement with the Valley transit authority, that came out late July, but still no deployment. They're hung up on something.

Right, right. Any moment now, and with flying cars, too. I am not holding my breath.

funnily enough self driving flying cars would be likely in a better shape to become a reality ai wise, because there are less human there and one can work within the framework of exclusion zones with governments to let them share the sky safely.

Except for the other part, i.e. failing to fly well. Which is sort of the point: "it's just around the corner, except there's always this one tiny inconvenience, but it will be completely awesome once we do something about it, so let's just handwave it off for now." (Yeah, yeah, I know that somebody made a render of a flying taxi this year. And last year, and the one before that, and the one before that, all the way to Kitty Hawk.)

The narrative is a function of a loud group with time and energy to spare, and downvoting power. A year ago you can still find the same issues raised today being raised, they’ll just be gray at the bottom of the page with a lot of angry replies like, “well human drivers kill a lot of people!”

Yandex Taxi looks really close. https://youtu.be/BbGEGDx59ZI

780 km (11 hours) at 99% autonomous. https://youtu.be/zljaMjLFqfI

> 99% autonomous

IMO the danger only increases as the autopiloted car approaches 100% coverage asymptotically.

As long as there's an expectation for the human to be able to take control I consider the technology worse than useless.

To me it looks like self-driving problem turned out much harder than anticipated by most players at first(10 years ago?)

Which kind of amazes me that it wasn't obvious. When I start to think of vision recognition issues, poor conditions, missing signs, markings, etc., unpredictable human drivers... then all of the edge cases I alone have seen in 30 years of driving... I would definitely have put the problem on the very edge of what might be solvable. And that's today, not 10 years ago. What amazes me more is that companies like Google have come as far as they have. My kudos to those who sought to tackle it.

Yeah... Just yesterday I was driving at night, through road construction, with a torrential downpour, and nowhere to pull off, and it was freaking scary. And I remember just thinking, "This kind of thing happens to me maybe a few times a year and there's no way in 30 years that we're going to trust autonomy with this"

but isn't that part of the point? I doubt any automated system would consider driving in those conditions 'safe', so it would deal with the situation by pulling over to a safer spot and stop driving. Humans make terrible risk decisions in cases like that - continuing to drive in horrible snowstorms, etc, when the risks a way higher than our already-risky roads in normal conditions.

By not having the human make that decision, you save lives, even if some people arrive home late.

It does raise the point of 'rescue' in certain dangerous conditions like winter storms. Extreme rain in the dark can probably normally be waited out, but snowstorms and other road-closure type conditions probably warrant a different proactive rescue type response if we'll have riders with no driving ability in self-driving cars.

"Not driving in the insane conditions when humans are foolish to do so anyway" would IMHO allow routine drives in good weather in known terrain without roadworks, about half the year. Which is a great and magnificent improvement, in all honesty - that is, once we can get the marketing types to cool down from their current hype "it drives itself, full autonomy, everything and a pony*!!!!!!!"

I think this is a critical point. We need to reign in the customer's expectations that have already been set too high. They're already expecting to just get in and go anywhere while watching things on their phone or reading a book... soon. We already see this with people posting videos of sitting in the passenger seat while their Tesla rips down the highway in traffic.

Like anything it should be a graduated phase in. It will handle some of the conditions some of the time, and in time it will get better. It would be like me being frustrated I can't carry on a conversation about philosophy with my Google Home. "... but you said I could talk to it and ask it questions!!!"

Expecting a conversation on philosophy would be completely understandable - if the vendor sold it to you with the tagline "it has all the parts it needs for a philosophic conversation!" Google Home doesn't do that, Tesla does. (Musk doesn't even try to weasel around it: says "full self-driving features", Tesla marketing materials repeat. That is, in my opinion, a blunt lie.)

Self-driving cars can automatically check weather before starting the planned route and drive and avoid those situations.

In the past few years there were leaps so gigantic in the field of Computer Vision that they could be said to have happened overnight within the general pace of electronic technological development; within an eye's blink in the general scheme of things.

This made us so, so cocky. And man, if it's that easy to make a website that opens a webcam and tell you your gender and age to a remarkable degree of accuracy, surely self driving (or even playing Mario Kart well) must be within the next two breaths, right?

It's the unexpected things..

A few weeks ago I was driving home from my parents and suddenly found myself diverted off my usual route, off from a motorway. Google maps was not updated and rerouted itself to try to put me back on the normal route, which was to double back and rejoin the motorway before the diversion - if I had followed its instruction I would have been stuck in a loop! Hard to see how a self-driving car would cope with that. I guess part of any diversion in a self-driving future would involve the authorities informing Google about it beforehand.

But it got me thinking of even more unusual events (that are nonetheless not so unusual that a driver wouldn't expect to meet them at least once in their life) - what if you come across a fallen tree blocking the road? A human being could make an executive decision to drive over a footpath or through a field or reverse back up the road a little way if it was safe... how could an automated system hope to cope?

Falling back to the human driver seems a cop out. What if the passenger cannot drive?

Perhaps just a failure of imagination on my part. Some sort of mission control that gets connected to where some remote operations steer you out of difficulty? Or does one just "await rescue"?

I just wonder if self-driving is one of those problems where you can get 95% of it solved, but the remaining 5% remains uncrackable.

> I just wonder if self-driving is one of those problems where you can get 95% of it solved, but the remaining 5% remains uncrackable.

It's felt that way to me from the start. I tend to dwell on small details more than necessary and more than I imagine most people do. When I take a drive it's not just a drive to me, it's a series of countless decisions with more ambiguity than I'd trust to an algorithm. "Clean" driving conditions have to be seen as the edge case, not the norm.

I believe the AI that gives us acceptable search results is so far removed from the kind of human cognition that understands the world and reliably resolves everyday ambiguity that we're not only not close, we haven't even started on a track that gets us there.

The traffic situations could be handled by some sort of beacon system informing drivers within say 500m distance that on given road they need to take over computer because something exceptional is happening.

The problematic part would be something unplanned (ie car crash ahead of you). You can't leave this to other drivers (replacing car hazard button), because it would be very easily attackable to create chaos on the roads.

There are so many problems and corner cases that we could sit all evening and keep coming with more or less exotic ones. Just look at how busy traffic looks like in places like India.

Self-driving is probably #1 on my wishlist (next to true immersive VR), but boy that's HARD problem to solve. Maybe 2040 if companies keep the interest(money) in it.

People were claiming self-driving cars around the corner for ages.

This is the real issue: https://en.wikipedia.org/wiki/Hubert_Dreyfus%27s_views_on_ar...

Did Uber figure out that if anyone can buy self-driving cars that their golden utopia vision of booting all drivers and simply maintaining a fleet of self-driving roving vehicles doesn't make sense? Did they figure out that someone else can run servers just as well as Uber and offer people the opportunity to lease out their self-driving cars, utterly destroying Ubers market (they would have to charge far more to maintain the whole lifespan of the self-driving car where someone only renting theirs out for a few hours a week would happily take far less)? Or have they just generally realized self-driving cars will never work because no company in the software space is capable of dedicating the years of testing necessary to make a truly reliable product, while those willing to cut every corner will be first to market.... and will proceed to destroy the market by turning society against them after some of their poorly-designed vehicles smear a few toddlers across the sidewalk.

Wasn't Uber's autonomous program basically a bunch of Carnegie Mellon researchers? My initial thought was that they actually had some of the best people in the field working on this problem. I vaguely even remember some criticism of Uber because they had plucked so many of the prominent researchers that there was brain drain at the university. What happened?

I personally think there are so many flaws in these self-driving cars safety features. There are quite many review by car owners on Youtube, forums, etc. I have also read an article that talks about this at https://www.lemberglaw.com/self-driving-autonomous-car-accid.... I think until car companies fix these flaws, I myself will not try one of these cars.

The article states:

>"The company, valued at $62 billion, has racked up billions of dollars in losses since it was founded in 2009 and needs to persuade investors that it can eventually create a sustainably profitable business. The self-driving efforts, which have been losing $100 million to $200 million a quarter, do little to help that case."

Can anyone say or speculate what percentage or Uber's losses are a result of the ATG efforts? Would they be profitable without it now?

What I don't get is why Travis thought whomever developed self driving cars wouldn't sell them to all comers? And even if the first guy didn't, under the guise hoping to create a vertically integrated monopoly, the 2nd, 3rd, 4th ... vendor would.

Reply on news.ycombinator.com