Expanding our testing in San Francisco | Hacker News Reader

Expanding our testing in San Francisco

blog.waymo.com

251 pointsra7243 comments

I took a road trip through AZ in January and took a detour to Chandler to try out a fully autonomous Waymo ride. I worked on self-driving at Lyft and I have a comma two devkit installed in my car. I have to say it was actually pretty magical having a robot car glide up to me in the Walmart parking lot and then take me for a drive. It's one thing to read about the launch and a wholly other thing entirely to see a fully empty car slink up to you in a parking lot.

The ride is most likely being monitored remotely. I changed my destination back to the start location after traveling about 20 minutes and an attendant's voice came over the speaker and asked if everything was alright. I said it was fine, but was interested that they had a live team monitoring and able to step in for customer service like that.

I did have to wait 17 minutes for the Waymo to actually show up. This success was after 3 or 4 failed attempts to request a car, so if I actually lived in the area I would probably not even open the app to check the ETA. This point is largely moot, as it's pretty clear that strategically AZ is a testing ground rather than something Waymo expects to make any money doing.

I can't wait for the autonomous future. Best of luck to all the entrants in the competition. I just hope you are able to turn a profit before investor money dries up. We've been waiting on promises of this elusive future for nearly a decade now ( ͡° ͜ʖ ͡°)

Do you think that SDV is a bit (just a bit) in a kind of skeumorphic zone where mobility was defined by human driven charriot and thus computers are having to replicate all the complex human perception and control.. while, in urban settings, we could pave the roads with markers and beacons so that moving objects could move around without all the sensing technology ?

What you're describing is a reliability nightmare.

Human drivers 'depend' on road markings, but if some paint gets scuffed or a stop sign blows down in a storm, things still work okay. For AI cars to be similarly fault-tolerant, they need to be generally capable drivers who don't absolutely need exacting external markers to get around.

You can't pave roads with markers and beacons to do self-driving without simultaneously banning all non-AI traffic participants, including pedestrians.

fair point

unless... we all were antennas under our skin

alright Bill Gates

Absolutely. In the same way that we add street lights for humans, we will add waypoints for cars so they make less mistakes.

Since you seem to have a lot of relevant experience how likely is it that you think Waymo could ever succeed from a business perspective? As you mentioned, they are going to need quite a few vehicles and at least some form of remote monitoring callcenter. I imagine the economics of this will be fairly unprofitable for many years to come still. Granted, Google has infinetly deep pockets, but thats a lot of money to recoup once it actually becomes profitable years down the road.

I think yes, but the economics are actually very different than their competitors (Lyft / Uber) will be. I also think their path to profitability will not have Waymo acting as the supplier (of rides) and the platform. Lyft / Uber have variable cost structures, while Waymo will be in a race of diminishing returns of extracting value from the fixed costs of their vehicles before they are no longer in service.

There's a great paper [1] that discusses this topic. In the 4th section the authors discuss a very straightforward model for this. They model a maximum time that a car could be in service (say 10 years) and a maximum mileage that these cars could operate. So then the car is in a race to earn as much money as possible to firstly pay for itself and then accumulate excess value before end of life for the robot.

It's harder when the cars are way more expensive ($200k - $1M) and especially when high utilization is table stakes for getting this to work. High utilization can only happen when you have a lot of people successfully getting rides on your platform (which the hard balancing problem of supply vs demand).

All this to say, yeah I think they can. I don't know if it will be the best business (race to low-margin operation of a fleet).

This paper also has an interesting argument against the manifestation of a future where personal car ownership is too expensive and everyone just relies on this fleet of self-driving robots to cheaply go places.

[1] Carpooling and the Economics of Self Driving Cars - https://web.stanford.edu/~ost/papers/sdc.pdf

As a small data point - the streets of the Sunset are being inundated by self-driving cars. They're well behaved and mostly pretty effective from what I can see, but I'll see literally 10 or 15 of them most times that I walk my dog (and they're out all night too). Primarily Waymo's vans and Cruise's Bolts but a number of others too. It's striking since overall traffic volume is down so much but there are just a ton of these guys collecting data.

(Edit; My word choice made it sound like I'm complaining, but it's all good with me. Would rather have these looking for me at intersections than the typical distracted driver.)

Since so many of them showed up in the Richmond I learned that my dashcam can see their 1550nm lasers, so they look like strobing party cars: https://youtu.be/56G34udNTfQ (one at the beginning and a second at 3:10 that you can also see reflecting off storefronts and parked cars)

At least I assume it's 1550nm based on https://patents.google.com/patent/US20160291134A1/en

Pretty sure both cars you mentioned in that video are Cruise, which is why you're able to see them as it's 905 nm. Your patent links to Google, but they don't make the cars in your video.

If you take a left on Arguello, and then right on Fulton, you won’t hit all the stop signs on Turk

Your dashcam is CMOS. It is not capable of sensing 1550nm, see i.e. [0]. I've worked with 1550nm sensors, they use different material.

[0] https://www.ir-photo.net/image/standard_camera.png

> I learned that my dashcam can see [..]

Given it was night, I'm impressed with the quality of your dashcam footage, mind sharing what make/model you have?

It's a mini0906-PRO (Novatek chipset) with a 24mm circular polarizing lens. I got it for the 60FPS support.

>It's striking since overall traffic volume is down so much

That's interesting. Ex-urban Boston intersecting with a couple of small cities, I basically don't immediately see material differences relative to pre-pandemic. Maybe a little less traffic congestion on the major roads? But certainly not "much less." (Which I find a bit surprising given how relatively little I go out but that's what I see.)

Pre-pandemic, there'd be 2-3 hour long periods with stop-and-go traffic inbound in the morning and outbound in the evening on Fresh Pond Parkway (2/3/16) in Cambridge.

I'd say that traffic is easily 30-50% of pre-pandemic levels now. Afternoon backups in front of my house are almost entirely gone and morning backups are quite a bit lighter and shorter.

Certainly the opening weeks of the pandemic, way back in late March of 2020, traffic was virtually nonexistent in my part of the USA; just outside of Portland in Oregon. And then traffic picked up just a tiny bit, but for months it was a cyclist's paradise; miles and miles of roadway open and very little traffic. For most of the summer it was an absolute delight. We're pretty much back to normal now.

I commute ~50mi into the Boston area daily. The volume is definitively down enough that the usual backup spots are no longer guaranteed backup spots except at the peak hour whereas before it was probable to hit a backup at those spots at any hour between 2pm and 6pm. My friends who ride the T say it is similar.

There was spectacularly little volume during the opening weeks of the lock-downs in March. It was like Tuesday volume becamee Saturday of a long weekend volume but without the downside of Saturday drivers generally driving slow because they're not commuters on those roads.

I wonder how much this is impacted by people avoiding the T like the plague?

I'm nowhere near the T. I'm about an hour west of Boston. And there's essentially no public transit out here. Based on looking at parking, commuter rail use looks extremely low but that shouldn't affect local traffic during the day that much (though some of it may admittedly be people who aren't otherwise in Boston offices).

What are the implications for you and your dog being scanned/captured by self-driving cars during these tests? Is biometric data usually discarded from these datasets before further processing and storage? Do you have any right to control your data or restrict its usage?

This is my favourite question! I don't think the stuff you described is any problem legally; lidar scan data and pictures from public places, not published, kept for a legitimate purpose (vehicle safety and training) should be fine.

But it's not a big step to turn these things into continuous scanners of the real world. Now every car can be a Street View car.

You could have them check parking lot occupancy, see whether storefronts look busy, scan license plates, measure foot traffic, do recognition on pedestrians and cars, enable real-time-ish "look around" in maps, scan for wifi networks or other signals, report weather conditions, and just generally act as massive mobile data collection network. They can bring the kind of invasive surveillance we have on the Internet right to your doorstep. There is a huge amount of money to be made.

The data might end up being more profitable than the rides.

I don't think there are any laws preventing any of this right now. We're going to need regulation eventually.

i think it would save a lot of headaches if legislation was drawn up today. it's not as if surveillance isn't possible already and doesn't happen. its an issue that's ripe for legislation. prevention is the best cure.

Solid disagree. Digital privacy has devolved into a luddite movement for privileged people who have nothing better to do than convince the less privileged that they'd be better off dealing with actual crime than allowing the nebulously constructed idea of their digital privacy to be violated.

There's nothing wrong with protecting the privacy rights of individuals when the largest companies in the world are running robot cameras down all of our streets.

there is some merit to your argument, by giving up some privacy we have created huge amounts of wealth for ourselves. But we can still fear a dystopian reality where ai and cameras will be used to create files on everyone.

I expect there will be political and law enforcement pressure to use those autonomous vehicles to scan for "Amber Alert" license plates.

The police are gonna be real annoyed when someone figures out how to make an OBD2 dongle or app that automates the "report/confirm a cop" portion of Waze

In person retail attribution and re-marketing! the next billion dollar ad tech startup. lol.

Though maybe if Google does it themselves they could actually make it work.

That's a fair question that I don't have an answer for. The Google Maps cars all blur faces (I'm sure in post processing) for their public data sets but I assume they haven't taken the time to do that with the self-driving ones. I guess GM and Google have tons of unflattering videos of my pandemic-self wearing the same clothes on multiple days in a row while I mutter to dog and my Airpods.

Right. It would be a small jump for Google to start using profile data on people to figure out who they should hire and who they should not.

Wonder in what way that influences the learning process, when all the algorithms are learning from each other.

Now I’m imagining an adversarial network of vehicles straight out of Mad Max & Death Race...

Now that would be a cool competition. Self-driving cars dropped into a random locale, it would be a no-holds-barred race to the finish line/battle royale. No human drivers means so safety concerns!

For this, train the cars in DC.

Clearly you haven't driven in some European cities — and I'm sure there are places that can top that :-)

As do I. I wonder in particular if there's a possibility that the (I presume) ML behind these things can start to distinguish human from machine drivers, how quickly that determination might be made, and react accordingly.

It'll be interesting when the characteristics of the cars are known. Eg you can cut in front of a Tesla and it'll stop, but dont do the same for a Chevy.

Same here, panhandle area. I walk the dog 1-1.5 hrs during the day, half hr at night. For years it was just the Cruise cars, but since last summer the Waymo vans/suvs started showing up and now it feels like they're constantly passing by. Late at night it's especially noticeable because of the lack of traffic.

It's the perfect time to test. Less civilians to hit, winter weather, still good data to use.

"winter weather" is a stretch

Heh I mean SF winter. So you know, slightly cloudy with a bit of water on the ground sometimes.

The coldest winter I ever spent was a summer in San Francisco.

https://pbs.twimg.com/media/EuTP4-0VoAAl-vd?format=jpg&name=...

A SF native, are you?

A solid competitor for the stupidest quote of all time. Even winter in San Francisco is massively warmer than basically anywhere in my home country. Summer is absolutely baking.

Certainly not. The average summer day in San Francisco is foggy, windy and cold. Coat weather all year.

Actually nice (>80F, sunny) days in San Francisco are rare.

"Baking" days, are more like once in a decade.

> Summer is absolutely baking.

No it isn't? The average July daily high in SF is 67 degrees. That's absurdly cold for summer, there's basically nowhere south of Alaska in the Northern Hemisphere that stays that cold in July.

There isn't a single SF climate. It changes radically between, say, the Financial District and the Sunset.

Similar here in Menlo Park. We are near Willow and 101, where the temp can easily be 10 degrees cooler in the summer than it is just a couple of miles away in downtown.

Not really relevant to the extreme stupidity of the quote. In a cold winter, it would snow.

Do you guys really not understand the concept of humorous hyperbole. At first I thought you were the same commenter who posted above but apparently there are two of you!

- Oscar Wilde

I always heard it attributed to Mark Twain

The Twain quote comes up every time people talk about SF weather, figured I might poke fun at it at the price of some karma

Or Keith Hernandez if you ever watched a Mets broadcast while they were playing the Giants.

Thanks for data point. How is it interacting with the self-driving cars? I'm concerned that a lot of thought has gone into experience for the passenger of self driving car, but much less for the rest of general public. Feels like they missed an opportunity to improve driving experience for rest of us as well with some more signals or communication between vehicles.

In my experience as a pedestrian - they're extremely cautious, which I appreciate.

For most human drivers, I'll wave them through intersections if I'm on the far side and crossing will slow them down substantially -- partially out of sympathy for driving through slow crosswalks, partially out of self-preservation as drivers are often impatient and chomping at the bit to get through.

With the self-driving cars, they really creep up to the intersection so I never really consider waving them through (and I have no idea how they'd react). They're very good about not entering intersections when I'm approaching and they wait until I'm all the way through as well, so in my experience, they're very cautious. It's nice too since where I am in the sunset has random blocks with 4-way stops adjacent to 2-way stops, which is fairly dangerous for human drivers but the robotaxis handle those without issue.

This is why I can't wait for self-driving cars to take over. They are sooo much safer for me as a pedestrian.

They could almost make cars tolerable in the city. Almost.

No better time for autonomous driver software to practice driving than when the streets are relatively empty. (Obviously, eventually they'll need to handle full traffic, but the current situation is probably good training wheels for that.)

This isn't the case in 2021. Driving empty streets is pretty easy, and the crazy corner cases that urban density requires is the whole pt of testing in San Francisco.

I think it's more likely that waymo's commercial product in sf will prioritize neighborhoods like the sunset (poor transit connectivity, easy driving).

San Francisco post-covid is gonna be so weird. Just electric scooters and Weymos. Like no parking, we just turned that into outdoor dining.

> Autonomous vehicles promise to change this, revolutionizing our cities by making transport safer, cheaper, and easier to access. But deploying autonomous driving technology to get there – perhaps one of the greatest engineering problems of our time – isn’t possible without developing a deep understanding of the nature of the challenges it must master.

Uh this problem has been solved for over one hundred years, its called public transit. This is just the most expensive, most complicated solution for a problem that has already been solved in many places many times. Its like using globally distributed micro services to serve your blog which could just be hosted on medium.

Yes, driverless cars will be useful for things at some point, but we don't need them in a dense city! I want fewer cars, not driverless ones

> This is just the most expensive, most complicated solution for a problem that has already been solved in many places many times.

Can you estimate how 'expensive' and 'complicated' it would be to retrofit San Francisco with a public transit system?

I would guess tens of billions of dollars and tens of decades, and that's if it would ever get approved in the first place.

London has spent $20 billion and twenty years adding one new rail line. In a city starting out as already famously enthusiastic about public transit.

What should we do in the half-century before this possible transit system turns up in San Francisco?

Don't let perfect be the enemy of better.

SF has spent more than 1.5 billion dollars and a decade trying to add 3 stations to the existing train network downtown.

https://en.wikipedia.org/wiki/Central_Subway

That's great, its a good investment.

Also, buses exist, and they work quite well actually, especially if you remove cars from the road

https://en.wikipedia.org/wiki/Bus_rapid_transit

Indeed, do not let perfect be the enemy of better. The best way to improve transportation in SF is to tell everyone in a car to fuck right off, and run a bus both ways down every street every 3 minutes. No rails required.

I do also support completely closing all cities to private cars (reasonable exceptions for genuine requirements.)

I don't understand why almost anyone needs, let alone wants, to try to drive a car into a city in the first place.

Relcaim the streets for people.

But again... how 'complicated' do you think it would be to achieve this legislation? I would guess impossibly complicated, so it's no more than a dream. So what do we do about it now? And what can a private company like Waymo do?

Do it incrementally:

Ban parking minimums, so developers stop putting parking lots in their buildings.

Tax lots based on their assessed land value and give them credits for providing housing units. Developers will build more units and less parking. Surface lots will close down or dramatically increase rates -- either way, people will find alternatives.

Require transit lanes for all major avenues, and enforce their use.

Increase taxes and spend that money on transit.

Make transit free, which will make it much easier for people to just hop on for their trips.

Invest in trains. Bus rapid transit is kinda terrible, but it could be a stopgap until your trains work well.

(Not on the same continent as San Fransisco)

When they first sought permission to build the building I live in, it was rejected for too few parking spaces, but by a decade later the penultimate plan was rejected for too many∆. My flat has a space (which I don't use) but some other residents don't get one. Public transit here is excellent, amenities are in walking distance, I can imagine wanting a car if I had kids or was disabled, otherwise it's unnecessary.

∆ the land didn't sit empty, it had a house on it, permission was sought to build a large multi-residence building on that land instead, once that permission was obtained the house was demolished.

>Ban parking minimums, so developers stop putting parking lots in their buildings.

SF already took that step:

https://www.reubenlaw.com/san-francisco-eliminates-parking-r...

Their permitting process is still a mess, based on what I have read, but that is a good start.

> 'complicated' do you think it would be to achieve this legislation?

Depending on how extreme you want to get, not that complicated. Part of market street has already been closed to cars. Many streets have been converted into pedestrian streets during covid.

I bet we will see more meaningful improvement on this front in SF before driverless cabs hit the streets

The new train line (Crossrail) in London is years behind schedule, mostly due to problems with the software control systems for the trains. [0] Integrating automated trains into an existing rail network isn't easy - imagine the scale of the challenge with automated cars on city streets.

[0] https://www.wired.co.uk/article/crossrail-delay-logistics

> I would guess tens of billions of dollars and fifty years, and that's if it would ever get approved in the first place.

So? I still think it is worth it. Also, these costs and delays are not given, they are artificially induced by mismanagement, and could be changed. I am not saying that they shouldn't build the cars, I'm just pointing out that the "solution" they are providing isn't that revolutionary.

Also, buses exist, which are much cheaper than what you are describing

> Don't let perfect be the enemy of better.

I don't really see how this is that much better than what already exists. How different would this really be from the status quo? I can already automatically hail a car which will take me where ever I need to go. Sure, autonomous cars _might_ be safer, but I doubt they will reduce congestion or travel times all that much. Plus, we'll have rolling surveillance cameras with a myriad of advanced sensors constantly driving by. I don't buy it

> So? I still think it is worth it.

The 'so' is that your complaint was that these cars are too complicated and expensive. I think an actual public transit system is undeniably more complicated and expensive, even if it's worthwhile.

I'm not complaining that the cars are expensive or complicated, I am just pointing out that it is a silly solution to problem which has already been solved.

> I think an actual public transit system is undeniably more complicated and expensive, even if it's worthwhile

I'm not so sure about this. We have public transit systems figured out in a lot of places. We don't even have self driving cars that we consider safe yet, let alone laws which regulate them

> [the cars are] the most expensive, most complicated solution

> I'm not complaining that the cars are expensive or complicated

I don't see how stating that something is expensive or complicated is a complaint. You clearly are not trying to understand what my original comment meant. I was trying to point out my perceived irony of the situation, then state my opinion, not complain about Waymo.

I'm not really sure why you are being so adversarial, looking at your other comments here, I don't think we disagree all that much

Like autonomous electric buses every 2-3 minutes in dedicated lanes that are demarcated by paint?

Autonomous because labor costs (and shift considerations) are a major factor limiting the frequency of service you can provide. Electric because they will cost less in maintenance.

They are also an important part of a possible solution for helping lower density suburban areas that suffer from housing shortages. It's worth listening to this whole talk by Peter Calthorpe, the planner who developed the now-well-known concept of transit oriented development: https://usa.streetsblog.org/2021/01/07/talking-headways-podc...

> I would guess tens of billions of dollars and tens of decades, and that's if it would ever get approved in the first place.

So a quarter or two of FAANG profits for scale?

Yeah great if we can get that much money.

Did you have some good idea how to achieve that?

I basically agree that cities of a certain density are too dense for cars, but SF isn't really there. Though I would love for SF to remove the zoning thay causes it to be this way.

I think it's also worth considering that self-driving tech ia fully compatible with buses and that the driver is a big component of the cost of bus service, which limits bus frequency and constrains ridership.

Trains are great, but we're basically incapable of building them at reasonable cost in the US.

> I basically agree that cities of a certain density are too dense for cars, but SF isn't really there.

Depends on where you live. If you live in the outer areas (south of bernal heights, outer sunset, etc) you might need a car but most other places you can get around easily with transit, walking, or biking

There is a difference between needing a car (i.e. I cannot live without it) and public transport being so good that you do not want one and the density being too great to support cars.

My experience living in the Lower Haight a few years back was that buses were slow, infrequent and required multiple transfers unless I was going to a few select places. Cycling was ok, but man were there some big hills (potrero) involved, e.g. cycling back from the dogpatch was pretty tiring even for a young person in pretty great shape at the time. And taking a train from dogpatch to the haigh required going back through the city.

So I could definitely get by without a car, but it constrained where I would consider going pretty significantly.

On the other hand, driving was usually fine, but finding parking was impossible, which self-driving cars don't have a problem with. Of course, parking was constraining driving, so removing the need for parking would make congestion worse, but even pretty close to the city center I felt like driving was pretty easy, and if we structure our streets (and bus lanes) right, the demand for self-driving cars will be in the routes that do not have enough demand/congestion to sustain good bus service.

While I agree in principle, it's not like Google or Cruise can change that.

The reason we have crap transit in the US is because of politics and culture, and tech/car companies can't unilaterally overthrow everything we already have, even if they wanted to.

Also, even in transit-friendly Japan or South Korea, there's still plenty of cars about, so self driving work still has value.

I think that’s a false dichotomy. Yes, public transport is much better than everyone driving a car. However, whenever I talk to a parent of 2, they say a car is a must even a city like Berlin (that has a vast public transportation network). There no one size fits all.

> I think that’s a false dichotomy.

I disagree, I never said cars should totally be banned, I was just pointing out that it is ironic how Waymo is claiming they are solving this age-old problem of transport in cities like they Gods when the problem has already been solved for a very long time.

> whenever I talk to a parent of 2, they say a car is a must even a city like Berlin (that has a vast public transportation network). There no one size fits all.

I don't think fleet of self-driving taxis would do much for these the people so I don't really see how this is relevant. Yes, people need cars for various reasons such as a disability, kids, etc. However, the problems those people face are not the problems Waymo claims to be solving. Waymo is claiming to solve general transit related issues, which again, is a solved problem

It’s a solved problem in densely populated cities like HK, Tokyo, Shanghai etc. I only know about US suburbs from the movies, but it doesn’t seem like public transport would solve the commute there.

Cool, if you wanna get downtown from the Mission.

SF hasn't had good public transit since at least the 50s. https://www.streetcar.org/end-of-the-b-geary-60-years-ago/

> Cool, if you wanna get downtown from the Mission.

Awesome, if we're calling that route good, let's close it down to non-professional car traffic. That's where people are dying https://i.imgur.com/NW3w0OF.png

I agree. So maybe we should improve it, instead of waiting a decade for rolling surveillance cameras to replace uber drivers, which is what I was suggesting

Oh boy! I can't wait for this to start actually operating. Hopefully it's ready in time for us to exit the COVID-19 situation since the only reason I use these ridehail things is when I'm going to be drunk or on drugs or both.

Hell yeah!

Appropriate username.

It would help if they said more clearly what 23% of people not feeling "safe" meant.

Does that refer to driving safety and hitting other vehicles or people? Does it mean personal safety while sitting in a taxi / Uber? Safety while riding Muni?

It makes kind of a lot of difference if that number is being used to justify whatever specific thing that addresses "safety" they're planning to do, doesn't it?

In the article they make it fairly clear that the survey was about people’s experiences driving in San Francisco, not general living etc.

>Does that refer to driving safety and hitting other vehicles or people? Does it mean personal safety while sitting in a taxi / Uber? Safety while riding Muni?

You forgot lidar in your eyes.

It seems like a very logical move. Compare to Gilbert, AZ, SF is more challenging. And of course, there will be way more media coverage. Excited to see full deployment!

The grim joke among cyclists is, "If you want to get away with killing someone just be sure they are riding a bike and you're driving a car."

Paraphrasing, if you want to get away with killing someone with your massive, high-velocity robots just be sure to make them look like cars (and maybe put a human "test driver" in there as a kind of moral "crumple zone" to deflect blame. They don't even have to be good at their job. In fact, it works out better if they're watching TV on their phone when your robot kills someone.)

Automotive death and mayhem was normalized in the USA by a deliberate campaign of domestic propaganda: "The Real Reason Jaywalking Is A Crime" (Adam Ruins Everything) https://www.youtube.com/watch?v=vxopfjXkArM Since then cars have killed more Americans than all the wars we've fought.

A self-driving golf cart that cannot go fast enough to hurt people could be built and deployed safely today. Fast heavy car-shaped robots are a fetish. Please, snap out of it before your robots kill. Again.

(Also, can we please call self-driving cars auto-autos? Please?)

When these go driverless there is going to be a lot of graffiti on them.

Etch bath on windows isn’t cheap to fix.

The perpetrator of the graffiti will be documented extensively, so while this might happen at first, when people realize it's a risky path to legal problems it'll probably die down.

https://twitter.com/SFPDTenderloin/status/134441625641848422...

SF: where you can vandalize police cars for 3 years, get booked 3 times, and keep doing it. Waymo cars are an easy target: you know they won't drive off or run you over.

If the person wears a mask...the camera & lidar footage wont do much good.

You also assume the SF police force will bother hunting down the perp even with a picture for a crime which likely will go unpunished in SF with its effective decriminalization of petty theft & crime.

> If the person wears a mask...the camera & lidar footage wont do much good.

Didn't Sony/Nikon/Fuji go through a scandal a decade or two ago where a camera's infrared-based, low-light enhancer was effectively taking nudes of fully-clothed subjects? All clothing is semi-transparent to IR, to varying levels. A mask may not be an effective way to hide your face from a sensitive IR receptor.

Edit: the camera model was a Sony[1] handycam, sold in 1998. It had a "NightShot" mode that removed the IR filter - activating this mode in daylight effectively enabled "X-Ray" vision.

1. http://www.aparchive.com/metadata/youtube/ad99e1a3930185817a...

Apart from the risk factor of being cought on camera as pointed out by others, I'd like to add that probably there will be few chances to actually graffiti self driving cars.

Autonomous cars have the ability to go places by themselves. So the most efficient way to use them is to have on the streets just the right number of cars based on the demand. Because demand, once the service is fully deployed, is quite predictable (this is already so for services like Uber and Lyft), the most likely scenario is that autonomous cars will be on the streets only insofar as it's necessary to meet the demand at any given time and place. I wouldn't be surprised if in a short time, thanks to ML, they'll be able to predict where and when pick-ups will be requested by users.

Since it's quite likely that autonomous cars will be busy going to pick up users, or transporting users, it is very unlikely that they will give any chance to "artists" to make graffiti.

Moreover, when cars are not used, given the relatively very expensive equipment mounted on them, it's very likely that these cars will be parked in surveiled areas.

So, the way I see it, it is very unlikely that graffiti will be a problem. There might be few cases, but nothing systemic.

And human drivers, who habitually drive at least 5mph over the speed limit, are going to experience some major frustration when these habitually drive 1mph under the speed limit.

I don't think people are ready for this part of it. I foresee dedicated lanes.

Imagine if driverless cars start reporting human operated cars around them that are speeding (excessively) or driving recklessly. Companies like Waze could even build up their own private databases containing the license plates of known “problem” cars they need to be extra careful around. Maybe that’s already happening.

If so I'm probably on those lists from driving and skateboarding around Mountain View. I consider myself an unpaid tester for self driving cars.

Hmm this sounds... familiar....

https://en.wikipedia.org/wiki/Bus_rapid_transit

What human drivers? :)

I can easily see cities becoming self-driving only.

Or maybe human drivers will finally start following speed limits

It's been shown that if you redesign the street, you can change human behavior on that street including lowering the average speed that drivers choose to drive at.

Designing a wide-open 4 lane boulevard then slapping a 30 mph speed limit on it is a terrible plan, yet how often we do exactly that in our cities.

The secret is to design terrifying, narrow roads with poor visibility! What an idea.

Look up "traffic calming". There are plenty of well-studied techniques for discouraging excessive speed that don't involve terrifying the drivers or limiting visibility, and make things safer for pretty much everybody involved.

https://en.wikipedia.org/wiki/Traffic_calming

It's ironic that it's called traffic calming, because the effect of adding chicanes and speed bumps and similar things is often to make drivers annoyed or angry.

Ultimately though, it does result in the desired reduction in speed. So yes, it works.

Narrow streets without road markings is what we generally do to slow traffic here in Europe. They are safer and slower without the need for speed bumps. We have full sections of the city centre without markings under free-for-all and pedestrians can freely walk on the street. Less crashes and safer strees.

No need to make them poor visibility tho. That would be just stupid.

People drive the speed that feels comfortable.

If you design a road like a highway: wide lanes, straight, flat, open, smooth, no traffic coming against you, you'll drive highway speeds.

If you have narrow lanes, narrow areas, trees, rough surfaces, twists, tight portions, raised areas, etc. you'll drive more carefully.

Putting up a sign with a number is just "We tried nothing and we're all out of ideas".

If they consistently got a ticket for exceeding the number on the sign, I'm betting people would suddenly feel a lot less comfortable at higher speeds regardless of what the road looks like.

The reason people drive as fast as they feel comfortable is that speed limits are seldom and inconsistently enforced, so people are more restricted by their own notion of safety than the risk of any legal consequences.

This is (mostly) correct as any country with a significant Gatso camera presence will attest. In the UK people more or less drive exactly at the speed limit at all times because of the profusion of speed cameras and heavy penalties for speeding. (Up to and including a lifetime ban from driving)

While true, this feels like a tangent. In this case, "cars following the speed limit" serve the same traffic-calming purpose as the urban planning choices you're describing.

Sure, if you can't pass and there's someone in front of you going slow, it will force you to slow down.

Most injuries and deaths by car in SF are of people outside of cars, and I suspect that the drivers who maimed/killed those people were usually in front / catching up, not following someone at high speed and had their victim jump between.

I used to live in Maryland, in the DC Suburbs (Rockville, Gaithersburg).

Anyone who has spent much time there, knows that they have wide roads.

Roads that say "Take me, you gypsy stallion! Do 65 on me!" (with 25MpH speed limits).

And they have some very aggressive cops that basically hide behind hedges, waiting for free spirits.

They might not like forcing people to obey the speed limits and red-light cameras. It would be a significant hit on the bottom line.

Maybe then update the speed limits? Definitely much easier than program cars to occasionally go past the speed limit.

Why would there be more graffiti on them?

Right now, it would be easy to go into a parking lot where there's parked cars (aka "driverless cars") and graffiti em, and it's not a very common occurrence for a car owner to get graffitied while parked.

These cars will be even harder targets because they'll have always-on cameras, likely even when parked.

Yes I'm wondering how they're going to deal with vandalism in a city like sf.

Cost of doing business. Pretty much nothing is cheap in SF, I dont see any reason autonomous cabs cant charge 10 or 20 or 50% more than tamer locales.

I'm staying near a main throughway in the mission, and I see Waymo vehicles CONSTANTLY. That said, I really noticed that they picked up the pace about a year ago.

Soon the streets will just be self driving cars driving around each other. Not the best data set.

If (when) self-driving goes mainstream, most traffic will be self-driving cars driving around each other.

These pie charts are awful: https://1.bp.blogspot.com/-HVdEnOjie4E/YC1KNXm4XgI/AAAAAAAAE...

There is no way to tell which colour is "active" without reading the number and mapping it back. Why not just skip the pie chart, it isn't like they draw much attention in the design anyways.

Finding parking seems to be the biggest problem by far, and for good reason

Finding parking is half the challenge. Hoping you windows aren't smashed when you get back to the car is the other half.

This is why I always try to take the crappier of my two cars when I go to SF.

My car's never been broken into when I'm in SF. I just realized my car might be crappy.

Just don't lock your car

That works during the day, but overnight you have to worry about people using your car for various things and leaving various potentially dangerous substances and items in your car.

Could they not still do that if they broke your window?

FWIW I have heard of many people having their window broken, I have not met anyone who had any of the concerns that people have about leaving their car unlocked actually happen to them.

Yes, they could do it either way.

Here's a study from 2019 on the topic of self-driving cars and parking.

> Self-driving cars will ‘cruise’ to avoid paying to park

https://www.universityofcalifornia.edu/news/self-driving-car...

The solution proposed by this study is "congestion pricing".

The obvious solution would seem to be "the owner of a fleet of cars builds a parking garage for them".

That's how you can tell it was a survey of drivers, not generally of people moving about in the city. Parking would not be a concern for the majority of trips that happen without a car.

Parking is not really a problem for self driving cars. Most people will use them as services.

If you want to own one, then its a problem. But by the time individual can buy one I would bet they can just drive themselves around or to a distant location and wait for recall.

> Most people will use them as services.

So we're just shifting the parking problem to a rush hour availability problem? Parking isn't merely an unnatural or unfortunate phenomenon.

> or to a distant location and wait for recall.

Which is essentially identical to public transportation.

> Which is essentially identical to public transportation.

I disagree. It's worse, from a strictly time of transportation point of view, than having your own car, but point-to-point automated transportation with some initial recall latency is still leagues ahead of public transportation.

Oh well, I guess we can all get ready for a future of every car constantly capturing all of our actions and shipping the data off to be immortalized in a Big G datacenter. Having some semblance of anonymity in public was fun while it lasted; I'm glad I got to enjoy some of the last years before it is gone entirely.

tldr

> We’ve now started limited rider testing in San Francisco with Waymo employee volunteers to gather feedback and continue to improve our technology. We’re conducting this testing with enhanced COVID-19 protocols to ensure the safety of everyone involved.

From watching Tesla FSD videos on YT it's clear that they aren't competetive with companies like Waymo that use lidar.

Edit: This comment once had 3 upvotes lol

Waymo drives on (vastly more) restricted routes, so comparing as if they are the same thing is questionable. The problem is several orders of magnitude more simple.

Simple? They are operating fully driverless. They don’t need a driver with hands on the wheel like Tesla. In that, you’re right that it’s not the same thing at all. What Tesla is doing is nothing more than glorified driver assistance.

Since you’re ok with limited routes and impressed by that, Tesla also doesn’t need a driver on limited routes. Doesn’t even need a passenger. I don’t know how you call it driver assistance when there is no human anywhere near the car, inside or out.

Tesla requires a driver. It constantly nags if you don’t keep your hands on the wheel. I can’t sit in the back seat and let it drive all by itself. That’s why it’s driver assistance.

Not for sentry mode. You’re impressed by limited routes, so if you’re going to give Waymo credit for their cargo cult self driving, give Tesla credit too.

You can totally sit in the back seat for sentry mode. Or up to 200 feet away.

Yeah because pulling out of a parking lot 200 feet away is the same thing as driving a 50 square mile city. This has got to be the weakest argument I’ve ever heard on this topic.

That's the whole point; now you are getting it.

"Yeah because driving a 50 square mile city is the same thing as driving on unrestricted roadways in dozens of countries around the world with totally different road markings, traffic patterns, rules, and even directions of travel… therefore we are justified in saying Waymo is more impressive since Tesla is not doing as well as Waymo with an orders of magnitude larger problem compared to the toy problem Waymo is dealing with."

This has got to be the weakest argument I’ve ever heard on this topic.

We both know that Tesla's FSD isn't working yet.

That goes without saying.

Where we disagree is that you think there is something wrong with this picture, when Waymo can solve the problem and Tesla can't. But the problem is a completely different problem, so I am saying the comparison is ridiculous. You are making the comparison as if it is not a ridiculous comparison, when it really is.

And Waymo keeps moving the goalposts. They gave up on Austin. They dropped the term "self driving." They limit their left turns and instead do three rights. They are just dumbing down the problem as they go. Easy path, I guess.

> Yeah because driving a 50 square mile city is the same thing as driving on unrestricted roadways in dozens of countries around the world with totally different road markings, traffic patterns, rules, and even directions of travel…

Except it doesn’t work in the unrestricted conditions you mentioned. It’s going to take years or even decades to make it work everywhere, if it all it works. You don’t get points just having a grander vision.

Meanwhile, they keep moving the goalpost on when FSD will actually arrive. They’ve been working on it for years and yet they’re nowhere near confident enough not to require hands on the wheel. That tells me everything.

Anyway, this is an argument that’s going nowhere since you seem to undermine by saying Waymo is solving a “toy problem”. So let’s agree to disagree which one is more impressive.

Moving the timeline is not moving the goalposts. But I agree they have moved both the goalposts and the timeline. I'm just not agreeing that Waymo's progress is more impressive than Tesla's progress. The problems are too different.

It's also several orders magnitude more simple to say, "our car is reliable most of the time, but drivers must always be ready to take over at a second's notice (or less)".

As long as final responsibility rests on the human driver, it's just autonomous testing, not true self-driving. When Tesla owners are allowed to take a nap as their car drives itself, that's when real self driving has arrived.

We all already know that though. The point is, they have different approaches to gradual rollout. Waymo is going narrow and deep; Tesla is going broad and deep.

Narrow and deep gives better naive demos earlier in the calendar. Like, now. Broad and deep wins later.

Meta comment: one thing I find particularly interesting here is the 4th point on their infographic - 23% of people don't feel safe getting around SF.

What amazes me is how SF responds to the endemic issues that are slowly destroying the city. Not be fixing the core issue, but by making it a startup opportunity. Cars getting broken into and muggings happening when you're on foot? Lower the barrier of entry to car transport without ownership, advertise the safety aspect of it. It's mildly surreal.

Tragically, people are injured and killed in muggings, but this is actually fairly rare, although it's a terrible experience to get mugged even if there's no injury. Hopefully we can work on building a world where no one is that desperate.

Even more tragic - because it isn't rare - is getting mowed down by cars. People get killed or injured by the dozen in car crashes in SF. It's insane that anyone with a license can just show up and drive in the Mission/Tenderloin/Downtown.

Not good enough for San Francisco, car property theft even happens now while the cars are being driven by people:

https://abc7.com/car-robbery-traffic-thief-jumps-out-of-robs...

I don't think they mean people worrying about muggings or breaking into the cars. Quote from the article "Worryingly, nearly a quarter didn't feel safe on San Francisco’s roads at all".

That for me heavily suggests that people don't feel safe while driving, due to most likely general chaotic environment of roads there.

It says in the article that these questions were specifically in reference to driving around San Francisco in a car, not general safety.

"SF" isn't the one responding to bad transportation with self-driving car efforts.

And SF is making progress with bus lanes and bike lanes and a new subway. Just...very, very slowly.

People need to stop believing in Hanlon's Razor.

When you have a hammer, everything looks like a nail.

> Meta comment: one thing I find particularly interesting here is the 4th point on their infographic - 23% of people don't feel safe getting around SF.

Be very careful when considering 'facts' like these. It's very easy to shape the response of a survey by asking particular questions in a certain order. Don't assume anything like this is actually true.

Does one really refer to people living in SF as San Franciscans?

Yes. New Yorkers, Chicagoans, Bostonians, San Franciscans... pretty common thing.

Why is "New Yorkers" in your list?

I believe it’s Sans Franciscan.

Right, it pluralizes just like Attorneys Helvetica.

Friscoians.

No. Not that one.

Do you have a better demonym to suggest?

Yes, that is the correct term.

Wikipedia lists the demonym for nearly every major city.

San Franciscan

San Francisqueño

Wow that's cool! TIL Savannahian, Atlantan

I think so!

Some of the first DuckDuckGo results were from out-of-town media, but here is a very, very local source (KQED):

https://www.kqed.org/coronavirusliveupdates/news/11856076/he...

23% of people don't feel safe getting around San Francisco -- wowza.

Worth nothing that the article clarifies that this is in reference to driving — as in "nearly a quarter didn't feel safe on San Francisco's roads at all"

One would need to know how this compared to other cities. As someone who used to visit SF a lot (though rarely drove much in the city),I'm not sure how it's notably bad. Sure there are the hills. But with an automatic transmission, that doesn't really make a big difference. And it's mostly grid-ish. It's not a particularly easy city to drive in but compared to NYC, Boston, DC, Seattle, etc., I'm not sure how it's especially awful.

I love how "on the roads" is taken to mean the same thing as "driving a car", rather than the vulnerable road users outside of cars who are hurt and killed the most on SF streets.

Is that traffic safety or other safety ? I honestly go to SF all the time, and it doesn't feel unsafe at all, unless I am in TL and some parts of SOMA. Places like Marina, Crissy Field, North Beach, Valencia, Noe, Richmond, Sunset just feels like beautiful suburbia with nice retail corridors, with a ton of people out and about.

I have to assume it is about general safety on the streets. Traffic-wise I've never felt unsafe there, but the article's claim about safety is ambiguously worded. It says "Worryingly, nearly a quarter didn't feel safe on San Francisco’s roads at all." But if you look at the question they asked in their infographic, I think people would have answered with general safety in mind more so than just road safety.

As for feeling safe - I used to also feel that most of the neighborhoods are 'fine' (although they seem to increasingly have trash and tents). But then I was physically attacked by a drug addict unexpectedly in broad daylight. And then had friends held up at gunpoint. And now if I ever have to visit SF, I am on high alert and can't turn off being on guard. I feel sad because I used to love SF but now I try to avoid it, both for personal and business tasks.

It says in the article that these questions were specifically in reference to driving around San Francisco in a car.

That's around the number of people who've lived all their lives before moving to SF in suburbs I would say ? Coming from any other big city, San Francisco doesn't feel more unsafe.

Coming from Chicago, SF felt a lot less safe to me. Would not even consider living there even though I've lived in a big city for 5+ years.

Chicago has a homeless population, but it's nowhere near as large or as emboldened as SF's.

Having lived in Chicago and SF, one big difference is how the problem areas are distributed geographically. In Chicago, the high foot-traffic areas (eg the Loop, Logan Square, Wicker Park) where tourists are likely to end up are very nice; you only end up in problem territory if you go wandering off into the Westside or Southside. In SF, the problem areas are the high foot-traffic areas where tourists are generally going: Union Square, anywhere touching Market, etc.

In the past, I probably would have still felt safer walking around eg the Tenderloin than some areas of Chicago's South/Westside. But beginning a few months after the pandemic began, the streets in SOMA/FiDi/Union Square emptied out (no more tourists or commuters) and SF has started to feel much, much less safe in those areas. In the last few months, my partner has been attacked on the street twice, and I narrowly avoided being mugged with a screwdriver in Union Square itself. That is much worse than anything I ever experienced in the Southside of Chicago.

It's a different world if you just avoid the eastern half of the city.

I think a lot of people, especially in the tech industry, get a somewhat distorted view by the cesspool near the Moscone and Market Street. Not that there aren't problems elsewhere but the area that conventioneers are exposed to is especially awful.

Coming from NYC, San Francisco feels much more unsafe. I’ve been randomly attacked by a deranged stranger on a busy street (16th and Guerrero in the Mission). I never felt so paranoid when walking around NYC.

> Coming from any other big city, San Francisco doesn't feel more unsafe.

Speaking from years of living in Shanghai, that didn't come close to passing the laugh test.

I wonder what subset of that is folks who are afraid of the virus and “don’t feel safe” taking public transportation or human driver-based ridesharing.

Is the waymo approach as economically in feasible as people claim?

Human labour is by far the most expensive aspect of taxis. If a tax driver earns 35,000 USD/yr, a self-driving car only needs to cost < 35,000/yr to turn a profit at scale.

IIUC, FSD costs are mostly captured by the running service, and the lidar equipment. The lidar equpiment is < 20,000/car for sure, and I can't imagine it costing 15,000/car to service it using waymo's ride hailing platform at scale.

And don’t forget that a human driver needs to rest. A FSD cab can operate 24/7 minus maintenance downtime.

And then on top of 24/7 operation if the lidar equipment can last at least a 5 year life time, it'd be about <$4000/yr/car since its amortized over the life of the lidar equipment.

Very good point. Also usable for any other transport activity (in near future?). Last mile packages, etc.

At least in places where it doesn't snow.

Uber and Lyft are two data points that ride hailing is economical, by market definition.

Which company's monumental and increasing net losses lead you to this conclusion?

Both are on track to be profitable this year. Lyft probably would have been already if not for Covid.

Lyft is "on track" to profitability in the same way that I am on track to be the queen of France.

Well then you'll have to start going by her highness starting in Q4 2021.

There's a market for taxis. And there's almost certainly a market for taxis with good ride-hailing apps (as well as private cars). But they're mostly a product for business travelers and well-paid professionals in urban areas. But I fully expect the VC-funded 50% or so subsidies to go away.

Profitability doesn't matter, revenue does in there being a market for ride hailing. You have to assume that Waymo doesn't have the cost of paying drivers income.

Why wouldn't it be economical?

Presumably because the cars would be substantially more expensive than an ordinary car due to all the high tech packed into them. Assuming that the components of the car have a finite lifetime, it might be possible that the overall cost is so much significantly more than a normal car that they won't be able to compete with a human driven solution like Uber.

Does anyone have any actual statistics/calculations on this sort of thing?

I have no good collections of statistics to share but the price of lidar seems to be dropping.

They built the most expensive solution (significant dependence on LIDAR) that scales the poorest (requires centimeter-precision 3D mapping of every street it drives on). Google Maps cars would need to do significantly more than they already do today on an even wider scale, and to be profitable it'd likely have to inundate you with ads on top of being the most expensive service.

...Meanwhile, the ADAS in many new consumer cars can pretty much drive itself already.

Waymo has to convince you their way is the "only" way to do self-driving cars, because nobody in their right mind would chose their solution otherwise.

> ...Meanwhile, the ADAS in many new consumer cars can pretty much drive itself already.

Yeah, not really. "Pretty much" drives itself isn't a thing. It either drives itself or it doesn't. Which ADAS system allows me to sit in the back seat while it drives itself fully?

> Waymo has to convince you their way is the "only" way to do self-driving cars, because nobody in their right mind would chose their solution otherwise.

Considering every self driving car company except Tesla (maybe MobileEye but even they have backup LIDAR systems) uses the exact same approach as Waymo, this is way off the mark.

There's no inherit reason why lidar can't be cheap afaik (and it's already getting there - prices are dropping pretty rapidly).

The cost of manufacturing 10,000 lidar units is much different than the cost/unit for a batch of 10,000,000 units.

Just look at EVs and the drop in battery cost over the years. Lidar being cheap at scale is just a matter of when, not if.

Convince me the customer? Why would I care how they do it as long as it works?

They seem very convinced, or at least are pretending to be convinced, that they can build FSD without human-level AI. I'm not convinced of that at all, though.

As a human with a driver's license, I'd say the job of driving a car demands ~1% of my mental system features.

Here's a few skills of my brain not needed for driving,

- compose poetry and alternate lyrics to "baby shark"

- recognize emotions on faces

- form opinions about music, movies, video games, etc.

- have a conversation

- remember that time I embarrassed myself in sixth grade

- contemplate my own existence

- divide other people into friend/enemy groups

- etc...

You still might need a functioning "vertical slice" of human-level AI to implement FSD, though. Leaving out irrelevant tasks doesn't make the one remaining task any easier.

The question isn't if driving the car requires you to know how to write poetry. It's if complexity of how our brain works, that enables you to be able to write poetry is also requirement to drive a car in real world, unpredictable environments.

> recognize emotions on faces

I'd argue this is a required skill occasionally, at least when you find yourself in a 4-way stop deadlock.

I find it depressing that cars aren't already talking to each other. I drive on a lot of winding, rural roads with zero visibility around the bend; I'd love to know that there's a car coming the other direction, even with no driving assist technologies at play.

Don't even need two way communication for that I think. A radar should do the trick (although the example video only shows it from a stationary vehicle) https://engineering.princeton.edu/news/2020/06/24/new-radar-...

In contrast, also as a human with a driver's license, I find that having a conversation conflicts pretty severely with the job of driving a car.

I'm not convinced either. I don't think I'll see a 100% FSD vehicle in my lifetime, at least the ones depicted in futuristic renderings without steering wheels.

Case in point, how does a FSD vehicle deal with extremely heavy snowfall on a unplowed road? I spent 15 minutes rocking the car back and forth, so I could slowly back my way out of a one-way road onto a main street. After that I had to drive on countless roads with no visible markings due to the snow. I don't think the way we're approaching self driving cars today will ever work for these edge cases.

I'm the opposite. I don't think I'll buy another car (current is 3 years old) and I'm even surer that my son (aged 2) is not going to learn how to drive.

Driving is a pretty well defined space. Stay in your lane, get to the destination, and don't hit anyone. Pretty much every similar problem that significant effort has been solved. Chess, Go, Starcraft, DOTA all were called unsolvable at one point and all were eventually solved.

Your edge cases are just that. Edge cases. Nobody is going to care if they can't drive in extremely heavy snow. To begin with, they probably shouldn't do that anyways. More importantly the vast benefits of self driving cars outweighs the inconvenience of not being able to drive a dozen times a year. We all still have cell phones even though you can't make calls in elevators.

Even then, if you can do it why can't a computer? Rocking the car back and forth seems doable. Following an unpaved road is much harder, but then again are humans actually good at it either?

You might be right that they will still have steering wheels for no other reason than there's little downside to including them.

Starcraft was solved in the sense that the winrate against top players was high, but the production model often found itself in pathological absorbing states that it couldn't escape from, which no human would ever get itself into.

That's important, since if we are using it as an analogy for driving, it's not just the win rate we're interested in, it's the ability to consistently avoid pathological states when presented with edge cases (e.g. not braking when there's a strange looking object in front). Some would call this "common sense".

Also, games like SC2 are distinct from the driving problem as far as RL goes, since you have unlimited training data which allows you to explore all edge cases given enough compute. In self-driving, the data is scarce and you can't explore all edge cases by just throwing compute at the problem.

>avoid pathological states

These just need to be relatively rare and not dangerous when they happen. If my car slams on its brakes every 100,000 miles for no good reason that's not really a big deal. At least it's not big enough to stop them from being used.

>unlimited training data

Unlimited is not possible, but we can get tens of millions of miles pretty easily. Then many many more as they roll out into production.

Edge cases can crop up while you're en route.

My go-to example: you're driving at night, two-lane road, and there's a minor accident in your lane ahead. You crawl forward to the point where you can pass the accident, and then what?

Someone involved in the accident is directing traffic. In the dark, with no safety gear. On the other side the traffic is backed up, so you see headlights facing you in the lane you have to use to pass the accident.

To me, that seems like an insanely difficult problem for a computer to deal with. A human driver, that's easy, but a computer?

(Having said that, I'm historically terrible at predicting just how fast technology is going to advance, so it's safest to assume I'm wrong and this will turn out to be easy.)

For edge cases this rare (and brief), there can be a small team of remote operators that take over, using the streamed camera data to decide when to tell the car to venture into the other lane.

The remote driver could also call the passenger's phone to double check that the sensor data is correct, and to reassure them that the AI isn't just guessing.

> Your edge cases are just that. Edge cases. Nobody is going to care if they can't drive in extremely heavy snow.

The problem when you have 1.4B vehicles in circulation is that your edge cases happen dozens of time per second.

> Nobody is going to care if they can't drive in extremely heavy snow.

A loooot of people will care. Snow isn't an edge case in a very large part of the world, it comes every year and stays for weeks or months.

What we're going to have is a system that's slightly better than existing "smart" systems [0] but much worse than "fully autonomous". People who will learn on these systems will suck at driving and be in complete panic mode at every single of the "edge cases" they'll face very often

People have to understand that the PCH from SF to LA and quiet Mountain View streets are the edge case to the rest of the world.

[0] ESP/ABS, collision avoidance, blind spot warnings, auto emergency braking

> Chess, Go, Starcraft, DOTA all were called unsolvable at one point and all were eventually solved.

Depends on what you consider 'solved'. AlphaStar was never able to demonstrate the level of superhuman performance in Starcraft that AlphaGo did. It got to grandmaster level, but even that was with significant mechanical advantages over humans, when what they wanted to test was the tactical/strategic thinking, not raw ability to issue basic commands (which an AI will easily win at, obviously).

That said, AlphaStar was vastly superior in performance and well-roundedness compared to any previous Starcraft AI, so it was still very impressive. It just wasn't "beating the best in the world at strategic thinking" impressive.

Solved meaning it was way better than the average Starcraft player. If we can get driving to even average levels that will be a huge win because it won't get distracted or drunk.

Yeah fair enough.

Though I'd like to see AlphaStar perform with comparable mechanical limitations to humans as well. That's a complete non-issue for Go or Chess, but for Starcraft it matters a lot.

Self driving cars will have a mechanical advantage too. They can go to maximum braking essentially immediately. It takes time for humans to move their legs and then some time for the pedal to move. Probably a 200-500ms advantage for the AI.

> Case in point, how does a FSD vehicle deal with extremely heavy snowfall on a unplowed road?

90% of the cars on my street didn’t try to deal with that this weekend, and most of them would have failed if they tried. None of them are FSD.

Two thoughts from your comment:

1. Agree on the edge cases, to do real driving we'll need AGI or some whole new way of looking at it, its not possible algorithmically. But, I would pessimisticly draw on other examples of how tech handles this problem: if the tech can't do it, it's not something people get to do. Try getting help from a chatbot, or getting help from uber or Amazon on a situation that doesnt fit the FAQ. So I expect self driving cars will be great for a range of things that young tech-savvy people want to do (and that google et al want you doing) but won't drive you to the hospital in a snowstorm and will balk when they realize they have to turn into a rutted backroad at night, because why would anyone live there?

2. More frightening, there are people (see one of the siblings) that think nobody should be doing "dangerous" things like driving in a severe storm, that are beyond the range of a self driving car, and want to force their value judgement on others. So expect there to be calls for outlawing certain activities, for our safety.

Do they have to operate in extreme weather? If they work for 99% of days and they are 100x more convenient than owning cars, I'm sure we would deal with transit during extreme weather some other way.

Trains don't have a steering wheel :)

But it never snows in Texas or Portland, OR so they'll be fine there :-/

well, many people driving around also don't reach the "human-level AI" too :) The computer beats me in chess, while it still would fail to find the proof i came up with in my Math MS thesis. Is the driving more similar to chess or to finding a math proof? I think it is closer to the former.

In a few years, giving the speed of sensor development and growth of computing power, the automated driving system can continuously capture and monitor 360 degree environment, at better resolution than human and across the specter of wavelengths (visual, IR, T/GHz) and react better/faster than human. Given that, i think human driven cars will be relegated in couple decades to the position of horse carriages - not allowed on highway for example where automated cars will be driving faster at tighter distances while being networked to each other in the vicinity. In some sense the driving may happen to be best suited for insect style distributed intelligence with relatively low intelligence of any given node which still possesses heightened senses. I wonder whether Google cars communicate to each other when meet at an intersection - either option has far reaching consequences and will define our near-term future.

Plus at some point insurance industry will say their word which most probably will be based on the sole metric of $/mile loss, not AI IQ score.

I'm not convinced either.

On the other hand, if we set it up well, each car will learn from the experiences of every car, so novel situations will (ideally) only be novel once.

R.I.P. Elaine Herzberg

I've never liked this company.

your username is waymo near to theirs than any other I've seen, so I can understand why

Reply on news.ycombinator.com