back
179 comments
This isn’t like the game of Go where processing power eventually wins. Driving in a major city, based on intuition, social cues, and the ability to drop a plan and make a new one, is the definition of too complicated for a computer. We could have public transportation, but we just can’t imagine a world where someone doesn’t privately own a thing someone else rents.
> This isn’t like the game of Go where processing power eventually wins...[it's] the definition of too complicated for a computer.

Isn't that sentiment how people used to think about computers playing Go, writing an essay, beating humans at Jeopardy, or transcribing spoken language? Betting on the side of "computers will never be able to X" seems like a generally bad bet.

> We could have public transportation, but we just can’t imagine a world where someone doesn’t privately own a thing someone else rents.

The concept of self driving cars isn't mutually incompatible with public transportation.

> we could have public transportation

> we can’t imagine a world where someone doesn’t privately own

I love public transportation and my favorite cities are ones that have the best public transportation.

But most of what I like about it is great when “visiting,” not living. When I’ve lived in cities that had great public transportation it was occasionally nice; I could go downtown and get drunk without worrying about/coordinating transportation.

But commute times on public transport are absolutely dreadful. Unless you win the lottery and your frequent destinations are along a single route with few stops, you’re transferring and waiting for a bunch of other folks.

And cities where public transport works well are densely packed, making getting downtown from the edges any other way difficult too.

Unless you’re talking about converting privatized point-to-point travel to public transportation, which sounds absolutely awesome and autonomous driving might bring closer to being a reality, I’d say I’m rather fond of going directly to my destination and back in less than 10min of drive time.

Edit: realizing from the comments here that many people’s experience with using a vehicle and public transport is a daily commute. I work from home and I don’t drive during “rush hour” when, for some reason, society pays to migrate an entire city full of knowledge workers from their home to a computer.

Living in the suburbs and not participating in daily commute culture looks a lot like “winning the lottery.”

An example of how I use my car: I choose to go to the grocery store with the best supply chains in my area and it adds ~5 minutes to my drive compared to the one that’s “walkable” as long as it’s not rush hour. I place my order on my phone, pull up, open my trunk, a store employee puts the groceries in, and I drive home.

That’s a hard baseline to compete with, especially for anyone who has a family they have to coordinate this around. I’d much rather stick my kids in a car seat and do the above than whatever the equivalent would be with public transportation.

To be fair, if they crack automated driving it'll revolutionize public transport too. Imagine buses routed dynamically, and sized perfectly to match demand. Or other automated, private cars on the road yielding to public transit vehicles.
> This isn’t like the game of Go where processing power eventually wins

It turns out transportation is not a zero sum game. There's no "AlphaGo beats Lee Sedol" watershed moment for AV. Instead we progressively work towards a good-enough level of AV driving that minimizes the risks compared to human drivers. Transit will always have accidents and deaths, even for public transportation, and we try to reduce that risk over time with new solutions.

Personally I'd like to see the evidence that Cruise and Waymo are riskier than human drivers in SF. My anecdata is that they drive by my house every 5-10 minutes during the day and are the best drivers on my street. Just yesterday, a Cruise slowed down nicely to edge into the other lane while I was getting into my car on the street. Most humans (including Muni drivers) speed past, inches from me.

In the case of this accident, sounds like the fire truck hit the Cruise. That actually happens a lot with non-AV cars. Emergency vehicles go into a red light expecting the green side to stop and they don't. So let's learn from it. Did the truck have its siren going? Could the Cruise car see the truck around the corner (it was at Turk and Polk)? Could the truck see the Cruise? What could the Cruise detect that could have indicated the truck was coming?

That's the kicker. We've had a solution for ages.

I've spent a good decade trying to poke holes into "why 'acthually' public transportation can't work in the US", and I've come out of it even more confused. The US has all the hallmarks of country that's well suited for public transit. Urban housing shortages, open space, flat land, pockets of density, overdue infrastructure overhaul, you name it. It is really THAT simple. There's so much low hanging to pick off and best-of-all, public transit spending goes back into the pockets of

That being said, you can't convince a person to like anything if they're reasoning from a place of comfort. I'm a broken record, but hearing an American praise European cities while simultaneously doing everything in their power to stop their city from ever becoming that, is the sort of hypocrisy that cannot be solved through convincing, logic or education.

Lastly, public transit also needs a strong police force, and there is some much needed 'cognitive dissonance' that YIMBYs need to figure out if they're going to convince those holding out.

_________

Prevention is better than a cure. Driverless electric cars are here to save the automobile industry at the expense of the environment, climate and public safety.

> Driving in a major city, based on intuition, social cues, and the ability to drop a plan and make a new one, is the definition of too complicated for a computer.

It's also too complicated for humans, though. They get into accidents all the time.

I feel like we're holding computer drivers to a higher standard than the humans we already allow to drive. The computers are absolutely going to make mistakes, and some of them will have catastrophic results, but, uh...

>too complicated for a computer

Too complicated for most humans too. I've seen human drivers t-boned by fire trucks and ambulances. Have had near misses myself a couple times. The solution is not "intuition and social cues" but better technology and instrumentation and especially Vehicle-to-Vehicle Communication (V2V) which all emergency vehicles (and eventually all vehicles) should have.

I don't think you understand what public transportation is.

Public transport is a local monopoly where profitable routes are used to subsidize unprofitable routes in order to guarantee base level of price and service to all residents. Public transport is not linked to any particular ownership or operating model, nor to any specific vehicle type. Sometimes agencies own and operate a fleet directly, other times they contract out operators and sometimes contract out whole fleets.

Self-driving vehicles are just another tech they'll have to incorporate the way they have been incorporating shared bicycles or scooters. In the long term, large bespoke expensive vehicles (busses, trains) requiring special operators and infrastructure will not be able to economically compete with mass-produced Autonomous EVs, so they'll replace much of today's fleets. Thinking that AEVs and public transit are at odds with each other is nonsense.

For sure. We often drive robotically, with our minds on a conversation, a podcast, or a daydream. That makes it easy to think a robot could really do it. But then there are the moments where driving requires our full attention, our knowledge of the world, or our ability to quickly read social situations and social signals.

As the article shows, a system that works 99% of the time won't cut it. Ditto the recent story where a robot car drove into wet cement. Or a zillion other edge cases, some of which will be fatal.

The problem here is that Cruise launched prematurely, not that automated driving is intractable.
Beating Go required dramatically improved algorithms. Simply increasing processing power was not sufficient.

I think Chess would be a better example - IBM was able to beat Kasparov with the Deep Blue super computer simply through hardware upgrades to essentially the same model. That system was searching 200 million moves per second, whereas with improved search algorithms modern (and better) chess engines beat GMs at orders of magnitude fewer moves per second on simple PCs.

The thing that gets me about self-driving vehicles is they just stop and block the road when they're confused (unsolvable situation, network disconnect, etc.) We can snicker at Cruise's meltdown during Outside Lands, but I keep thinking of the videos of people driving out of the Paradise fire, or more recently the Maui fire. One confused vehicle could kill dozens, hundreds, even thousands of people behind it.
It's like the article a few days ago about the Air France 447. We expect a competent pilot to form a situational awareness model, but then discard that model when confronted with novel information. Sometimes they have to do this in a matter of minutes, other times it's just a few seconds, in the worst case sub-second accuracy is needed. It remains an unsolved problem in both the human and machine worlds.
At the same time, other companies have cars self-driving around cities for years outside of the headlines.

I imagine after we all start just using an app to summon one of those cars to go around, there will still be people quick to call self-driving impossible every time something interesting happens.

(Reminds me of solar energy.)

Many places are probably past the point where neighborhoods could be connected with public transit simply due to how they're laid out and how sparsely they are populated. Cities though I agree. Extreme individualism and convenience probably.
Regulators are playing way too fast and loose in this case, but given the number of human drivers that collide with firetrucks - just drop the "for a computer" from your definition.
While Go is theoretically calculable with infinite processing power, that is not the lever with which the best algorithms have claimed dominance over others.
Couldnt agree more, well said
Why is it always Cruise in these incidents? From afar, it seems that Waymo is generally safer & more methodical in their rollout while Cruise keeps announcing new cities before they've quite solved things. I get that they need to be bold if they're going to try to compete, but is it worth injuring people and creating negative press for the industry? ...Ultimately I'd love to see multiple players in this market, but I feel like Cruise has jumped the gun a bit here, and I am afraid there will be wider backlash.
There's a risk/reward trade-off to be made.

Waymo had a huge head start, a ton of advantages, and more cash to burn. So they chose one operating point, under the theory that public perception and trust were vital to get right.

Later entrants chose a different operating point, because they had to catch up and had more limitations.

This is why Tesla rolled out autopilot before it was ready and killed a bunch of people; getting data collection running at scale was the only way for them to compete. Keep in mind that Musk promised a million robotaxis within a year of their roll out. That didn't happen.

I've ridden both and Cruise just sucks. Riding in their car is terrifying.
So both Cruise and Waymo are subsidiaries of giant corporations rather than VC funded and needing to push tempo before they collapse. I don't think GM is particularly less risk adverse than Google.

Maybe the depth of the respective corporate coffers and the power of the executives backing the respective subsidiaries? I'm pretty sure that Waymo is supported directly by the Google founders, and I don't even think there's an equivalent person at GM to guarantee Cruise's continuation if they don't show results.

I'm not going to comment on Cruise specifically, but keep in mind that AV companies have an insane burn rate in the order of 9 figures yearly or more, without any meaningful revenue at this point in time. They're under a lot of pressure to show some kind of growth, especially considering the current environment in which money isn't as easy to get anymore.
Yes, if I were Waymo I would be so pissed. It looks like protestors and regulators want to lump the two together. I'm pretty supportive of careful public testing of AV, but even I look at these recent Cruise incidents and find it very understandable that a city would want to kick out these AVs.
Sounds like it’s been a bad week. Tuesday a Cruise car drove into wet cement and got stuck.

https://www.businessinsider.com/driverless-cruise-car-stuck-...

That's not great, but crashing into a fire truck has the potential to injure other people who need firefighters.

Maybe it's time to pump the breaks on self driving cars.

And they were randomly getting stuck around the concert.
This is what happens when you don't solve a CAPTCHA correctly. People get hurt.
I know this is purely anecdotal but Cruise cars have begun testing in my neighborhood in Austin (they've been in the city for months already) and in two separate incidents on two nights this week I've seen one completely roll through a stop sign. In the first it seemed to know there was a stop sign because it was going along at the speed limit, slowed down like it was going to stop, and then just didn't stop. I was driving through the intersection and I had to move into the opposite lane to avoid the Cruise car. The next night I was just out for a walk and observed the same behavior at a different intersection. I've had several unpleasant run-ins with these cars over the last few months. Not sure how to feel about it all.
You feel these cars are not safe but for some reason abandon your observations to be open minded or objective. It’s crazy how far the public utility commission is willing to jeopardize public safety in the interest of autonomous experimentation.
I believe you. A Cruise car went out of turn at a stop sign in SF and almost hit my car. Folks do too, but if these AVs are just as bad as human drivers, what is the point?
It has seemed to me throughout this year with the contentious CPUC ruling that Waymo has made a huge strategic error by not denouncing Cruise. They should have totally thrown Cruise under a bus, but it's too late now. The press and the government simply conflate them as "Cruise and Waymo" or "the AV industry".
How many driver-full cars collide with fire trucks each year?
Interesting that 100% of the instances I've seen are Cruise. Has anyone actually seen a Waymo vehicle in a significant collision?
I know the trope "Americans spent millions inventing a pen that works in space, and Russians just brought a pencil" isn't entirely accurate. But, I can't help but get similar vibes as Americans spend billions developing self-driving cars, and as the rest of the developed world just builds trains.
Okay, but this isn't a Cruise car colliding with a SF fire truck - it's a fire truck colliding with the car! Now, it's still not good that they didn't yield to the fire truck properly. But that's a very different level of fuckup than "ran into a fire truck".
It's amazing to me that driverless cars seem to have so much trouble with emergency vehicles. They already know how to avoid other cars; seems like another car that's very big, very red, and very blinky would be even easier to avoid.
Strange because around there SF fire trucks are super cautious: many human drivers ignore them and keep driving. And there are often random humans in dark clothing shuffling around in the middle of the streets. This is not an area where fire truck charge through. It would take some effort for a human driver to get hit by a fire truck.
It seems this is one of the situations where just focusing on the immediate area around you is faulty.

An emergency vehicle with lights on has the right of way through an intersection even if the lights don’t match.

However, you would probably need longer range cameras and object detection to identify an emergency vehicle with lights approaching the intersection.

AI is becoming more and more human every day! Though, honestly, at least these driverless cars are backed by billion dollar companies, and can pay for the damages. I've seen plenty of less than savory people in junky cars nearly hit emergency vehicles, and they would've never had insurance to cover the damages.
I wonder how big of a fix it is to respect emergency vehicles. Like, is it tweaking a few weights? Or is massive retraining necessary?
I hope this car will go to jail for some time
> One passenger (…) was treated and has non-severe injuries.

Injury attorneys will make those injuries cost couple millions.

Do these cars have microphones listening for cues like sirens?
maybe it's time to add ears to driverless cars? It could have heard the siren and stop until it went away.
Past time to mark emergency vehicles with beacons, and force e-cars to respect them. This seems an important pain point to address early.
This is a technical use of the word "collides".

The Cruise car failed to yield, continued at a green light, and was struck by the fire truck. So presumably it did not detect the truck.

It is sad that automated cars are being asked to detect emergency vehicles at all.

A system that signals the location of any actively flashing vehicles to all nearby automated cars is technically achievable. It should have been one of the first steps in the development of any of this technology.

Unfortunately it requires careful coordination between many different entities so it hasn't happened.

Edit: Rarely have I had a comment so misinterpreted. I am not excusing Cruise in any way. This is a major design failure for the entire industry.

For clarity, automated cars should be required to have a rock-solid radio-based system that is aware of the position of any nearby emergency vehicle.

All emergency vehicles should have a transponder to signal their location to automated cars.

There's no reason to be relying on visual detection of flashing lights for a fire truck. A fire truck costs most of a million dollars and a transponder is not an expensive requirement.

There should be a standard for radio frequency announcement before automated cars are rolled out and there is not.