Apologies for repeating myself, but it's hard to convey how far ahead Waymo is in every area: sensors, vehicles, training data, markets served, simulation, infrastructure, operations, regulatory.
They're growing fast, but I do wonder what they see as the bottlenecks: municipal approval, training, manufacturing? It's really unfortunate that lobbying is slowing self-driving rollout in some of the biggest and most important cities: NYC, Chicago, DC.
I just took a Waymo in San Jose, having taken a few from San Francisco - it dropped me off in a place that would have made sense in SF (across the street from my destination at a light, avoiding a U turn) but was terrible in the suburban streetscape where the road was literally 10 lanes 4 / 4 / 2 turning and no one would ever drop someone off like that instead of U turning so they didn't have to cross scary giant traffic.
So weird little things for scaling, but overall it was a good ride till that point. Also, the dropoff was at a hospital/clinic and Lyft for the return trip had no issue with going into the property to the actual destination.
That seems like something not hard to fix, I am surprised they haven't already addressed it. Uber, as a comparison, absolutely narrows down the drop-off & pickup to the side of the road you want. I would demand the same of Waymo.
Waymo also lets you pick, but only from a predetermined list of pickup/dropoff locations. Sometimes there's not a spot on the side of the street you'd want, other times it will default to a spot on the other side of the street but you can override.
So are they? I know that tesla is behind, but they have so many simulation miles, several generations of their own hardware, and lots of people behind the wheel taking over.
I've also seen the autonomous waymo vans, which seem nice too.
Is there a "theoetical" ahead and a "practical" ahead?
How do the finances look and does financial solvency count? will someone run out of money first?
There have been two waymo articles (about sensors, and now about hardware), is this a sign they are worried?
Is it unfortunate, or just a different priority in a different place?
Those are also cities with a great public option for transit (unlike, say, Dallas or the Bay Area where different municipalities squabble). One might laud New York driving people away from car culture when the subways/buses can serve more people, greener.
Taxis and barely-legal app taxis both require a human to drive. Someone who's earning a living doing this, and paying taxes to NYC. Waymo is driven by a robot, owned by a company out in California, who has a lot of lawyers doing their best to avoid as much financial and legal liability as possible. Not that Uber and its ilk isn't doing the same thing but they've still got a tax-paying citizen in the loop.
The current mayor of NYC is a DemSoc so he might have some ideas about keeping NYC's money in the pockets of middle-class NYC citizens.
I don't think all of the economic benefit stays in NYC even when you take a yellow cab. A significant share of NYC taxi medallions are owned by entities based outside the city. Marblegate Capital in Connecticut owns about 2,146 medallions—nearly 16% of all NYC taxi medallions. They make more profit off the medallion financing business than actually operating the taxis which is basically break-even AFAIK.
Nobody is keeping money in anybody's pockets. What you are advocating for is a tax on NYC people to force them to use human drivers instead of new technology. Just like the toll booth operators of the 90s.
It is so disappointing to see the tech class advocate for laws that keep jobs that are unhealthy and lead to nowhere (cabbies) instead of moving the country forward (and those people getting other, hopefully better, jobs). But again, that type of short-sighted, zero-sum/socialist thinking is becoming more common, sadly.
No contest when the bus is full, but the overall efficiency (/"greeness") of the system is much less clear considering: buses are rarely full, most buses are diesel, heavy buses tear up the pavement on their routes (especially near stops), and buses don't take riders directly to/from their destinations.
Most of the city buses in my UK home town are now battery electric double deckers. These carry up to 85-90 people (with 25 standing), weigh 11 tonnes, and is 10.5m long.
A typical jaguar Waymo usually carries up to 4 people (usually just one), weighs 2.1 tonnes and is 5.2m long.
So the double decker bus should be more environmentally efficient with maybe 5 passengers making independent journeys, though operational costs for paying a driver probably make the Waymo more financially efficient in that situation.
I suspect robotaxis will become very important in suburban areas with distributed jobs where buses don’t work well. In busy cities there simply isn’t the roadspace to replace buses with robotaxis (or any other form of car) for commuting.
> A typical jaguar Waymo usually carries up to 4 people (usually just one), weighs 2.1 tonnes and is 5.2m long.
A more reasonable taxi for single riders is going to weigh around 1.5 tons (e.g. the prototype Tesla Robocab).
You also incorrectly assume that the efficiency depends linearly on weight. It doesn't, it depends more on aerodynamics, speed, and the number of stops. London buses (11 tons) average around 2.2 kWh/mi, Tesla Model 3 is at 0.21 kWh/mi while weighing 1.8 tons.
Different 30 people in the US. I’d drop my car if Waymo was around and financially made sense. You’ll never catch me wasting my time on a bus though. We’ll never even be close to Europe or elsewhere wrt where public transit was even decades ago in those places.
People here only ride the bus when there’s no other option for them.
Waymo could make a bus too anyway, especially after they’ve been in an area a while and have data about where they’d have the most impact. Internet at large would probably decry it as enshittification.
Yes, if you live within about 1 mile of downtown and also work there. If you live downtown and your job is in Beaverton, you're likely in for a 90 minute commute. Or if you live in Beaverton and your office is downtown. Portland's mass transit is adequate for a subset of people and completely impractical for most.
Source: I've lived in and around Portland for 50 years. Worked downtown much of it, and have lots of friends and former colleagues who used public transit. You have to really want to not have a car because the time commitment is real.
Okay, but clearly there’s a difference between, “the bus makes sense if you live or work near downtown,” and, “people only ride the bus when there is no other option,” as the GP said.
An average bus load in the US is around 18 people. So if you just compute the raw bus emissions, then an electric bus is maybe about 2x more efficient than personal EVs.
But wait, there's so much more!
A single bus/subway ride costs about $20, which is a surprisingly consistent figure for various US cities. In NYC it's $17, for example. I'm talking about the _true_ cost, as in: "yearly operational budget divided by the number of trips". And most of this cost comes from 3-4 drivers that every bus needs, not from the hardware itself.
Price is a pretty reliable proxy for CO2 emissions for general market goods and services. And with these kinds of numbers, personal cars are competitive. Even if we factor in subsidized road maintenance cost ( https://taxfoundation.org/data/all/state/state-infrastructur... ).
If we now add mild carpooling, so that a single robotaxi drives 2-3 people, then classic transit just dies. It's like comparing horse-drawn carriages with steam trains.
It's been 16 years since Google announced they were working on driverless cars. Waymo's been offering service to the public for six years. There's been plenty of news coverage and people already know about them. Being in the spotlight all the time just wears people out.
>it's really a shame that they aren't the one capturing all the media spotlight
This does seem at least a bit intentional on Waymo’s part. They appear to have adopted an under promise, over deliver mindset (unlike some other robotaxi alternatives that come to mind). They’ve seen a lot of their competitors (Uber, Cruise, etc) capture a lot of ultimately negative attention and folded as a result; I would probably try to avoid the limelight if I were them too. The tech speaks for itself.
Every Waymo on the street is already an advertisement, especially since they're so distinctive with their LIDAR units and now they're driving Chinese cars. Every time a Waymo is remotely associated with anything bad, the media jumps on them anyway, like the times they've "hit" a pedestrian but literally after a human driver hit that same pedestrian.
Tesla Robotaxis need to hog the media spotlight because they're trying to prop up the stock price, not because they work.
It reminds me of putting on noise-cancelling headphones in an already quiet room. You suddenly gain an appreciation for just how much quieter things could've been thanks to the contrast.
Similarly, having a vehicle accelerate/decelerate consistently, always taking turns with the same sharpness, etc.. all of it builds to this subtle feeling of the car ride fading into the background of the experience. You didn't realize those were features of a drive you could appreciate until experienced.
The more I think about it the more autonomous cars look like extremely demanding edge computers. You have limited power and cooling, unreliable connectivity and very strict latency requirements.
It's probably one of the few places where designing the hardware around the workload makes more sense than trying to make a general purpose platform do everything
Very cool. To be honest, I'd assumed they were using commodity hardware, but no real reason for that. Cool chip. It is quite interesting that so many large companies have their own chips for these things. I almost managed to get my hands on a couple of accidentally liquidated MTIA chips but I was too slow. I wonder what the chip is like and what the dev experience is writing for it.
It would just be cool to see what these guys have cooked up dev tools wise since Google has always been famed for good developer experience (often to such a degree that Xooglers are useless outside the exosuit).
REing Meta's custom AI HW to get it running with something would be one hell of a project, and one hell of a flex.
It would rely on them being nice enough to leave at least some binaries on the thing though. Because I'm not sure how would one even get an in on how to boot the things without.
Pretty much. If you compare the size of the hardware, power requiremements, and what is fed in ... Tesla doesn't just win overall, it wins in every category. They don't need to brute force while Waymo is grasping for straws in profitability
Interesting that the article uses "ML" everywhere instead of "AI". I'm sure some marketing person chose that term deliberately, and I wonder what went into that decision.
All of the industry leaders of AI can’t stop bragging about how their AI is committing what would be felonies if a human did it, warning that it’s about as likely as not that AI will literally lead to human extinction, and threatening that it will erase all white collar work (or not depending on the day) and investing resources in a manner that will only be worthwhile if it actually does.
I don’t think there’s any mystery as to why AI is not a good marketing term.
I thought that was stalled because the mayor didn't like it?[0] Any new developments?
> New York abandoned its proposal to allow self-driving taxis earlier this year. New York City’s new mayor, Zohran Mamdani, has said he would heavily weigh the interests of taxi drivers in deciding on rules for the technology.
I'm very open to being wrong about this, but will they actually work in NYC? Waymo requires some grace from other drivers and pedestrians in order to function and I just can't imagine New Yorkers will offer it.
>high-fidelity data from 13 high-resolution cameras
is it me or their example photo doesn't look even like 1 high-resolution camera (and very noisy at that, looks like a pretty high ISO setting). Their example looks similar and i'd say a bit worse than what my Fuji camera did 25 years ago in the dark after i'd adjust dynamic range of the image in GIMP. (and the cameras on the Waymo cars i see around have about the same lens size as my Fuji)
If it is really a showcase of their visual pipeline than it makes sense that they are still so reliant on lidar - which is pretty expensive because of the number of rays you have to support to maintain good situation-awareness and which is still hard and expensive to scale to get good resolution. That also means that their tech can't really be put onto other platforms like drones for example.
Absolutely not an expert on this, but I would imagine that they would train the pipeline on a raw-ish image data to have better consistency? I would think that some kind of algorithm "bettering" the image suddenly might cause a discontinuation in the pipeline that is harder to account for versus the ML machinery being trained to do this job as well.
Also, what looks good to us may not be "good" to a machine.
>they would train the pipeline on a raw-ish image data to have better consistency? I would think that some kind of algorithm "bettering" the image
of course, whatever "bettering" algorithm makes sense, the neural network would probably emerge it inside itself as a result of training. My comment was more about hardware in the sense that "garbage in - garbage out", and so you'd want to have a decent hardware which is available and pretty cheap today while Waymo seems to be using, i'm stating it politely :), very-very-commodity level.
>Also, what looks good to us may not be "good" to a machine.
the kernels of the first layers of visual CNNs (at least the ones i looked at) converge to the Gabor-like kernels what are in the first layers of the human visual cortex. Bunch of years ago, before deep-learning CNN, etc. i saw an article (in neurophysiology field) proving an optimality of such kernels for those visual tasks. Of course things may and probably diverge a lot in the deeper layers, yet deep dreaming, etc. suggests to me that the nature of the divergence may be less severe than expected.
Custom silicon while it still looks like they've got multiple gaming PC's worth of compute gear mass isn't a flex when you remember what Tesla has with their FSD computer comparably just sipping power, while processing signals that are less straightforward as unscalable LiDAR
Apologies for repeating myself, but it's hard to convey how far ahead Waymo is in every area: sensors, vehicles, training data, markets served, simulation, infrastructure, operations, regulatory.
They're growing fast, but I do wonder what they see as the bottlenecks: municipal approval, training, manufacturing? It's really unfortunate that lobbying is slowing self-driving rollout in some of the biggest and most important cities: NYC, Chicago, DC.
I just took a Waymo in San Jose, having taken a few from San Francisco - it dropped me off in a place that would have made sense in SF (across the street from my destination at a light, avoiding a U turn) but was terrible in the suburban streetscape where the road was literally 10 lanes 4 / 4 / 2 turning and no one would ever drop someone off like that instead of U turning so they didn't have to cross scary giant traffic.
So weird little things for scaling, but overall it was a good ride till that point. Also, the dropoff was at a hospital/clinic and Lyft for the return trip had no issue with going into the property to the actual destination.
That seems like something not hard to fix, I am surprised they haven't already addressed it. Uber, as a comparison, absolutely narrows down the drop-off & pickup to the side of the road you want. I would demand the same of Waymo.
Waymo also lets you pick, but only from a predetermined list of pickup/dropoff locations. Sometimes there's not a spot on the side of the street you'd want, other times it will default to a spot on the other side of the street but you can override.
So are they? I know that tesla is behind, but they have so many simulation miles, several generations of their own hardware, and lots of people behind the wheel taking over.
I've also seen the autonomous waymo vans, which seem nice too.
Is there a "theoetical" ahead and a "practical" ahead?
How do the finances look and does financial solvency count? will someone run out of money first?
There have been two waymo articles (about sensors, and now about hardware), is this a sign they are worried?
I am actually wondering.
Is it unfortunate, or just a different priority in a different place?
Those are also cities with a great public option for transit (unlike, say, Dallas or the Bay Area where different municipalities squabble). One might laud New York driving people away from car culture when the subways/buses can serve more people, greener.
Perhaps, but I can't think of why Waymo should be prohibited but not Uber or taxis.
Taxis and barely-legal app taxis both require a human to drive. Someone who's earning a living doing this, and paying taxes to NYC. Waymo is driven by a robot, owned by a company out in California, who has a lot of lawyers doing their best to avoid as much financial and legal liability as possible. Not that Uber and its ilk isn't doing the same thing but they've still got a tax-paying citizen in the loop.
The current mayor of NYC is a DemSoc so he might have some ideas about keeping NYC's money in the pockets of middle-class NYC citizens.
I don't think all of the economic benefit stays in NYC even when you take a yellow cab. A significant share of NYC taxi medallions are owned by entities based outside the city. Marblegate Capital in Connecticut owns about 2,146 medallions—nearly 16% of all NYC taxi medallions. They make more profit off the medallion financing business than actually operating the taxis which is basically break-even AFAIK.
I wouldn't be surprised if that's somewhere on Zorhan's to-do list :)
Allow Waymo and leave the choice up to residents. If they want to pay a driver they can.
Nobody is keeping money in anybody's pockets. What you are advocating for is a tax on NYC people to force them to use human drivers instead of new technology. Just like the toll booth operators of the 90s.
It is so disappointing to see the tech class advocate for laws that keep jobs that are unhealthy and lead to nowhere (cabbies) instead of moving the country forward (and those people getting other, hopefully better, jobs). But again, that type of short-sighted, zero-sum/socialist thinking is becoming more common, sadly.
Buses and subways are not necessarily "greener".
30 people in 30 waymos vs 30 people in an electric bus?
No contest.
No contest when the bus is full, but the overall efficiency (/"greeness") of the system is much less clear considering: buses are rarely full, most buses are diesel, heavy buses tear up the pavement on their routes (especially near stops), and buses don't take riders directly to/from their destinations.
Most of the city buses in my UK home town are now battery electric double deckers. These carry up to 85-90 people (with 25 standing), weigh 11 tonnes, and is 10.5m long.
A typical jaguar Waymo usually carries up to 4 people (usually just one), weighs 2.1 tonnes and is 5.2m long.
So the double decker bus should be more environmentally efficient with maybe 5 passengers making independent journeys, though operational costs for paying a driver probably make the Waymo more financially efficient in that situation.
I suspect robotaxis will become very important in suburban areas with distributed jobs where buses don’t work well. In busy cities there simply isn’t the roadspace to replace buses with robotaxis (or any other form of car) for commuting.
> A typical jaguar Waymo usually carries up to 4 people (usually just one), weighs 2.1 tonnes and is 5.2m long.
A more reasonable taxi for single riders is going to weigh around 1.5 tons (e.g. the prototype Tesla Robocab).
You also incorrectly assume that the efficiency depends linearly on weight. It doesn't, it depends more on aerodynamics, speed, and the number of stops. London buses (11 tons) average around 2.2 kWh/mi, Tesla Model 3 is at 0.21 kWh/mi while weighing 1.8 tons.
Different 30 people in the US. I’d drop my car if Waymo was around and financially made sense. You’ll never catch me wasting my time on a bus though. We’ll never even be close to Europe or elsewhere wrt where public transit was even decades ago in those places.
People here only ride the bus when there’s no other option for them.
Waymo could make a bus too anyway, especially after they’ve been in an area a while and have data about where they’d have the most impact. Internet at large would probably decry it as enshittification.
Bus was just fine in Portland, Seattle, and San Francisco.
> Bus was just fine in Portland
Yes, if you live within about 1 mile of downtown and also work there. If you live downtown and your job is in Beaverton, you're likely in for a 90 minute commute. Or if you live in Beaverton and your office is downtown. Portland's mass transit is adequate for a subset of people and completely impractical for most.
Source: I've lived in and around Portland for 50 years. Worked downtown much of it, and have lots of friends and former colleagues who used public transit. You have to really want to not have a car because the time commitment is real.
Okay, but clearly there’s a difference between, “the bus makes sense if you live or work near downtown,” and, “people only ride the bus when there is no other option,” as the GP said.
Maybe it's different where you live, but 95% of the buses I see driving around here are completely empty.
Are you near the city center, or out near the suburbs? It's common for a bus to be crowded on one part of the route and nearly empty at the end.
And this is a fundamental issue with transit. It _has_ to travel as mostly empty on avereage to be useful.
An average bus load in the US is around 18 people. So if you just compute the raw bus emissions, then an electric bus is maybe about 2x more efficient than personal EVs.
But wait, there's so much more!
A single bus/subway ride costs about $20, which is a surprisingly consistent figure for various US cities. In NYC it's $17, for example. I'm talking about the _true_ cost, as in: "yearly operational budget divided by the number of trips". And most of this cost comes from 3-4 drivers that every bus needs, not from the hardware itself.
Price is a pretty reliable proxy for CO2 emissions for general market goods and services. And with these kinds of numbers, personal cars are competitive. Even if we factor in subsidized road maintenance cost ( https://taxfoundation.org/data/all/state/state-infrastructur... ).
If we now add mild carpooling, so that a single robotaxi drives 2-3 people, then classic transit just dies. It's like comparing horse-drawn carriages with steam trains.
Most of the cost comes from paying the drivers, but also cost is a proxy for emmissions?
Yes? Why is this strange?
Drivers emit a lot of CO2, just like every person in the US. A driverless car allows these emissions to be reallocated to other more productive areas.
FWIW they started actively testing in Chicago a few months ago. I frequently see them around town now
Yes! I've seen cars as far north as Evanston and as far west as Oak Park.
I’m glad you said it because as a 40 something nerd roboticist taking a Waymo is nothing short of a miracle to me.
I mean drafting code on my phone with chatgpt in a Waymo is legitimately some Jestons shit and I’m not hearing otherwise.
Amazing
They really are the impressive tech, it's really a shame that they aren't the one capturing all the media spotlight.
I don't know if we'll get there eventually, but if we do, I won't be surprised if it's because of Waymo.
It's been 16 years since Google announced they were working on driverless cars. Waymo's been offering service to the public for six years. There's been plenty of news coverage and people already know about them. Being in the spotlight all the time just wears people out.
>it's really a shame that they aren't the one capturing all the media spotlight
This does seem at least a bit intentional on Waymo’s part. They appear to have adopted an under promise, over deliver mindset (unlike some other robotaxi alternatives that come to mind). They’ve seen a lot of their competitors (Uber, Cruise, etc) capture a lot of ultimately negative attention and folded as a result; I would probably try to avoid the limelight if I were them too. The tech speaks for itself.
Every Waymo on the street is already an advertisement, especially since they're so distinctive with their LIDAR units and now they're driving Chinese cars. Every time a Waymo is remotely associated with anything bad, the media jumps on them anyway, like the times they've "hit" a pedestrian but literally after a human driver hit that same pedestrian.
Tesla Robotaxis need to hog the media spotlight because they're trying to prop up the stock price, not because they work.
Riding in one is pretty mind-blowing at first. But what's more shocking is how fast you get used to it. After a while it's just a boring car ride.
I wish more people would experience it and see that the future is now.
It's beyond boring after a bit.
It reminds me of putting on noise-cancelling headphones in an already quiet room. You suddenly gain an appreciation for just how much quieter things could've been thanks to the contrast.
Similarly, having a vehicle accelerate/decelerate consistently, always taking turns with the same sharpness, etc.. all of it builds to this subtle feeling of the car ride fading into the background of the experience. You didn't realize those were features of a drive you could appreciate until experienced.
The more I think about it the more autonomous cars look like extremely demanding edge computers. You have limited power and cooling, unreliable connectivity and very strict latency requirements. It's probably one of the few places where designing the hardware around the workload makes more sense than trying to make a general purpose platform do everything
A self-driving car is an "AI robotics" platform.
Both self-driving car and general purpose AI robots are emerging technologies right now. My bet is that eventually, the HW will converge accordingly.
Very cool. To be honest, I'd assumed they were using commodity hardware, but no real reason for that. Cool chip. It is quite interesting that so many large companies have their own chips for these things. I almost managed to get my hands on a couple of accidentally liquidated MTIA chips but I was too slow. I wonder what the chip is like and what the dev experience is writing for it.
It would just be cool to see what these guys have cooked up dev tools wise since Google has always been famed for good developer experience (often to such a degree that Xooglers are useless outside the exosuit).
Downside of the MTIA chips is no open backend - so quite a lift to get them working if possible at all :(
REing Meta's custom AI HW to get it running with something would be one hell of a project, and one hell of a flex.
It would rely on them being nice enough to leave at least some binaries on the thing though. Because I'm not sure how would one even get an in on how to boot the things without.
It really doesn't say much. Tesla had more detailed talks on their HW and AI stack. This reads like a glorified marketing blurb.
Recruiting
That's exactly what this is, a marketing blurb. Was this unclear?
Pretty much. If you compare the size of the hardware, power requiremements, and what is fed in ... Tesla doesn't just win overall, it wins in every category. They don't need to brute force while Waymo is grasping for straws in profitability
Interesting that the article uses "ML" everywhere instead of "AI". I'm sure some marketing person chose that term deliberately, and I wonder what went into that decision.
All of the industry leaders of AI can’t stop bragging about how their AI is committing what would be felonies if a human did it, warning that it’s about as likely as not that AI will literally lead to human extinction, and threatening that it will erase all white collar work (or not depending on the day) and investing resources in a manner that will only be worthwhile if it actually does.
I don’t think there’s any mystery as to why AI is not a good marketing term.
Apple did exactly the same thing a few years back:
Apple avoids “AI” hype at WWDC keynote by baking ML into products (2023)
https://arstechnica.com/information-technology/2023/06/at-ap...
Would love to get them in nyc
I thought that was stalled because the mayor didn't like it?[0] Any new developments?
> New York abandoned its proposal to allow self-driving taxis earlier this year. New York City’s new mayor, Zohran Mamdani, has said he would heavily weigh the interests of taxi drivers in deciding on rules for the technology.
0: https://www.nytimes.com/2026/06/17/technology/waymo-driverle...
I'm very open to being wrong about this, but will they actually work in NYC? Waymo requires some grace from other drivers and pedestrians in order to function and I just can't imagine New Yorkers will offer it.
Do you think they will?
Contact you mayor
>high-fidelity data from 13 high-resolution cameras
is it me or their example photo doesn't look even like 1 high-resolution camera (and very noisy at that, looks like a pretty high ISO setting). Their example looks similar and i'd say a bit worse than what my Fuji camera did 25 years ago in the dark after i'd adjust dynamic range of the image in GIMP. (and the cameras on the Waymo cars i see around have about the same lens size as my Fuji)
If it is really a showcase of their visual pipeline than it makes sense that they are still so reliant on lidar - which is pretty expensive because of the number of rays you have to support to maintain good situation-awareness and which is still hard and expensive to scale to get good resolution. That also means that their tech can't really be put onto other platforms like drones for example.
Absolutely not an expert on this, but I would imagine that they would train the pipeline on a raw-ish image data to have better consistency? I would think that some kind of algorithm "bettering" the image suddenly might cause a discontinuation in the pipeline that is harder to account for versus the ML machinery being trained to do this job as well.
Also, what looks good to us may not be "good" to a machine.
But I'm really just guessing in the dark here.
>they would train the pipeline on a raw-ish image data to have better consistency? I would think that some kind of algorithm "bettering" the image
of course, whatever "bettering" algorithm makes sense, the neural network would probably emerge it inside itself as a result of training. My comment was more about hardware in the sense that "garbage in - garbage out", and so you'd want to have a decent hardware which is available and pretty cheap today while Waymo seems to be using, i'm stating it politely :), very-very-commodity level.
>Also, what looks good to us may not be "good" to a machine.
the kernels of the first layers of visual CNNs (at least the ones i looked at) converge to the Gabor-like kernels what are in the first layers of the human visual cortex. Bunch of years ago, before deep-learning CNN, etc. i saw an article (in neurophysiology field) proving an optimality of such kernels for those visual tasks. Of course things may and probably diverge a lot in the deeper layers, yet deep dreaming, etc. suggests to me that the nature of the divergence may be less severe than expected.
they now are doing a 5nm asic, what did they have before this?
Custom silicon while it still looks like they've got multiple gaming PC's worth of compute gear mass isn't a flex when you remember what Tesla has with their FSD computer comparably just sipping power, while processing signals that are less straightforward as unscalable LiDAR
Compute is under the trunk ??? Noted. At the price of RAM nowadays it's like the new copper !