Waymo the Leapfrog
loeber.substack.com
loeber.substack.com
Riding it is now a mundane experience, and that's a marvel in itself. Every time I'm forced to get an Uber in another city (or do airport rides as Waymo doesn't do pickups/dropoffs in LAX yet), I feel like I went 5 years back in time.
My Tesla has FSD and that has gotten progressively better the last few updates, however Waymo still feels ahead. I can truly "relax" in a Waymo, where FSD still makes me uneasy at times, like I'm supervising a teenage driver.
The cars are well maintained and clean inside and out.
They are very careful and considerate drivers.
They navigate complex and ambiguous situations with unprotected turns, pedestrians, bicyclists, double-parked cars, construction zones and narrow streets.
They're better than Uber in every respect.
And I dare say I think they're now a safer and more capable driver than I am, at least within their designated operating territories.
I think if there's something weird/bad going on, someone manually takes over (but not sure if that's confirmed or not). Still a cool experience, but the real world is a lot more complicated than it first seems.
Even with all that, it definitely feels like the future.
Remote agents give the cars "advice" or label objects, but the car is free to not believe them. There's no actual remote control.
If the advice is not enough, roadside assistance goes and drives the car manually.
Interesting. Are you sure they aren't honking at you? I think your framing might be right but I'm not totally sure
Some thoughts:
* For the purposes of transit efficiency, self driving cars are very similar to Ubers. They have a low passenger density (being a regular car), and once the passengers disembark, they still take up space on the road with 0 passengers. Better experience and lower costs will basically just induce more demand over more efficient mass transit options. If you imagine everyone at a bus stop ordering an Uber, or have ever seen the flurry of Ubers after a big event, it's clear why self driving isn't really addressing the core issue.
* You can't really make direct cost comparisons to the infrastructure costs of bus lanes or subways like that. Infrastructure is ungodly expensive in the US, yes, but there are very well understood reasons to make dedicated bus lanes and subways: they don't compete with cars on the road. They're high density transit options, so having more reliable service will impact a lot more people (and reduce car congestion on the road!) A rideshare service is wholly unprepared to deal with the transit demands of a larger city, and imagining that we'd replace existing mass transit options with it is silly.
* I don't really understand the point about suburbs. You can already get that experience today by ordering an Uber to and from work. If there's more demand, it's just going to make traffic even worse while promoting more suburban sprawl.
1) With superior scale, the utilization of each Waymo approaches 100%. "Empty rides" going to a new pickup spot become more and more rare because they immediately get a ride close-by.
2) I think you underestimate the case for shared rides. Any Waymo with >= 2 passengers is reducing congestion, not adding. When you consider small bus-shuttles, e.g. 6 or 12 or 18 seats, it gets even more impactful.
3) The point with suburbs is that they become more accessible when the price point comes down. If you have a 45-minute commute to drive yourself, an Uber might cost you $60, which would be prohibitive. If a seat in a carpooled Waymo costs only $5 or $10, the incentives and customer behavior change significantly.
> *”because they immediately get a ride close-by”
That’s is unsubstantiated utopia (from you POV). People do not have uniformly distributed locations for starting and ending routes. There are concentrations of where people work, live, go to restaurants, schools, etc. More so if you are assuming suburban homes. The direction of routes also have very strong biases.
You very strongly and, again, unsubstantially overestimate shared rides. Shared rides could already be happening at scale with a human driver and are not. There is nothing about self-driving tech that solves any urbanistic, social or economic barrier to shared rides. Also, a shared ride negates most of the benefits of self driving cabs. You are not alone in the confined space, you are sharing with others, others that will likely be more intrusive than a professional driver. The vehicle will not be as clean, as comfortable, as silent, etc.
About commute getting cheaper seems a pipe dream. There is nothing in the self driving tech remotely indicating that a currently $60 ride will cost $5 with self driving cars.
Most of the dedicated bus lanes I've seen were once open to all vehicles, but later restricted. Some were built specifically for busses, but not many.
Buses on regular roads create a "shadow" that affects other vehicles behind it and reduces 2 lanes in one direction to an efficiency of much less than 2 lanes, due to stops and starts. And except for peak usage times outside of a few major US cities, I would argue it is a net congestion creator rather than congestion reducer. A bus has the same effect as 10 or 15 cars, from my observation; so if there are less than 15 people on a bus, it's worse.
Part of the reason that buses have wraps or tinted windows is so you can't see how few people are riding on it.
At least the Waymo is not moving and not on the road when not in use.
Peak usage times are obviously what matters for congestion
> I would argue it is a net congestion creator rather than congestion reducer
I think buses even outside of peak hours are usually net congestion removers... but if you want them to be congestion removers during peak times, you also need to offer service during other times.
Even triple the space usage, is leas than what you claim to observe. What is your methodology for determining this?
When you get a chance, follow behind one during a non-congestion period and count the cars that pass you in the other lane (meaning, your lane is slower) and estimate how many additional cars it would take to have the same speed of traffic (how many cars would need to be in your lane, instead of 1 bus, for the lane to be as slow?).
Imo everyone driving unpredictably to "get ahead" and save (literally, do the math) 1 minute on a 30 minute commute is what causes a lot of congestion.
Imo even if a bus took up 10 car spaces, the more predicable driving and slower accel/decel might still be a net positive for traffic flow.
Also, do you care about the state of traffic at off-peak hours? who cares if buses don't run at peak capacity when there's no traffic?
But like I said, triple the “physical space” the bus actually takes up and that is still less than their claimed “space consumption” of a bus.
And Waymos would need to also stop to pick people up as well…
This is quite the stretch. Even in the best case scenario, Waymos won’t beat well run public transit lines in dense cities, especially in east and south asia.
This piece is way too optimistic about Waymo. They’ve mastered a couple of cities over many years. To do that for more cities would require just as much time. It’s conceivable that ride share will continue to exist until that happens, which is likely several decades.
The main advantages of trains are:
* Dedicated right of way - which you cant't get with cars.
* High route capacity - even a unrealistic 10x improvement still doesn't compete with trains. At 2x it's worse than busses.
* Cheap running & maintenance costs - doubtful running costs are similar, certainly not maintenance.
Trains do however congest, you get that a lot in the USA when you don't have dedicated passenger lines.
> Even having a modest percent of all cars on the roads cooperative autonomous vehicles using basic control theory to relieve congestion would aid the entire cities traffic (there have been several studies proving this).
There have also been several studies into something called induced demand. The result of reducing car traffic is that more people choose to drive, resulting in more car traffic. The only way to actually reduce traffic is viable alternatives to driving.
My main point is that taking away easy/street parking will be more effective than just not building roads or even going with congestion pricing.
You build more lanes, that reduces the time of a commute, that reduces the "price" and therefore more is "demanded" until you back up at the equilibrium. It works exactly the same as every other good except that the price is in a different currency.
No one is surprised when, if the baker lowers their cost, more people buy bread. Yet somehow, people are flabbergasted when, if you reduce travel times, more people travel.
https://humantransit.org/2012/09/the-photo-that-explains-alm...
They are also not convenient. They don’t provide direct point to point transit. They follow a schedule and require transfers to get to any location not on the local route. As a practical matter they’re generally not compelling for most people outside of specific uses like commutes. Even in NYC for the decade I was there I almost never used a bus because the experience was so unpleasant, I stuck to the subway and Uber.
So, without some sort of draconian mandate requiring divestiture of cars, autonomous cooperative cars are a substantial advantage.
I’d note you also don’t need autonomous cooperative cars to have as much seating. A single person autonomous vehicle could be a single person sized car and it wouldn’t feel like a death trap because accidents would be effectively nonexistent. Traffic density could be bumper to bumper with high rates of flow. In that sense you could actually jam many more people into the space of a current (largely or completely) empty bus.
When public transport is overcrowded it's unattractive- nobody takes the bus anymore, it's too crowded. When it's not overcrowded it's inefficient. Catch 22! The US can't have public transit. Everyone else in the developed world seems to manage just fine but the US is special.
Here's an idea: use all those big server farms and AI talent to optimise public transport. If there's too many passengers for a route, add more buses. If there aren't enough, reroute the buses where there's more demand. Maybe get those leapfrogging self-driving AIs to drive buses, instead of personal cars, and get the buses to flock instead of cars? There are social solutions, institutional solutions and technological solutions to the problems in mass transit. A nation that considers itself a leader in science and technology should be able to make a dent there. Companies who advertise their self-driving AI as beneficial for its safety and efficiency could make more than a dent if they used their considerable means to attack a real problem. But - noooo! It's gotta be self-driving cars. More cars! More cars to solve the problems caused by too many cars in the first place.
The idea that buses must be enormous vehicles belching along the roadway packed with people is an anachronism. A much smaller vehicle with a smaller engine and less mass overall designed to transport a small number of people directly to their destination in a large collaborative fleet of autonomous vehicles would be more efficient in literally every dimension. The obsession with the form factor of the bus as being somehow ideologically ideal weirds me out.
Here’s a middle ground - in US cities where buses are largely empty but are still enormous behemoths, why not reduce them down to vans and do more dispatched transport? This is specifically your idea but is more achievable because the issue with rerouting buses and sending more out etc is the cost per bus of the typical form factor bus costs between $500k-$1m. You could buy a fleet of 10 EV vans for a single bus. The second greatest cost for operations is the driver. Autonomous cooperative vans lets you scale that to 15-20 vans for the cost of one bus.
If you move from vans to smaller 2-3 person passenger vehicles rather than a larger van you can increase that to 30-40. Now you’re transporting 60-120 people point to point with the same cost structure as an empty bus taking a few people on long circuitous routes belching noise and air pollution.
The problem people seem to have with this idea is that it’s not a bus, which, again, appears to be more ideologically driven than rational. If you took that bus, carved every row of seats into its own autonomous EV, you would end up with a more efficient better utilized infrastructure that would entirely obviate the need for private car ownership.
Oh, absolutely. I've been taking buses (and trains, and trams, and trolleys) all my life and I've never been on an enormous vehicle belching along the roadway packed with people. Where does that imagery come from? Mad Max?
Where I live now, in the UK, buses look like this:
https://tfl.gov.uk/modes/buses/
You can see the people beside the bus for size comparison. I don't have the exact numbers but the typical double-decker takes about 5/3s the space on the road that an SUV takes and a substantial fraction are electric or hybrid, so no, I don't think your maths pan out.
Edit: btw a single-decker EV bus will set you back ~£340,000. Hybrids and hydrogen cell ones are more expensive, around £500k (around $590k). Dirty petrol and diesel double deckers cost £20-30k but those are no longer allowed according to the mayor's green policy:
https://www.mylondon.news/news/zone-1-news/how-much-london-b...
What bus is it that costs $1 million?
Earlier this month, The Denver Gazette reported two of the four e-buses Colorado Springs' Mountain Metropolitan Transit acquired in 2021 are not running. They cost $1.2 million a piece, mostly paid for by government grants.
https://www.foxbusiness.com/politics/electric-buses-sitting-...
A London double decker is 30 feet long
https://www.quora.com/What-is-the-average-length-of-a-London...
(I had better PDF specs but was loading slowly)
A suburban - a massive SUV is 18ft long. This is not a normal SUV though.
Where I am, in seattle, buses are from 60 ft to 30 ft long. This is the same as a London bus, but the double length buses are pretty common too.
https://metro.kingcounty.gov/advisory-groups/service-guideli...
I really wonder what this looks like logistically, when you realize how inefficient it is for everyone to be in their own individual Waymo during rush hour.
The obvious first step is:
> Many people who currently drive themselves would probably be happy to carpool in a self-driving vehicle if it’s reliable and easy
But if it's a 4-person vehicle, I'm not sure that's enough -- that's still a lot of vehicles clogging bridges and tunnels. And if you get to 12-person self-driving minibuses, it feels like too much time picking up and dropping people off if you're trying to stick to one vehicle.
So I kind of wonder if there will be intermediate ~20-person self-driving buses to bring people in and out of cities, that respond instantly to demand?
So in the suburbs, you take an 8-minute individual Waymo that brings you to a perfectly timed Waymo Bus that you wait 3 minutes for, get on with 19 other people for 42 minutes, everybody gets off at the same point at the edge of downtown, and everybody gets into carpooling individual Waymos already waiting, so you take 7 mimnutes to get to your office, dropping off 2 people along the same route you would have taken anyways.
But the buses don't have fixed routes or stops or timetables -- they just aggregate the demand along the common long corridors. Essentially replacing light commuter rail.
Reporting often misses just how quickly vehicle costs have dropped. When I started in the industry, vehicle costs often exceeded $500k. Almost everyone I'm aware of today is targeting sub-$100k.
But also -- I've been thinking about what it would be like to have nobody driving for over a decade, ever since the promise of self driving cars became a popular topic here on HN. I still really want to own one - once they're good enough to get rid of any interior controls and face the front seats backwards with a gaming table in the middle. If board game day could also have a destination dinner, I would pay a lot of money for that car.
In L.A. for public transportation, they'd have to be really cheap. Public transportation here is crowded because it's realtively cheap. For a long while I took nothing but bus around L.A. It cost me around $50/month to go to nearly all the places I needed to go everyday. I still bus/train often, but sometimes now I have to drive for work and my wallet notices.
Took twice as long as driving but I did not need to pay for parking or drive in the snow when I took the bus.
My guess is that Google will try to get out of the business of owning the vehicles and make Waymo an enabling technology. But there is a lot of experimentation to be done before anything other than how Waymo has evolved to become part of what Waymo does.
Winner-take-most markets, like Uber, tend to be two-sided marketplaces with major network effects: folks use Uber because they’re pretty ubiquitously available and relatively inexpensive; they’ve available because the many users request many trips which makes working for Uber as a driver appealing. (Ditto AirBnB, etc.)
But Waymo’s competitive advantage is pretty much only technology here. There is an argument to be made that this technology has a “network effect”-style moat: you need massive amounts of data from cars to build the tech; you get that data by being first to market. But this post doesn’t make this argument, it simply points to Uber/Lyft/the-taxi-app-graveyard and says “see?”
It’s also not clear that this set of data will forever be as critical as it is now, nor that in 10 years it won’t be comparatively easy to collect.
If the only edge that Waymo has here is technology, that’s not a natural monopoly position on its own. Autonomous driving is surely hard, but it’s an absolute-performance-baseline task, not a zero-sum market like ads. The existence of one set of 0.0001%-accident-rate-drivers doesn’t prevent anyone else from producing a 0.0001%-accident-rate self-driving car. There’s a small market stickiness effect from being first, sure, but this is not the same kind of natural monopoly as a two-sided marketplace. It’s clear why users use Uber and AirBnB over alternatives: it’s where the drivers and properties are, and that takes time to build up and it hard to take away. It’s not clear why someone would be a tied to Waymo if another equally-available self-driving taxi were available that were substantially less expensive, even if it were marginally less safe. (Of course, everyone has their trade-off point here—consider airlines: while people are more likely to pick an airline based on convenience and price, rather than safety record, these days—United perhaps excepted—that’s only true because the overall accident rate is so low.)
Assuming the baseline for “safe operation” doesn’t escalate, improvement in technology (e.g., more capable GPUs, better sensors, more public datasets) will only make this baseline easier to reach for newcomers. If the tech gets easy enough, all it will take for a new entrant is CapEx, and it’s not clear how Google would differentiate, and could end up as a single commodity provider among many, rather than a two-way marketplace owner benefitting from network effects on both sides.
That said, to construct a monopoly here, Waymo does have a couple of options:
1. Argue for regulation around self-driving cars that protects incumbents, by (e.g.) setting safety standards that are impossible for new entrants to meet. The NRC has a “good” model for this: by analogy to nuclear plants, any new vehicles on the market would need to prove that they are better than existing vehicles in the market.
2. Cut deals with major auto manufacturers so that incumbents just can’t get the data through partnerships with camera-equipped vehicle manufacturers.
Number 2 seems like it could be clearly anticompetitive, but number 1’s playbook is being written by the big AI companies as we speak.
Technology is not the reason it's a winner-take-most market. It's helpful, but in the long term, many players will have comparable technology.
Utilization is the reason why it's a winner-take-most market. If you have the most popular app with the best drive liquidity, that's a self-reinforcing feedback loop. In turn, it means that your economics are better because your cars are going to be utilized at a higher % of the time. This in turn means that you can charge lower prices, and so forth. (A similar dynamic has been the case for Uber, for example.)
In many industries large players will benefit from economics of scale. This is definitely true for Uber. But it's not a natural monopoly position -- if someone could offer the same service for half the price, people would switch, as we all expect they will when Waymo does this.
This is different from, e.g., AirBnB, where there's not that much a competitor could offer, necessarily, and they'd have to build a huge set of inventory from scratch, a true two-sided marketplace. A competitor that manages to cut the service fee in half would still face an uphill battle building up that marketplace.
Do not underestimate quality of the experience/product on the whole. Whenever I take an Uber/Lyft/Taxi, I find myself in the uncomfortable position of having to deal with a stranger in a small space for a period a little longer than an elevator ride (if you know what I mean)... and that driver in turn has to figure out how to interact with me (or not). We're social animals and we feel compelled to do something or to expect something even if that interaction doesn't come. In a Waymo this forced social situation doesn't exist: and I for one find it superior enough to pay the small premium... which in my experience is on par with what I'd have to tip (another fun thing to have to deal with in the Uber/Lyft/Taxi scenario).
Also, I've mentioned this recently elsewhere. Uber/Lyft/Taxis don't always show up, drivers cancel trips and that can happen several times in a row. Waymo hasn't done that to me yet. So the certainty and dependable expectation setting of Waymo vs. Uber/Lyft/Taxis make a compelling case for premium pricing. And again, in my experience, the pricing of Waymo hasn't been so much more than the alternatives to justify calling those alternatives "substantially less expensive". Once the tipping is figured into the equation they're roughly on par.
Would I consider other self-driving entrants over Waymo? Sure if they can match the quality of service and are price competitive. But the comparison with traditional driver based services needs to account to the higher overall product quality and not just the pricing model or even safety margins (so long as they are within some reasonable baseline).
I don't consider Uber/Lyft to be self-driving taxi services. There's no question these are different.
> The once-dreaded long commute is about to come back in a big and pleasant way. I would have no issue at all sitting in a Waymo for 45 minutes each way every day. It’s just a nice time to myself that I can use to nap, work, or read.
In places with working public transit, you can already have that experience. While it may make the experience less stressful (because you're not driving), it is still a huge time sink. You can do something entertaining while on a bus or in a Waymo, but it still blocks off time you could spend more productively.
Unfortunately, sitting in a Waymo is experientially not at all the same as riding the BART or NYC subway.
Alas, not if you're from Europe it seems. I'd love to try one when I'm in the Bay area
For good reason sure, but might as well accept it.
Do you see what I mean by 5-8 years?
Also VW invested long time in autonomous driving.
I wouldnt be sure if Google has such an advantage.
You can complain about Boston drivers, but they are pretty predictable compared to SF tourists.
Snow is an issue, but they'll get that done.
The stock is down 55% in the past three years.
> Why wouldn't Musk just partner with Waymo on this?
Because they sell cars worldwide, not just in San Francisco and Phoenix. Because Waymo's required sensor suite is far too expensive and maintenance heavy for a consumer-owned vehicle. Because what they have is already far better than any other car manufacturer's system. Because they have a huge head start in data collection over anyone else including Waymo.
> Isn't this a situation where the more people using the tech will make it better faster than trying to have a dozen different companies chasing it essentially duplicating effort and training
No, because many different companies are trying different approaches with incompatible hardware and it's not certain which approach will ultimately have the best tradeoffs in the real world. Centralization is not the way to produce innovation.
Tesla prioritized building a highly generalized stack that can offer incremental improvements from L2+ on all roads (today) to L3/L4 some time in the future. Waymo is prioritizing building a tightly geofenced L4 system and gradually expanding the operating domain.
Neither is more of a "correct" approach, since they have different design goals, and it's hard to even directly compare them at this stage of development.
But I think that as AI continues to improve and more sensors are manufactured, the cost goes down and feasibility goes up for more companies to implement self-driving.
Definitely. Waymo has partnered with Zeekr to produce purpose-built taxi vehicles: https://www.youtube.com/watch?v=Ew32ydR2O3c
1 HP is over 700 watts.
A few extra HP to generate power isn't any big deal.
They don't need to be electric. You can make electricity out of gasoline with an ICE engine. This is not a problem in reality.
The practical issue is t hat Waymo uses more advanced sensors than Tesla vehicles use. Partnering with Waymo would mean telling Tesla owners and Tesla shoppers "your vehicles won't be self-driving."
If Tesla partners with Waymo, they're telling investors that their self-driving technology won't cut it. It's not just about ego. If Tesla is worth 10x more than GM or 100x more than Mazda because of a self-driving/AI/robo-taxi future, then partnering with Waymo is essentially saying "we'll provide the easily replaceable piece of the self-driving system."
Let's say that this works by Waymo buying Tesla vehicles for its self-driving fleet. That leaves all the power in Waymo's hands. Maybe they ink an exclusive deal for a few years. When that deal is up, Waymo is going to be pitting Tesla/GM/Volkswagen/Toyota/etc. all against each other on price. Waymo has the thing they don't have. They all have mostly comparable substitutes.
Let's say it works by Tesla licensing Waymo's technology and ditching their own AI plans. Again, it leaves the power in Waymo's hands. A few years out when the deal is set to be renewed, Waymo has a dozen car companies clamoring for its technology. Waymo knows it can seek a very high price, especially if Tesla wants to keep it exclusive. If Tesla doesn't want to keep it exclusive, then Tesla's valuation needs to be a lot lower like all the other car companies.
And what's in this for Waymo? They could partner with any number of car companies. What is Tesla bringing to the table? A CEO likely to damage Waymo's reputation and blame them for anything that goes wrong?
In terms of sharing things vs. not: you're kinda getting down to a fundamental inefficiency in our whole society. If all the car companies worked together on engines, we could get better engines faster. If all the drug companies worked together, we could accelerate research. But there are problems to that idea too. Having a single way of doing data collection can lead to blind spots, having a single hierarchy can mean that different things don't get tried, etc. There's also the issue of whose work gets deemed worthy of compensation.
If you've ever worked at a large enough company, you'll have seen instances where people with entrenched ideas stymie progress. If everyone partnered on self-driving, who decides what sensors are going into the vehicles? Maybe I come up with a new sensor and SelfDrivingPartnership says "nah, we don't think it's worth the cost," and I can never test whether I'm right or not.
In some ways, open source tries to solve some of this: everyone can still compete to make something better with the freedom to do so, but you can just take other people's code so you don't have to duplicate efforts. Of course, the problem there can be that companies don't want to share the pieces that are worth a lot of money. If self-driving tech is worth a trillion dollars, I might rather get that trillion dollars for myself and you can wait another 2, 5, 10, 20 years for it to be available.
Yes, open sourcing and sharing training data would likely get us to safer self-driving cars faster.
> Of course, the problem there can be that companies don't want to share the pieces that are worth a lot of money. If self-driving tech is worth a trillion dollars, I might rather get that trillion dollars for myself and you can wait another 2, 5, 10, 20 years for it to be available.
I guess I'm not sure it's a trillion dollar idea ultimately if it becomes widespread - is ABS a trillion dollar industry? Also, I wonder if auto companies will become wary of the liability issues. A full self-driving car (no steering wheel) likely won't generally be owned by consumers. You'd hail one like you do an Uber. Consider the liability issues: Who carries the insurance? Not the rider as they have no input (even if someone chose to buy one it's tough to see that they would be liable in the event of a crash since, again, they had no input). It would have to be the maker of the vehicle that carries the insurance.
The other thing to consider is that ultimately in order for these systems to work most efficiently (and most safely) we need some communication between vehicles and with traffic signals. Another area where cooperation would be better than competition.
Maybe, but the Tesla Taxi seems to be a separate, non-consumer vehicle, more like what Waymo is doing.