California and Nevada are the only two states where the company can legally sell the technology to consumers. The two state DMVs gave Mercedes approval to begin selling the cars last year—Nevada in January, and California in June. Mercedes announced in September its plans to begin sales, but this is the first news of the cars actually reaching consumers.
Mercedes takes all legal responsibility for any accidents caused during autonomous operation.
So far, autonomous operation is limited to relatively few situations, but most importantly it includes traffic jams (which is where human drivers quickly lose focus).
I'm always terrified I will doze off without realizing when I'm in traffic jams. There's just so little stimulants that my brain demands going into standby mode.
They are opening themselves up to massive liability even with “fine print” — this is meaningful and should not be disregarded. Just because there’s some super indem clause in their customer agreement doesn’t mean the courts will rule in its favor and not strike it out.
The vast majority of traffic accidents does not result in jailtime for anyone involved.
Since algorithms cannot have malice, the only offense left is negligence which for companies usually means a heavy fine and/or legal restraints, maybe prosecution.
California being the most populous state is a pretty big deal. I'm guessing the focus on two states is more a result of Mercedes' focus on geofencing/mapping test areas than anything.
Well yeah that’s what the name “United States” means - different states in one country.
And each state has its own set of laws. Dunno why this example popped into my head but - back during COVID when we had the first case in the state of Washington, we said “first case in the U.S.”
> As of April 11, there were 65 Mercedes autonomous vehicles available for sale in California ... One of those has since been sold, which marks the first sale of an autonomous Mercedes in California ...
Those are pretty small numbers and strike me more as them testing the waters. It's neat to see some potential in the space after what felt like a total collapse with Waymo and Tesla especially not living up to expectations. It's not clear how good the tech is here, but it's a political victory for the space if nothing else.
What are you referring to with Waymo? They were still operating in LA and SF as of this week. Do you mean Cruise (owned by GM)?
In any case, the Mercedes system is what's known as a conditionally autonomous system. It can drive itself under specific conditions, but a human is still involved in operating the vehicle as soon as it exits that design domain. Waymo (and Cruise) are fully autonomous systems. Humans are not involved in the control loop and the vehicles will stop if they leave their operational domain. Tesla FSD is not an autonomous system at all and requires a human to always be in control.
Waymo and Cruise have failed to scale and it doesn't look like they ever will. It's difficult to find a single article with the full context, but around 2017 as Waymo started giving limited rides to the public, everyone though the problem had been solved. There was a flood of startups hoping to be a strong second mover in the space, and then basically nothing. Expectations have shifted over time, but 7 years later and there's been only small incremental progress.
> In any case, the Mercedes system is what's known as a conditionally autonomous system.
Yes, but it's significant progress if it reaches consumer hands. This sounds like the typical Waymo marketing of "you can't get to Level 4 through Level 3". I bought into this at one point, but Waymo and Cruise have demonstrated that going straight to Level 4 with $100k+ (probably much more than that) of sensors and corresponding maintenance doesn't scale.
Waymo is the only one actually scaling driverless taxis. They’ve gone from providing public rides in 1 city (Phoenix) to 3 cities (+SF, LA) with a 4th one (+Austin) coming soon. They’ve given more than a million rides, with the majority of them coming in the last year. No one else is doing this.
If other companies thought they’d hitch on to Waymo after they started giving rides in 2017, then that’s on them. Because Waymo has been pretty clear that their scaling is going to be methodical and not instantaneous.
"Taxi service in a congested urban area" is probably the market where it makes the most sense. Trips are pretty short, cars stay in the a local area where they have hyper-accurate mapping data, and are likely to quickly find another passenger or can return to a nearby parking/waiting area.
Waymo won't keep a monopoly on this forever. There are still high barriers to entry, but they will fall sooner or later. If there is money to be made, competition will come.
Indeed congested urban areas are their market. That's exactly why they are in two of the largest taxi markets (SF and LA) where it doesn't snow.
I'm not sure about barriers to entry lowering soon. First off, this is an incredibly hard problem no one else seems to have cracked (and even Waymo has a long way to go). Second, they could help influence regulation that requires extremely high safety records, mandatory sensors, etc. which could make it harder for others to compete. It's all very early though.
i also find it difficult to find data on how much they're scaling but anecdotally, i've been seeing a steady increase in waymo vehicles all year, here in tempe i encounter at least 3-4 every day it seems, driving in some pretty impressively heavy pedestrian traffic and on freeways too
I'm curious why Waymo is considered to be not living up to expectations? They operate in limited number of places, but seem to be just fine driving there (of course some people will not like them regardless). Tesla - yes, not comparable..
"Even if Waymo's schedule slips a few months and it introduces a self-driving car service in the middle of 2018 instead of late 2017, that will still give the company a multiple-year head start over most of its rivals. And it would confound skeptics who insist that full self-driving technology is still years away."
Ok, maybe the as-of-2024 product is living up to the 2018-horizon expectations from 2017 (as long as your expectations were not at the high end of the range at the time).
The people who didn't get a driving license in 2017 because what was the point if cars would drive themselves in a couple of years may care. Even more so if they finally have to wait to 2050.
Anyway, the question was about expectations - and those came with a timeline.
> Anyway, the question was about expectations - and those came with a timeline.
The mention was of a "collapse" in the space, too-- which can't be true if we're still getting there.
I expect the timeframe of 2030 is when we get most of the way there. Even exponential growth takes time to reach a big world. Most of what Waymo has left to do is rollout.
Waymo does not sell cars. They are in a different business.
I don’t really know the numbers but Waymo adds something like 50k worth of hardware into the cars. That is not a viable approach for consumer cars (Mercedes, Tesla, etc).
Maybe. Not everything ever manages to make it to low cost, high volume- either because there is no reasonable path to making it that cheap, or because there are not sufficiently large markets in the in-between prices.
Waymo is amazing and has fully replaced Uber for me in SF. They're also expanding to the entire peninsula from SF down to Cupertino - you're out of date on this.
Waymo driving in SF is way better than human. It's not timid either.
This Mercedes post is marketing - it's not really what people consider autonomous given the constraints.
> "Drivers can activate Mercedes's technology, called Drive Pilot, when certain conditions are met, including in heavy traffic jams, during the daytime, on specific California and Nevada freeways, and when the car is traveling less than 40 mph. Drivers can focus on other activities until the vehicle alerts them to resume control. The technology does not work on roads that haven't been pre-approved by Mercedes, including on freeways in other states."
The above makes it nearly useless and far less useful than what Tesla has been shipping for years imo.
How is Waymo pricing compared to Uber or other ride share?
I originally thought autonomous cars would greatly reduce costs but considering the tens of billions dumped into these companies, they can't charge lower prices as they have to recoup investments.
It's a dollar or two more expensive than the corresponding Uber (non-shared) fare in my experience, so a bit less after tips. Keep in mind that the "true" price of these services probably isn't the biggest pricing factor right now. The cost is a way to manage demand, get some additional income, and investigate what the market will bear.
> As of the latest available information, Waymo is not yet profitable. Alphabet’s Q2 2023 earnings report indicated that its “Other Bets” category, which includes Waymo, had revenues of $285 million and an operating loss of $813 million. This was a decrease in operating loss from $1.34 billion in the second quarter of 2022, but still significant. Over the last five years, “Other Bets” has generated $3 billion in cumulative revenue but incurred $20 billion in operating losses, with Waymo constituting the bulk of these losses.
I just don't see the end game. Are marginal rides not profitable even at Uber fates? Uber for instance was losing money because they were taking rev from profitable cities to give incentives and build up drivers in the new markets. But Waymo doesn't have that problem. Is it just GPU spend that they can eventually stop?
Nothing about that quote indicates that waymo loses money on marginal rides, rather that it loses money in aggregate which is not particularly surprising considering low overall volume , r&d overhead of both head count and things like cars and facilities to store said cars
I suspect it's vaguely near gross profit, that is excluding almost all the expensive engineers, marketing, R&D costs, new silicon development costs, etc. It's not remotely operationally profitable though, you're right.
The business model is pretty simple. Robotaxis can operate much cheaper than humans and simultaneously provide a better service. Competing at the same cost as existing services makes them massively profitable. You can unlock additional efficiencies through things like fleet management, not supporting unprofitable cities the way Uber has to, vehicles designed for longer service lives and so on. Most companies plan to use that market (and closely associated ones like food delivery) as a profitable beachhead to scale down the technology to the point where it can be applied more ubiquitously in things like consumer vehicles. Some are more vocal about this than others.
I wonder if anyone has done an analysis looking at a set of tasks and ordering them according to the ease of automating them compared to the cost of labor.
Self-driving cars seem like a poor choice of market, since the task is very complex and there are also many people worldwide who are willing to work as drivers.
The sweet spot would be a task that is easy to automate and also doesn't have much human competition. Obviously the OG automation case of assembly lines fits this. LLMs can occupy this space since there are probably more people who can drive a car than can write grammatically correct natural language. I would put self-driving trains in this category as well but that's an outsider's perspective.
In addition to what others have said about pricing, you need to compare Waymo with Uber Black, not UberX or shared rides. Waymo says they are aiming to be a premium service because the cars (Jaguar I-pace) are much nicer and, of course, there’s the novelty of not having a driver.
At least in my city (state, really) the police have vacated their duties to enforce roadworthiness because hand-wringing quasiliberals said that enforcing the laws against bald tires and missing headlights was hurting the poor. Also my state does not have and never had annual safety inspections.
It's not a novelty that the driver in the driverless car won't hit on, or harrass a woman taking a trip late a night in an altered state. Uber and Lyft have paid out to settle thousands of lawsuits brought by women who get assaulted by drivers.
> originally thought autonomous cars would greatly reduce costs
In general, it's a mistake to think that selling price has anything to do with the costs. The price is simply what the market will bear. If people are paying Uber $x for a service, waymo would be stupid to price the same service much lower than that. If their costs are much lower, they would rather make that much more profit. If the rider wants a small piece of that pie they can invest in the business (via stocks in the case of Google).
I think more realistically google right now does not want to undercut uber and by extension real drivers who're doing it for a living.
They don't need that kind of bad publicity when potentially the licenses they operate under could be pulled given enough political pressure.
I wouldn't consider current pricing indicative of what it might ultimately be, consider that if aggressive enough google could divert a big portion of the countries car spend to themselves.
> "Drivers can activate Mercedes's technology. called Drive Pilot, when certain conditions are met, including in heavy traffic jams, during the daytime, on specific California and Nevada freeways, and when the car is traveling less than 40 mph. Drivers can focus on other activities until the vehicle alerts them to resume control. The technology does not work on roads that haven't been pre-approved by Mercedes, including on freeways in other states."
It's remarkable how often these significant limitations are ignored.
The difference between SAE Level 3 versus Level 2 is liability, not functionality. Conceptually, it would be relatively simple to create a Level 3 system that only worked in parking lots and never drove over 5 MPH. And yes, such a system would be "Level 3," which naively sounds better than "Level 2" because the number is higher.
But you could compare such a system that works only in parking lots against Tesla's Supervised FSD Level 2 system which controls the vehicle at prevailing speeds on all city streets and highways, executes left and right turns, stops at traffic controls, executes U-turns, parks, and everything else. Tesla doesn't want to assume liability yet because they are still iterating fast. The last few builds have been released about two weeks apart.
The functional domain of a Level 2 system can be significantly greater than a Level 3 system. And that is the case when comparing Tesla Supervised FSD versus Mercedes Drivepilot. It remains commendable for Mercedes to take on liability, but we should not kid ourselves. Doing so is a play for cheap media wins and punchy-sounding milestones like what we see in this article. It's not actually moving autonomy forward in any meaningful way when compared alongside Waymo and Tesla FSD. The scope of the Mercedes system is simply too narrow.
> The difference between SAE Level 3 versus Level 2 is liability, not functionality. Conceptually, it would be relatively simple to create a Level 3 system that only worked in parking lots and never drove over 5 MPH. And yes, such a system would be "Level 3," which naively sounds better than "Level 2" because the number is higher.
No one’s going to take liability if their technology is not ready for the conditions they want to work in. And no one’s creating systems that only work in 5 mph parking lots just to claim the L3 title. So comparison to hypothetical systems aren’t useful.
The limitations of the Mercedes system are only a little higher than the hypothetical parking lot Level 3 system. The point is to emphasize that a narrow Level 3 system is not very interesting compared to a broad Level 2 system. Technical capability > appeasing lawyers.
Uhh but a narrow Level 3 system will also in general do broad Level 2. You think these Mercedes can't do adaptive cruise control at 60mph wherever you want to activate it?
I would take a car with reliable, narrow L3 and reliable broad L2 (especially when the L3 is reliable in the most frustrating driving conditions to encounter) over broad pseudo-L3 any day.
Waymos are incredible, but Teslas are a much worse tradeoff than what MB is proposing here in my opinion.
> Uhh but a narrow Level 3 system will also in general do broad Level 2. You think these Mercedes can't do adaptive cruise control at 60mph wherever you want to activate it?
The Mercedes system will absolutely not follow navigation routes, change lanes, execute right and left turns, execute U-turns, stop at traffic controls, and so on. All of which FSD will do.
A broadly capable Level 2 system that is suitable for 99% of driving scenarios (with supervision) is way more interesting than a tightly constrained Level 3 system that is only suitable for <1% of driving scenarios.
> The difference between SAE Level 3 versus Level 2 is liability, not functionality.
No it's not. The difference is functionality, proof that the vendor believes they've achieved that difference is most clearly communicated via liability.
> The scope of the Mercedes system is simply too narrow
Producing systems that perform reliably under the prescribed conditions is called "good engineering."
Level 3 actually lets you take a break while driving. L2 systems don’t and therefore IMO don't provide significant benefits.
More range and more driving conditions just don’t matter here if I can’t read a book on the way to work then it’s not self driving just advanced cruse control.
You can read a book on the way to work. Before going fully remote, I did it every weekday, using a magnificent piece of technology called public transit.
And level 5 beats level 3, however I don’t have any public transport options on the way to work and nobody is selling level 5 self driving cars. Mercedes hasn’t mapped my roads either, but that’s a lower barrier than getting nationwide public transportation setup.
And this is the why I think pouring money on autonomous driving is a net negative for most but a small group of shareholders: a public transit infrastructure is expensive, but the net benefits are immense, and can be had today.
Every dollar spent on private cars reduces that infrastructure, and that's not just my own thinking, but also that of car manufacturers, as evident by their lobbying to derail public transit projects.
Further I don’t live in a city so my local government has zero interest in public transportation beyond school buses. We just don’t have the density where it makes any sense. When I did live in a city the issue wasn’t drivers it was the inability to scale roads for local demand. Self driving cars don’t change anything about that equation.
...those same Mercedes cars that have Level 3 also have a much more "powerful" and flexible Level 2 system built in, but this requires to keep a hand on the wheel and pay attention to the road (mentioned towards the end of this video, at around 6:15: https://www.youtube.com/watch?v=ZLz95Gw7g8c)
> Tesla doesn't want to assume liability yet because they are still iterating fast.
I would guess the reason for not wanting to assume liability is not their update cadence, but the same reason that made Mobileye fire them as a customer.
Are there human “watchers” in your waymos? Or are they completely autonomous?
They just started testing in my city and have humans sitting there not doing anything. So it makes for kind of an uncomfortable ride. Chitchatting with the Uber driver is the worst part of my ride, so I’m looking forward to just robots.
In SF, all the Waymo rides I've taken are completely autonomous... no watchers in the car. This has been the regular mode of operation here for quite awhile.
This isn't to say you won't see Waymo cars with drivers/watchers, but most of those seem to be transiting areas were they don't operate.
Of course there are no people in the car. I can’t take a walk anywhere without seeing at least one Waymo drive by without anyone in the front seat. This is old news. It’s really really funny reading these takes. We’re living in the future and there are still people making predictions about it.
Given that Mercedes is taking responsibility for accidents caused by the system I can see why they would launch cautiously. Less of an issue if it goes wrong.
The first actually autonomous car (under very specific circumstances) sold to the general consumer should be a historic day for Mercedes US, and the auto industry as a whole.
But there doesn't seem to be even a press release on mbusa.com?
Considering there are grand total of 65 cars available for sale, it's obvious that the company is still testing the waters. There's no reason to do a massive PR push until they are ready and can scale production.
Those 65 cars could easily be >10% of all new S classes available on dealer lots in the two states, the majority of customers order custom made or order special chauffer variants.
There was a bit of press a few years back when they got approval in Germany, but I’d imagine that Mercedes might be reluctant to bet the farm on this or link it to their public image; self driving cars have not, ah, exactly been an unqualified success thus far, and raising expectations could come back to bite them later.
"Drivers can activate Mercedes’s technology, called Drive Pilot, when certain conditions are met, including in heavy traffic jams, during the daytime, on specific California and Nevada freeways, and when the car is traveling less than 40 mph."
This is not what people think of when they see a headline like this. It's just a bunch of PR bs.
Exactly. As a mental exercise, people should consider how broad or narrow the functional domain needs to be for Level 3 (liability assumed by manufacturer) to be truly of note versus Level 2 (liability remains with owner).
Would a Level 3 system that only navigates parking lots at up to 5 MPH be notable versus a Level 2 system that drove everywhere? For me, a very narrow Level 3 system like Mercedes' is absolutely less interesting versus a very broad Level 2 system like Tesla's. The technical capability of Tesla's system is substantial. How important is taking liability, really, to the bigger picture of increasing the technical capability of autonomy? A narrow Level 3 system is not an entirely empty milestone, of course, but compared to the actual work of increasing driving capability, assuming liability seems like a game played primarily by deploying lawyers, not programmers.
> Drivers can activate Mercedes’s technology, called Drive Pilot, when certain conditions are met, including in heavy traffic jams, during the daytime, on specific California and Nevada freeways, and when the car is traveling less than 40 mph.
So this isn't any more of an advancement than Tesla's autopilot, which has been standard in every car for years.
Basically if Tesla decided to self-insure when autopilot is engaged, it would be the same.
Misleading headline. The cars aren't autonomous at all. They have self-insured cruise control under specific restricted conditions.
Isn’t this just a swing of the pendulum to a similarly misleading extreme?
Somewhere between “this car is autonomous” and “not autonomous at all” lies the truth of the matter.
These cars have the capability of driving autonomously under specific conditions. If you drive a lot under those conditions, this functionality may be compelling.
Comparing this to cruise control (the automatic variety of which has been available for many years at this point) also seems quite misleading.
The important distinction is that the Mercedes does not demand the driver be paying attention and ready to take over immediately if the car decides to pop out of self-driving mode. Tesla does not offer that.
You are guaranteed 12 or so seconds to take over. That means the system can handle anything, from moving over from emergency vehicles to unexpected debris while you take over
> Basically if Tesla decided to self-insure when autopilot is engaged, it would be the same.
So it's not the same as Tesla. Reading Reddit now that FSD has been available to not just those having bought into the hype has been interesting. It seems more stressful to use than just driving yourself.
Can confirm Tesla FSD is more stressful than driving yourself. I’ve been using it for a few weeks since they offered everyone free trial.
Seems to be improved from the previous version, but still requires a ton of interventions every single drive. It’s cool to experience for people who are tech savvy, but nowhere near the realm of being useful and replacing your driving needs.
Agreed. Every time I’ve used it or been in a Tesla as a passenger with FSD engaged, human intervention has been required several times each trip to avoid accidents, harm to pedestrians, or moving violations. It’s frankly embarrassing to use and terrifying to imagine the masses relying on it in any way in its current form.
Tesla FSD capabilities far exceeds all other car manufacture systems by a wide margin. Mercedes has Level 3, but that's only a legal thing, it says nothing about what the car can actually do.
Right now hundreds of thousands American Tesla owners are in their one month free trial FSD month and drive around with it on all kind of streets and weather conditions. There is no one else beside Tesla that is capable of doing this (no, not even Waymo).
Found the TSLA long. This is not the place to talk your book.
I have no idea what you're talking about with Waymo. The disengagement rates are orders of magnitude different. FSD is just simply not the magical secret sauce that you want it to be.
Tesla requires the driver to pay attention to the road at all times, even with "FSD". (And having tried it, you would definitely want to do so.) This system does not have that requirement, on roads where it's supported. What's the lie?
They are misleading the public in thinking that Mercedes is a leader in autonomous driving, but they are not. They are, including all other legacy car companies, so far behind Tesla, that they should be ashamed.
Tesla still has this one [1] up where the "not required driver" had to save the vehicle after it ran off the road into fences and etc numerous times [2] [3].
FSD ver. 12 is a massive improvement over ver. 11, but not something I'd want to trust my life on. The vision system alone is reason not to trust it (it's just a "multiple choice" image classifier). The actual driving in ver. 12 (now neural net vs C++) looks much smoother.
The continuing problem is that at the end of the day these are just rule-based systems with zero intelligence. They'll work fine 99.9% of the time, and then do something life-threateningly inappropriate in the 0.1% of cases they were not trained to handle.
The only way to make these things as safe as a human driver is to base them on human-level AGI, which is still a long ways off in the future. I don't know if Musk is planning to add his Grok "AI" into the mix, but having a hallucinating chatbot as backup system or driver is not my idea of safety.
Mercedes stands behind thir L3 and takes the responsibility for it taking the liability away from the driver.
Tesla can not do this no matter what they claim they can do. When their system fails is fails spectacularly and no company in their right mind would want to be liable for that.
It doesn't matter what Tesla can do in 80% of the time when it can't do it 20% of the time.
Consistently running down stationary children in the middle of a empty, well-lit road is not exactly what I am looking for in a autonomous vehicle. In my mind that qualifies as somewhere between criminal negligence and depraved indifference to human life.
It does, however, remain to be seen how the Mercedes system fares in its operating domain, so it is prudent to reserve positive judgement until more evidence emerges.
So I have to be ready to react to a mistake the car is making, not having any insight into the thought process of the car. I have to be a constantly vigilant backseat driver.
In theory the point of this system, unlike level 2 systems, is that you _don't_ need to do that. Of course, not having used it, I can't comment on how successful it is.
This is L3, so you don’t have to constantly supervise like you do a Tesla with FSD (L2). Mercedes even says you can watch movies or read books. They will achieve a minimum safety condition on their own, which for this system is stopping in the lane with hazard lights on. They also say they will take liability provided you haven’t misused the system or failed to properly maintain the car.
But L3 has this grey area where you can be notified to take over with sufficient time. IMO, It’s the most ambiguous level of automation. How this will work in practice, I have no idea.
am I the only one in the comment section that thinks this sounds great for commuting in heavy traffic at low speeds, which is the absolute most miserable type of driving?
ya sound of an alarm when we get through the traffic and I can actually drive, fine with me, I'll play on my phone
It operates under such specific conditions, it’s not very useful right now. Unless you spend a large part of your commute on clogged California highways, there’s not a big use case. But Mercedes would be hoping to expand that operational design domain (ODD).
Ultimately, I don’t think their L3 technology is the groundbreaking part (I think they use Nvidia Drive). It’s that they are willing to take liability inside their ODD. This will be a good experiment to see how liability shakes out for an automaker putting it in the hands of customers to be used conditionally vs someone like Waymo who doesn’t want anyone in the driver’s seat at all.
It’s embarrassing for Tesla because they pretended they had this solved and practically ready to ship eight years ago, started taking customer money; still take customer money on promises of “Real Soon Now”; and yet it’s a bunch of staid old Germans who actually ship it.
I’m so glad I didn’t fall for it. Nearly bought one, then took it for a test drive at the dealership and it couldn’t even make a left turn correctly. Then I rented one on another trip because I wanted to test it more extensively and the car locked up because I activated some stupid easter egg that started playing a song with an animated gif all over the screen…even getting out of the car and locking it didn’t stop the damn thing and I literally just had to sit on the side of the road and wait it out. The weird thing is I haven’t been able to find that easter egg mentioned anywhere.
We have seen those cars on the road for a while (this has been over the news for quite a bit already), they just didn't market the tech as L3 due to legal reasons. Also the liability cover clearly shows a level of confidence above that of other car makers.
Yes it is a very limited, but it is still quite a bit ahead of everyone else and if you're thinking we will get full selfdriving on any roads under any conditions in the near future than I think you're kidding yourself. The failure modes I have observed in driver assist (which is much easier than full self driving) with several car makes, really shows the tech is nowhere close yet.
I don’t quite understand Tesla’s insistence on vision-only. I don’t believe vision-only “FSD” will ever be good enough for mass adoption – even if it performs at human-level. Autonomous systems need to be better than humans to gain acceptance, and that’s simply not possible with a vision-only let’s-mimic-humans approach.
I agree with you, it seems like it's the most valuable first step. Just eliminating the daily rear-end collisions during commute would be a huge step forward.
This. It matters how you achieve level three. MBs solution requires a lot of legwork to build the datasets describing the specific roads where the L3 capability can then operate. There’s a massive amount to do in this approach to even come close to supporting a useful subset of all roads in their key markets.
To my mind, Tesla is playing a long game betting on achieving a more generalised solution that doesn’t rely on having a detailed database encompassing every inch of many roads in vast amounts of detail that have to be captured in a costly, resource intensive effort. That, or their vast fleet of vehicles is gathering the data for them to leapfrog the likes of MB to higher levels of autonomy using a fused approach in the future. No other manufacturer yet has a comparable on the road data gathering capability. It’s a matter of opinion right now which of these stands a better chance of succeeding at higher levels of autonomy alone - the best chance may be a synthesis of both, or something else entirely. Fanboy holy wars aren’t going to resolve any of this.
Definitely takes something to imply someone working in the space (on actual self-driving, not faked demos and empty promises) is engaging in a "fanboy holy war".
The simple fact is Mercedes made a massive leap that other manufacturers have not been able too, despite repeat implications they would do so many years ago. The fact that some of the existing L2 systems can even be activated on non-major highways is a farce, and overall this is a lot of waxing poetic about what could be, vs what is.
-
> He’s forever talking up hard stuff where the timeline always turns out longer than his prediction - well, he would say that wouldn’t he? Needs to keep people buying shares - albeit balancing that against losing credibility
Actually, I can see why someone saying this stuff with a straight face would assume everyone else is also a fanboy of something.
there were a couple automated urban lines in pre-90s. If we hadn't cannibalized the inter-urban trolley cars for public roads but private rails in the 1920s, who knows what would have happened.
I work on L3 technology but my opinion is of my own and not my employers. It’s laughable how people think FSD is better than L3 as of today. If it is, then why doesn’t Tesla sell it as one? The big difference is liability, can you play on your phone legally and safely while the car drives itself when autonomy is engaged? If so, it’s L3. If not, it’s L2.
140 comments
[ 5.1 ms ] story [ 212 ms ] threadSo far, autonomous operation is limited to relatively few situations, but most importantly it includes traffic jams (which is where human drivers quickly lose focus).
I have a feeling there is enough small print to allow them to wiggle out of most situations.
Since algorithms cannot have malice, the only offense left is negligence which for companies usually means a heavy fine and/or legal restraints, maybe prosecution.
And each state has its own set of laws. Dunno why this example popped into my head but - back during COVID when we had the first case in the state of Washington, we said “first case in the U.S.”
Those are pretty small numbers and strike me more as them testing the waters. It's neat to see some potential in the space after what felt like a total collapse with Waymo and Tesla especially not living up to expectations. It's not clear how good the tech is here, but it's a political victory for the space if nothing else.
In any case, the Mercedes system is what's known as a conditionally autonomous system. It can drive itself under specific conditions, but a human is still involved in operating the vehicle as soon as it exits that design domain. Waymo (and Cruise) are fully autonomous systems. Humans are not involved in the control loop and the vehicles will stop if they leave their operational domain. Tesla FSD is not an autonomous system at all and requires a human to always be in control.
No OP but there were multiple headlines of Waymo incidents just recently
- https://techcrunch.com/2024/02/13/waymo-recall-crash-softwar...
- https://www.theverge.com/2024/2/7/24065063/waymo-driverless-...
> In any case, the Mercedes system is what's known as a conditionally autonomous system.
Yes, but it's significant progress if it reaches consumer hands. This sounds like the typical Waymo marketing of "you can't get to Level 4 through Level 3". I bought into this at one point, but Waymo and Cruise have demonstrated that going straight to Level 4 with $100k+ (probably much more than that) of sensors and corresponding maintenance doesn't scale.
If other companies thought they’d hitch on to Waymo after they started giving rides in 2017, then that’s on them. Because Waymo has been pretty clear that their scaling is going to be methodical and not instantaneous.
Waymo won't keep a monopoly on this forever. There are still high barriers to entry, but they will fall sooner or later. If there is money to be made, competition will come.
I'm not sure about barriers to entry lowering soon. First off, this is an incredibly hard problem no one else seems to have cracked (and even Waymo has a long way to go). Second, they could help influence regulation that requires extremely high safety records, mandatory sensors, etc. which could make it harder for others to compete. It's all very early though.
"Even if Waymo's schedule slips a few months and it introduces a self-driving car service in the middle of 2018 instead of late 2017, that will still give the company a multiple-year head start over most of its rivals. And it would confound skeptics who insist that full self-driving technology is still years away."
https://www.ft.com/content/3355f5b0-539d-11e8-b24e-cad6aa67e...
Waymo forecast to capture 60% of driverless market
Group’s dominance by 2030 will force carmakers to adopt its technology, says report
Investment bank UBS estimated global revenues from self-driving technology by 2030 will be up to $2.8tn, with Alphabet’s Waymo unit the global leader.
I read the comment as saying that Waymo's product itself is not living up to expectations.
It looks like it's slowly but steadily expanding, at least in sunnier climes, and offering a decent user experience.
Anyway, the question was about expectations - and those came with a timeline.
The mention was of a "collapse" in the space, too-- which can't be true if we're still getting there.
I expect the timeframe of 2030 is when we get most of the way there. Even exponential growth takes time to reach a big world. Most of what Waymo has left to do is rollout.
I don’t really know the numbers but Waymo adds something like 50k worth of hardware into the cars. That is not a viable approach for consumer cars (Mercedes, Tesla, etc).
Waymo driving in SF is way better than human. It's not timid either.
This Mercedes post is marketing - it's not really what people consider autonomous given the constraints.
> "Drivers can activate Mercedes's technology, called Drive Pilot, when certain conditions are met, including in heavy traffic jams, during the daytime, on specific California and Nevada freeways, and when the car is traveling less than 40 mph. Drivers can focus on other activities until the vehicle alerts them to resume control. The technology does not work on roads that haven't been pre-approved by Mercedes, including on freeways in other states."
The above makes it nearly useless and far less useful than what Tesla has been shipping for years imo.
I originally thought autonomous cars would greatly reduce costs but considering the tens of billions dumped into these companies, they can't charge lower prices as they have to recoup investments.
> As of the latest available information, Waymo is not yet profitable. Alphabet’s Q2 2023 earnings report indicated that its “Other Bets” category, which includes Waymo, had revenues of $285 million and an operating loss of $813 million. This was a decrease in operating loss from $1.34 billion in the second quarter of 2022, but still significant. Over the last five years, “Other Bets” has generated $3 billion in cumulative revenue but incurred $20 billion in operating losses, with Waymo constituting the bulk of these losses.
I just don't see the end game. Are marginal rides not profitable even at Uber fates? Uber for instance was losing money because they were taking rev from profitable cities to give incentives and build up drivers in the new markets. But Waymo doesn't have that problem. Is it just GPU spend that they can eventually stop?
The business model is pretty simple. Robotaxis can operate much cheaper than humans and simultaneously provide a better service. Competing at the same cost as existing services makes them massively profitable. You can unlock additional efficiencies through things like fleet management, not supporting unprofitable cities the way Uber has to, vehicles designed for longer service lives and so on. Most companies plan to use that market (and closely associated ones like food delivery) as a profitable beachhead to scale down the technology to the point where it can be applied more ubiquitously in things like consumer vehicles. Some are more vocal about this than others.
Self-driving cars seem like a poor choice of market, since the task is very complex and there are also many people worldwide who are willing to work as drivers.
The sweet spot would be a task that is easy to automate and also doesn't have much human competition. Obviously the OG automation case of assembly lines fits this. LLMs can occupy this space since there are probably more people who can drive a car than can write grammatically correct natural language. I would put self-driving trains in this category as well but that's an outsider's perspective.
In general, it's a mistake to think that selling price has anything to do with the costs. The price is simply what the market will bear. If people are paying Uber $x for a service, waymo would be stupid to price the same service much lower than that. If their costs are much lower, they would rather make that much more profit. If the rider wants a small piece of that pie they can invest in the business (via stocks in the case of Google).
They don't need that kind of bad publicity when potentially the licenses they operate under could be pulled given enough political pressure.
I wouldn't consider current pricing indicative of what it might ultimately be, consider that if aggressive enough google could divert a big portion of the countries car spend to themselves.
It's remarkable how often these significant limitations are ignored.
The difference between SAE Level 3 versus Level 2 is liability, not functionality. Conceptually, it would be relatively simple to create a Level 3 system that only worked in parking lots and never drove over 5 MPH. And yes, such a system would be "Level 3," which naively sounds better than "Level 2" because the number is higher.
But you could compare such a system that works only in parking lots against Tesla's Supervised FSD Level 2 system which controls the vehicle at prevailing speeds on all city streets and highways, executes left and right turns, stops at traffic controls, executes U-turns, parks, and everything else. Tesla doesn't want to assume liability yet because they are still iterating fast. The last few builds have been released about two weeks apart.
The functional domain of a Level 2 system can be significantly greater than a Level 3 system. And that is the case when comparing Tesla Supervised FSD versus Mercedes Drivepilot. It remains commendable for Mercedes to take on liability, but we should not kid ourselves. Doing so is a play for cheap media wins and punchy-sounding milestones like what we see in this article. It's not actually moving autonomy forward in any meaningful way when compared alongside Waymo and Tesla FSD. The scope of the Mercedes system is simply too narrow.
No one’s going to take liability if their technology is not ready for the conditions they want to work in. And no one’s creating systems that only work in 5 mph parking lots just to claim the L3 title. So comparison to hypothetical systems aren’t useful.
I would take a car with reliable, narrow L3 and reliable broad L2 (especially when the L3 is reliable in the most frustrating driving conditions to encounter) over broad pseudo-L3 any day.
Waymos are incredible, but Teslas are a much worse tradeoff than what MB is proposing here in my opinion.
The Mercedes system will absolutely not follow navigation routes, change lanes, execute right and left turns, execute U-turns, stop at traffic controls, and so on. All of which FSD will do.
A broadly capable Level 2 system that is suitable for 99% of driving scenarios (with supervision) is way more interesting than a tightly constrained Level 3 system that is only suitable for <1% of driving scenarios.
Only one of them allows you to take liability.
No it's not. The difference is functionality, proof that the vendor believes they've achieved that difference is most clearly communicated via liability.
> The scope of the Mercedes system is simply too narrow
Producing systems that perform reliably under the prescribed conditions is called "good engineering."
More range and more driving conditions just don’t matter here if I can’t read a book on the way to work then it’s not self driving just advanced cruse control.
Every dollar spent on private cars reduces that infrastructure, and that's not just my own thinking, but also that of car manufacturers, as evident by their lobbying to derail public transit projects.
Further I don’t live in a city so my local government has zero interest in public transportation beyond school buses. We just don’t have the density where it makes any sense. When I did live in a city the issue wasn’t drivers it was the inability to scale roads for local demand. Self driving cars don’t change anything about that equation.
...those same Mercedes cars that have Level 3 also have a much more "powerful" and flexible Level 2 system built in, but this requires to keep a hand on the wheel and pay attention to the road (mentioned towards the end of this video, at around 6:15: https://www.youtube.com/watch?v=ZLz95Gw7g8c)
I would guess the reason for not wanting to assume liability is not their update cadence, but the same reason that made Mobileye fire them as a customer.
If you read the article, they essentially just have Tesla Autopilot with self insurance and driver monitoring removed.
This "landmark achievement" only works on freeways and under 40 mph.
Waymo is better than anyone else out there by a mile. Source: I take one daily.
They just started testing in my city and have humans sitting there not doing anything. So it makes for kind of an uncomfortable ride. Chitchatting with the Uber driver is the worst part of my ride, so I’m looking forward to just robots.
This isn't to say you won't see Waymo cars with drivers/watchers, but most of those seem to be transiting areas were they don't operate.
But there doesn't seem to be even a press release on mbusa.com?
I was pretty thrilled to start using ChatGPT as a math tutor, nowadays I'm just "meeh" (while still using it as a math tutor).
This is not what people think of when they see a headline like this. It's just a bunch of PR bs.
Would a Level 3 system that only navigates parking lots at up to 5 MPH be notable versus a Level 2 system that drove everywhere? For me, a very narrow Level 3 system like Mercedes' is absolutely less interesting versus a very broad Level 2 system like Tesla's. The technical capability of Tesla's system is substantial. How important is taking liability, really, to the bigger picture of increasing the technical capability of autonomy? A narrow Level 3 system is not an entirely empty milestone, of course, but compared to the actual work of increasing driving capability, assuming liability seems like a game played primarily by deploying lawyers, not programmers.
So this isn't any more of an advancement than Tesla's autopilot, which has been standard in every car for years.
Basically if Tesla decided to self-insure when autopilot is engaged, it would be the same.
Misleading headline. The cars aren't autonomous at all. They have self-insured cruise control under specific restricted conditions.
Isn’t this just a swing of the pendulum to a similarly misleading extreme?
Somewhere between “this car is autonomous” and “not autonomous at all” lies the truth of the matter.
These cars have the capability of driving autonomously under specific conditions. If you drive a lot under those conditions, this functionality may be compelling.
Comparing this to cruise control (the automatic variety of which has been available for many years at this point) also seems quite misleading.
So it's not the same as Tesla. Reading Reddit now that FSD has been available to not just those having bought into the hype has been interesting. It seems more stressful to use than just driving yourself.
Seems to be improved from the previous version, but still requires a ton of interventions every single drive. It’s cool to experience for people who are tech savvy, but nowhere near the realm of being useful and replacing your driving needs.
The fact that Tesla doesn't self insure tells you everything you need to know.
Writing that off as some footnote is missing the forest for the trees.
Right now hundreds of thousands American Tesla owners are in their one month free trial FSD month and drive around with it on all kind of streets and weather conditions. There is no one else beside Tesla that is capable of doing this (no, not even Waymo).
It is not only. Tesla is unable to provide same guarantees or does not trust FSD.
I have no idea what you're talking about with Waymo. The disengagement rates are orders of magnitude different. FSD is just simply not the magical secret sauce that you want it to be.
Watch a Tesla FSD 12.3.4 video, then one of Mercedes and see for yourself.
Tesla still has this one [1] up where the "not required driver" had to save the vehicle after it ran off the road into fences and etc numerous times [2] [3].
[1]: https://www.tesla.com/autopilot
[2]: https://www.classaction.org/media/matsko-v-tesla-inc-et-al.p...
[3]: https://www.nytimes.com/2021/12/06/technology/tesla-autopilo...
The continuing problem is that at the end of the day these are just rule-based systems with zero intelligence. They'll work fine 99.9% of the time, and then do something life-threateningly inappropriate in the 0.1% of cases they were not trained to handle.
The only way to make these things as safe as a human driver is to base them on human-level AGI, which is still a long ways off in the future. I don't know if Musk is planning to add his Grok "AI" into the mix, but having a hallucinating chatbot as backup system or driver is not my idea of safety.
Tesla can not do this no matter what they claim they can do. When their system fails is fails spectacularly and no company in their right mind would want to be liable for that.
It doesn't matter what Tesla can do in 80% of the time when it can't do it 20% of the time.
Consistently running down stationary children in the middle of a empty, well-lit road is not exactly what I am looking for in a autonomous vehicle. In my mind that qualifies as somewhere between criminal negligence and depraved indifference to human life.
It does, however, remain to be seen how the Mercedes system fares in its operating domain, so it is prudent to reserve positive judgement until more evidence emerges.
Am I on the hook for accidents?
But L3 has this grey area where you can be notified to take over with sufficient time. IMO, It’s the most ambiguous level of automation. How this will work in practice, I have no idea.
ya sound of an alarm when we get through the traffic and I can actually drive, fine with me, I'll play on my phone
There's debate about this:
http://safeautonomy.blogspot.com/2023/09/no-mercedes-benz-wi...
Ultimately, I don’t think their L3 technology is the groundbreaking part (I think they use Nvidia Drive). It’s that they are willing to take liability inside their ODD. This will be a good experiment to see how liability shakes out for an automaker putting it in the hands of customers to be used conditionally vs someone like Waymo who doesn’t want anyone in the driver’s seat at all.
It’s embarrassing for Tesla because they pretended they had this solved and practically ready to ship eight years ago, started taking customer money; still take customer money on promises of “Real Soon Now”; and yet it’s a bunch of staid old Germans who actually ship it.
Yes it is a very limited, but it is still quite a bit ahead of everyone else and if you're thinking we will get full selfdriving on any roads under any conditions in the near future than I think you're kidding yourself. The failure modes I have observed in driver assist (which is much easier than full self driving) with several car makes, really shows the tech is nowhere close yet.
L3 is 100% the groundbreaking part because it's when the manufacturer has to take liability and it's where people can actually afford it.
I believe in the technology "trickling up" for L3 much more than I believe tech in our million dollar robotaxis trickling down to consumer vehicles.
The simple fact is Mercedes made a massive leap that other manufacturers have not been able too, despite repeat implications they would do so many years ago. The fact that some of the existing L2 systems can even be activated on non-major highways is a farce, and overall this is a lot of waxing poetic about what could be, vs what is.
-
> He’s forever talking up hard stuff where the timeline always turns out longer than his prediction - well, he would say that wouldn’t he? Needs to keep people buying shares - albeit balancing that against losing credibility
Actually, I can see why someone saying this stuff with a straight face would assume everyone else is also a fanboy of something.