Mercedes is the first automaker to offer Level 3 self-driving in the US
insideevs.com
insideevs.com
Rephrasing it: I don't want to be the one preventing the car from doing stupid things. I want the car to be preventing me from doing stupid things.
For example my wife's previous Toyota would gently shake the steering wheel if I were to switch lanes without using the blinker. Or it'd warn me if I were to put my blinker while a car was in the lane I was planning to go to. Or take start and stop trafic, bumper to bumper in trafic jams... I want the car to prevent me from bumping in the car in front of me if I'm not careful enough (the Toyota didn't do that).
Stuff like that.
I find it very stressful to have an "autonomous but not really" car that I must baby-sit. I want instead to feel more relaxed by knowing that the car is probably there to help me should something be about to go wrong.
You get way too "distracted" with those technologies imho, but you still need to check what the car is doing.
So if before I was relaxed and driving in auto-pilot mode not even thinking about it, I'm now more tense and looking over what the car is doing.
Obviously blind spot, simple cruise control or beeps that warn you in traffic the car ahead is moving are fine, but the more complex assistive technology gets the more you need to check what it's doing.
Stuff like lane assist is just...unreliable in many situations. I was more than once driving on streets that from a second to the other had the white marks disappear and the car drift dangerously off road.
I just find more relaxing and safer normal driving (unless I was doing thousands of kilometers per week I guess) than being this controller who has to check what the car is doing.
As far as "the car preventing me from doing stupid things" goes it's good in theory, but requires an excellent implementation to not suck. When driving in traffic it's basically 100% of the time that my car beeps at me when I put my turn signal on, because I'm either passing or waiting for a car to pass me when I signal. I would turn it off, but I can't do so without also disabling the RTCA which is a great feature.
There is a post going around just this morning about a Tesla that stopped itself on a major bridge in San Diego and caused a huge accident and pileup. I don't want technology that has the ability to take action like that on it's own. I don't trust it's decision making enough for that.
The idea is baffling to me that anyone lets self-driving software pilot a multi-ton steel missile that contains the only body they will ever get hurtling through an asphalt field full of other humans and an environment that is often not predictable and actively hostile. I write automation for a living, and like everyone else I know that does the same I want automation in safety-critical applications to be as simple, bulletproof, and limited as possible. Self-driving software does not fall into that category right now. Some day it probably will. Until that time I won't be buying it.
There is still almost no information on this crash, probably because Tesla HQ stopped Elon from writing blog posts to put the record straight on PR-harming issues [0]. The fact of the matter is that the person was not using "Full Self-Driving Beta", given it does not activate on highways yet; the only real possibility is Enhanced Autopilot, ie. the driver hit the left turn stalk, the car moved over for them, and it detected a stopped car in front of it, either triggering the AEB system (which can slam on the brakes completely independently from every other system) or TACC detected a car and came to a safe stop as if the traffic in that lane was already stopped.
My guess is confused pedals or both pedals being depressed at the same time, since the accelerator will not spin the tires if you're on the brake whatsoever.
This is why I think the only direction in which this can all be headed is “human roads” and “robot roads”. When the robot cars all start talking to each other/the road 100,000x/second, there won’t be a need for traffic lights, and so anytime an accident happens, it will be because of a human.
BMW and Mercedes switch places periodically in technical and design leadership, and I think it's Mercedes's time. BMW has been going off the rails recently...
I was all excited about replacing my 2007 Prius with the new one, but now I'm not so sure. Perhaps I'll wait for the Lightyear 2 which is hugely more slippery at a Cd of 0.175.
The prototype design looks much more sleek to me, but I'm not a car guy, and I think all modern cars are ugly.
There was never a reason, aerodynamic or not, to make the design look so ridiculous except if you’re the GM of the 3 Series or S class you don’t want someone else eating into your sales with a different product line.
There's a lot of visual conflict between the two motifs, as shapes that normally have boxy, sculptural depth get squashed down into basically 2D features on the surface of the smoothed out, mouse-like volume of the overall package.
The designers seem to have been stuck with the assignment of putting a Mercedes mask onto a wide, flattened jellybean. The result is mostly successful from the side views, but seems pinched and overwrought from the front.
Additionally, some of the grille shapes on the front are likely vestigal, as EVs don't require the same ventilated cooling as combustion engines. Much to the frustration of car designers, one of the key identifying features of any brand has historically been the design of the now obsolete front grille; the functional purpose that served as an aesthetic starting point is now gone. (You can trace the gradual acceptence of grille obsolescence in the evolution of Tesla's designs.)
- OEDR is required, which is "monitoring the driving environment (detecting, recognizing, and classifying objects and events and preparing to respond as needed) and executing an appropriate response to such objects and events (i.e., as needed to complete the DDT and/or DDT fallback [fallback to driver])".
- for DDT fallback, "At Level 3, an ADS is capable of continuing to perform the DDT for at least several seconds after providing the fallback-ready user with a request to intervene. The DDT fallback-ready user is then expected to resume manual vehicle operation, or to achieve a minimal risk condition if s/he determines it to be necessary." It also mentions a scenario like a blown tire, disappearing lane lines, or a malfunctioning sensor/spatial input device.
- "DDT performance by a Level 3 ADS assumes that a fallback-ready user is available to perform the DDT as required. There is no such assumption at Levels 4 and 5."
- "A Level 3 highway overtaking assistance feature automatically performs the lateral and longitudinal vehicle motion control actions, as well as associated OEDR, necessary to pass a slower-moving vehicle on a multi-lane highway when activated by the driver or fallback-ready user." .
Otherwise, the ODD (operational design domain, ie. what conditions the level 3 system can perform in) is entirely defined by the manufacturer. The document includes low speed traffic as an example of level 3 operation:
> A Level 3 traffic jam feature performs the complete DDT on a fully access-controlled freeway in dense traffic, but requires a human driver to operate the vehicle upon ODD exit (e.g., when traffic clears, as well as before entering the congested freeway, and again upon exiting it).
Some other goodies:
- level 4 "May delay user-requested disengagement", and level 4 may still have ODD, although it cannot assume a driver fallback is available
DDT = dynamic driving task
ADS = automated driving system
Level 2 is 0 seconds; the driver has to watch constantly.
Level 4 is indefinite, or alternatively we could say "minutes". The car can pull over if necessary.
For this particular level 3 system, it's 10 seconds. That seems pretty reasonable for level 3.
And as you said, the ODD of levels 3 and 4 is entirely up to the manufacturer. Only level 5 has a domain by definition, which is matching human ability.
I think Level 4 is clearer because it's closer to public transit. Somebody else (the machine) is driving, so if I'm day dreaming, or indeed asleep, nothing bad happens. I can listen to tunes, have an argument with someone, breast feed a baby, whatever, the driver is driving the car, I'm just a passenger.
As I understand it, there's no reason you'd need a Driving License to ride in a Waymo for example. Waymo is driving, you're in the back like a cab, why would you need a license?
You "need" a license and insurance because Waymo and other self-driving companies will never ever allow a scenario where they're liable for accidents. They'll move heaven and earth to prevent it.
Because according to the Waymo One public service rules [1] you need to be 18 or older to use the app and order a Waymo, but they don't mention anything about requiring a license or insurance.
It would also be a weird requirement, since you sit in the back and couldn't control the car even if you wanted to.
What if staying where he is poses a greater risk to Jim than driving on the road? Why should the car decide for Jim what an acceptable level of risk is, when Jim (presumably) has more information to make that decision?
1. The elevator does not belong to you. The building manager setting whatever policies on its usage that they want does not violate your right to use your property however you like. For it to be analogous, during a blizzard the road would have to refuse to let your car onto it.
2. While you use the elevator, no one else (such as fire crews) can use it. You driving your car does not prevent anyone else from doing the same.
3. In a fire, there are alternative routes of escape other than the elevator. I suppose the same is not true in a skyscraper. Do elevators shut down in those too? If you're several kilometers from, say, a hospital, you may not have any reasonable alternatives to driving to the hospital in inclement weather.
I wouldn't think that the idea that hardware should obey its operator would need to be defended in HN of all places.
Generally, yes. In very tall buildings it will often be possible to seek authorization to use elevators for evacuation but you need to specifically arrange for the elevators and rest of the building to be suitable, in most cases it makes more sense for the evacuation plan to rely on stairwells instead. Walking down even 100 flights is a huge annoyance, but it's not impossible for able-bodied adults.
You'll know if you visit a building where the elevators are to be used in evacuation because they'll say very clearly on them that's what you should do, it's part of the regulations authorizing such things in the US and other countries I looked at with this allowance.
In addition most tall residential buildings have "Remain in place" strategies where evacuation would rarely be needed because the building is supposed to be designed so that most occupants are in no danger from likely emergencies such as a fire in another residence.
> I wouldn't think that the idea that hardware should obey its operator would need to be defended in HN of all places.
"I am entitled to whatever I want" is indeed a pretty common libertarian nonsense we see on HN but it tend to get shot down more often than not. There's no reason that hardware should be designed to do unreasonable things just because that's "obeying its operator".
You're not talking about designing a piece of hardware to do unreasonable things. You're talking about designing a piece of hardware to be unable to do arbitrary things. Why do I say arbitrary? Because the designer is not omniscient. They cannot conceive of all the possible future situations in which their product will be used. Are you really unable to imagine situations where it might be necessary to drive on the road regardless of the conditions outside?
That aside, this has nothing to do with libertarianism or "doing whatever you want". I'm not saying if you decide to drive on the road in such conditions you should not be liable if you cause an accident. I'm saying, simply, "people should be able to use their property however they like". That doesn't seem like such a controversial idea to me. Or are you saying you shouldn't be able to do that? You don't trust yourself to use your property responsibly?
I would put it up to the "reasonable person" test. I'm not sure what's dubious about the fact that "human level" doesn't and can't mean you're a perfect match to every human.
> I think Level 4 is clearer because it's closer to public transit.
I don't know if it's clearer in the general case. Basically any set of operating restrictions that is sub-human counts as level 4, and you could have a detailed and nuanced list of operating restrictions on level 4 that's just as complicated or more complicated than a level 5 setup.
From the opening paragraph [0]:
“This document does not provide specifications, or otherwise impose requirements on, driving automation systems (for further elaboration, see 8.1).
[1]: “8.1 This document is not a specification and imposes no requirements. This document provides a logical taxonomy for classifying driving automation features (and ADS-equipped vehicles), along with a set of terms and definitions that support the taxonomy and otherwise standardize related concepts, terms and usage in order to facilitate clear communications. As such, it is a convention based upon reasoned agreement, rather than a technical specification.“
0: https://saemobilus.sae.org/download/?fileFormat=pdf&cid=1000...
1: https://saemobilus.sae.org/download/?fileFormat=pdf&cid=1000...
https://blog.codinghorror.com/the-2030-self-driving-car-bet/
(Note that there's an escape hatch here, which is that the AI falls back to some human in a call center for guidance to get out of sticky situations. In my opinion that still doesn't qualify as level 5.)
Then this is close to impossible :-)
The existing road infrastucture maintenance burden is already crushing financially, imagine if we add smart roads to it.
In this case, the original goal is to get people from A to B. And, pardon the pun, but cars and roads are a dead end solution.
https://youtu.be/knbVWXzL4-4?t=21
Just think about any large scale governmental software project and amplify that 10-100x. That's what I imagine :-)
> The design of the website was overseen by the Centers for Medicare and Medicaid Services and built by a number of federal contractors, most prominently CGI Group of Canada. The original budget for CGI was $93.7 million, but this grew to $292 million prior to launch of the website. While estimates that the overall cost for building the website had reached over $500 million prior to launch and in early 2014 HHS Secretary Sylvia Mathews Burwell said there would be "approximately $834 million on Marketplace-related IT contracts and interagency agreements," the Office of Inspector General released a report in August 2014 finding that the total cost of the HealthCare.gov website had reached $1.7 billion and a month later, including costs beyond "computer systems," Bloomberg News estimated it at $2.1 billion.
That's not an example of what it would cost, though.
> Just think about any large scale governmental software project and amplify that 10-100x. That's what I imagine :-)
If I imagine the software costing ten million dollars, and then balloon it to a billion, I think that's still only half a percent of the annual road budget.
As for smart roads, what would these imply? Bear in mind maintenance costs and the fact that asphalt is 100% recycled and reusable.
The car would always have other sensors to fallback on, but this would reduce the amount of time that the DDT disengages due to inclement weather or other road hazards.
Add to this emitters in cones and possibly emergency vehicles.
I think useful "smart" road infrastructure would necessarily be horribly complicated, since the point of it would be to offload otherwise necessary complexity from the vehicles using the road. Perhaps the road could track the real time kinematics of all the cars on the the road using some combination of cameras, lidar, UWB, etc. It could then communicate that information via RF to the vehicles that want to receive it, along with local mapping information.
The automotive industry has been excited about "vehicle-to-vehicle" comms for the past 15 years aka V2V. Technically, all the standards that have emerged from the industry's efforts seem lame and nothing interesting has come of it. The FCC granted a chunk of 5Ghz spectrum for automotive DSRC use, and have subsequently taken back most of it for lack of use. It's still allocated for automotive, but it's now in the hands of the cell modem providers to do something useful with it.
Ah yes, trains. :)
The design and engineering activities involved in solving this would be very enjoyable work, and the human benefits when (eventually) operational would be huge.
Every solution would need to be custom for its city/region, but one of the common ideas is concentric circles of trains, spaced at appropriate(ly large) distances, with trains running between in perpendicular direction. Basically wheels and spokes.
The wheel trains could stay in motion always, only slowing at connection points where the perpendicular trains would temporarily turn and run in parallel. Both trains would slow, cars would connect, passengers would come and go, and then trains would separate. Perpendicular train curves back in for its perpendicular direction of travel, and "wheel" train picks up speed and heads for the next spoke connection. Obviously any problem causes the joined trains to slow, even to a stop if necessary (to prevent human harm).
There are so many different possibilities here, but ultimately I think it should be possible for a human to walk 5-10 minutes to some station, get on some people mover, and eventually get out of another people mover within 5-10min walking distance of their ultimate destination... all without ever setting foot outside of a train aside from start/end points.
The usual challenge is that the cities are old and organically structured. So the possible paths for public transport are limited; and there's very little space for big interchanges.
But what exists is way, way better than nothing. With trains, metros, trams, and buses (and walking), you can get around quite well. It just can take a long time due to timing mismatches between modes.
When the tram comes only once every 15 min, and your metro drops you without enough time to walk (or run!) to the tram, it means you now wait 15min for the next one. When I was commuting to work in Amsterdam via tram+metro, it took me at minimum 30 minutes with walk+tram+metro+walk. If I rode my bike, it was the same. But if timings didn't work out well, the public transport option could be +25 min longer... so 55min. And on cold sh*tty rainy days (half of winter), the metro would be jammed with people such that you might need to stand and wait for the next one. The bike option was always possible, but it's no fun to ride in cold rain with wind.
In summary, it was better than 60 minutes of bumper to bumper driving commute in Dallas/Fort Worth, but it had much room for better alternatives... if one could go greenfield.
(Or https://archive.org/details/isbn_9780739410516 if you want to actually read it.)
> It’s much saner to have one car break down and allow 999,999 people to keep on schedule.
Obviously that's very wrong, as we know that one car breaking down on the freeway has a major negative effect on everything behind it. And if it's something bigger than one car... I recall one evening in Dallas when a chemical tanker turned over and created a situation which required all traffic on a major freeway to take an exit, use the access road, and then eventually return to the freeway. Even though it was after evening rush out, the ripple effect eventually spanned to two sequentially related freeways. Basically long road travel in the city stopped for hours.
In my mental fantasy, I see two vertical levels. One is for passengers and operates with more connections and stops, and one is for cargo (where transfers can be fully automated like with a successful airport luggage routing system... or probably some kind of system which Amazon has already perfected at scale).
They two improvements I see over today's public transport realities are:
1. If greenfield is possible, which unfortunately it really is not (unless we are planning a whole new urban environment), there is a physical network design which balances maximum travel time from any point to any point. This would be something elevator designers have already managed for big towers, I would think. But this time, it's in two dimensions.
2. There are novel temporary linkings of different vehicles to allow passengers to transfer from one to another without stopping either vehicle. I suppose in the extreme case, each passenger could be in a small sit/stand capsule, and some plucking and moving of capsules could be done by robotic systems which are aware of everything (including route/dest.)
Really, these are just lame solutions based on our current understanding of nature. Teleportation is the best answer ;).
This would solve a lot of the "last mile" problems that we currently have with train stations in the US (need to park at one terminus, need to get picked up at the other, can't deal with accidents on rails, etc.).
When they are on water, we call them "ferries".
Looking into it, it seems they were popular decades ago, but they've been all but shut down by the 2010s due to unprofitability, and only a handful of seasonal routes remain now.
Though who knows, if gas prices keep increasing, they might enjoy a brief resurgence for a decade or so until enough people have switched to EV.
But it will take quite a bit of an increase. I looked at the longest route, 1300km Hamburg-Verona (which would actually interest me!) and it costs 400 to 600€, depending on the date. That's well over twice as much as the current gas prices, even being generous; while I guess it includes sleeping arrangements, you could stay in a very fancy hotel with the money you saved.
There's a high mountain pass with a glacier being in the way, the Lötschenpass. There's only a hiking path over it. A few kilometers to the west, there's another high mountain pass, the Gemmipass that has an extremely steep drop to the south such that in medieval times not even a mule track was possible. People avoided the rockfaces by going over an even higher saddle to which they got on by a steep grassy ridge. Today there's a cable car and hiking path blasted into the rocks, somewhat like the path to Angel's Landing in Zion National Park.
Up to today there are no roads over that mountain range between the Cantons of Bern and Valais for about 100 km. Only the Lötschberg railroad tunnel. People from the eastern part of Valais "have trains you can just drive your car into".
I myself have a customer card of the railroad operator because it's the most direct way to the south, and we go about once a month. They also offer transportation through a second mountain pass to Italy. After 35 minutes of driving and after one hour of relaxing we enjoy Italy's dolce vita.
There are other railroad operators taking cars. I know one in the Canton of Grisons, and between France and England there's another one.
The problems of last mile train stations in the US are completely self-inflicted, to cater for the current status quo of car-centric development, you don't need massive parking lots in central train stations for the majority of big population centres in Europe.
Your suggestion requires that we demolish and redo entire swathes of cities before we would gain any benefits. My suggestion can be implemented piecemeal on specific corridors and start gaining benefits immediately.
Path dependence is a thing. Political will is finite with short time horizons. Something which can be implemented piecemeal is less optimal but more likely to gain enough momentum to deliver real benefits.
At some point it'll be needed to do it, the longer the wait the more painful it will become. As you pointed out, it's already pretty fucking painful to achieve it, introducing new features unto an inadequate system is just compounding the issues in the long-term for just a small short-term gain, it's not really smart to do it.
Instead of starting to properly address the issue of transportation at scale, your suggestion will just add a component of "smartness" to a system that is already overloaded and definitely not good for society. Car-dependency is not a net positive. I'm not against people using and having cars but depending on them for transportation between large population centres is an absurd waste of resources for a society...
Simple and easy solutions have a tendency to create even larger problems, it's unfortunate there's no political will in the USA to revamp its transportation system because it's clear to not be the most efficient way for your society to move around.
https://www.statista.com/statistics/449539/europe-railway-sh...
Mercedes appears to be indemnifying drivers when its tech is engaged. That's frankly the bigger first for me than SAE Level 3.
Didn't Ford do a similar thing with their explorer? They knew about a roll over issue but decided that it was finacilly cheaper to pay the lawsuits than recall the vehicle.
Sure hope it was! Point is they're putting their money where their mouth is. That's new.
Yes, the standard requires that the driver be given "a few seconds" to resume before the system disengages, but things can change dramatically in a few seconds at highway speeds.
> things can change dramatically in a few seconds at highway speeds
In the immediate term, "brake enough not to hit things" is a very good default response and a computer should be better than a human at doing so.
Really?
Tesla has done exactly that - refusing liability because autopilot returned control to the driver in less than a second before the crash.[0]
No one called them out on it. There was a lot of hand wringing and feigned outrage but Musk's ass needs to be kissed and it ain't gonna kiss itself.
[0](https://www.google.com/amp/s/techcrunch.com/2022/05/18/nhtsa...)
On a page of statistics they put together, they said "we count any crash in which Autopilot was deactivated within 5 seconds before impact", and that seems sufficient for a level 2 system.
Not good enough to take a nap but good enough to be texting or reading.
I’d be surprised if that wasn’t the case in Germany, independent of any decision Mercedes would make beyond the one to make and sell the car.
Tesla doesn't have anything that's eyes off yet. You need to monitor the road. Tesla seems to be going for breadth first (FSD beta working everywhere and doing every manouver) and not depth (autonomy only for highway without changing lanes but reliable enough for eyes off)
It's most likely also much easier to make a car drive autonomously on highways versus on smaller roads, city roads, dirt roads etc.
They're promising 10 seconds of warning when the car wants you to take over.
"Warning, take the wheel! Death imminent" takes a good 6 seconds to say :P
Just "Warning, take the wheel!" gives you time to pause between each repetition and still say it 5 times.
Tesla's Autopilot has long been able to do this in this domain, and it does it well; they just don't claim liability yet because $reasons (I'm sure they're gearing up for it with Tesla Insurance).
As we see in the announcement, Drive Pilot doesn't even perform automated lane changes yet, which even Ford's BlueCruise system can do.
Assuming liability for malfunction is the stellar opposite of "gimmick." It's an actual SLA on a GA feature rather than pre-alpha toy tech that a certain despicable company that won't be named hoists on all public roadway users in the vicinity of their garbage vehicles.
I doubt your liability is waved when reading a book while using the adaptive cruise control in your car.
“Like related Level 2 driver-assistance systems, Mercedes' Level 3 technology handles driving speed, following distance, and steering. In the meantime, it keeps tabs on a preset route and traffic signs and signals. The system can also brake and/or maneuver to avoid potential situations.”
So, it reacts to traffic signs, and can change lanes to pass slower cars.
Also, only certified in Nevada currently (but California to follow soon). And seems like available only in S class or EQS class.
Basically, for now benefits chauffeurs of expensive limo companies ferrying customers along the Las Vegas strip (admittedly an over simplification for the purpose of humour)
Source: https://www.heise.de/news/Hochautomatisiertes-Fahren-bis-130...
Cannabis legalization and everything else: "We need to study the potential effects for 20 years to be absolutely sure it's safe."
Allowing Mercedes let cars drive 130km/h on their own: "Seems about fine, signed."
Not complaining about the lack of bureaucracy in this case but it is crazy how fast they got the initial law change passed and now they've even updated the speed to 130, even though no car is capable of qualifying yet.
Apparently said by former chancellor Gerhard Schröder.
https://gutezitate.com/zitat/152677
Also https://www.welt.de/wirtschaft/article167476785/Ohne-Volkswa... (use translation services, damn it!)
Furthermore https://www.tagesschau.de/wirtschaft/vw-gesetz-faq-101.html (see above)
I guess similar applies to Mercedes, BMW in their respective states.
The first part of the article mentions the driver being able to take a meeting or watch a movie. That’s complete horseshit if they need to be ready to take control of the car at all times.
Source: https://www.rambus.com/blogs/driving-automation-levels/#leve...
I created a mySAE account to download the full spec. What I found is that the manufacturer generally has fairly free reign to define their "operational design domain (ODD)" for level 3 systems, aka what conditions the car's "level 3" operation is designed for and can operate under. The only real requirement for level 3 is that OEDR is required (monitoring the driving environment (detecting, recognizing, and classifying objects and events and preparing to respond as needed) and executing an appropriate response to such objects and events (i.e., as needed to complete the DDT and/or DDT fallback [fallback to driver]).
Props to SAE for making their spec free do download, modulo a gratuitous account.
Later, they add the nag. 2018 I think.
In that case, this paragraph from the article doesn't make much sense (in fairness, it was on pretty shaky ground to begin with):
> Mercedes’ Level 3 conditionally automated driving assistant can, on suitable highway sections [emphasis added] and where traffic density is high, offer to take over the driving, leaving the driver free to do something else, like watch a movie or participate in a meeting.
If the car is able to recognize far enough in advance that it doesn't know what to do in a situation and reduce speed, pull over, etc., then that could be enough time for a person to put down their phone and take control of the vehicle.
When I interpret "being constantly ready to immediately take over" language, it's unambiguous that I need to maintain full situational awareness as if I were driving the vehicle.
Physically and mentally transitioning from partial to full situational awareness obviously takes some time, and yet from the graphic in the SAE link, the chart remains entirely ambiguous about the permitted time between "when the feature requests" and when "you must drive".
Furthermore, a quick review of SAE J3016 Rev 2021-04 reflects equally ambiguous language in § 5.4 NOTE 4 (my emphasis):
> NOTE 4: In the event of a DDT performance-relevant system failure in a Level 3 ADS, or in the event that the ADS exits its ODD, the ADS will issue a request to intervene within sufficient time for the fallback-ready user (whether in-vehicle or remote) to respond appropriately.
I'm curious what the normative minimum of "sufficient time" will be in practice.
The way I’m reading the SAE, it’s closer to the second. Like, it’s actual real self driving, but might need the user to take back control with a modicum of warning if it sees, say, a construction zone ahead or if it detects too much standing water on the road.
This is marginal. Volvo was trying for a level 3 system where, if the driver did nothing, the vehicle ended up safely stopped, preferably out of traffic. That's the gold standard for level 3.
"Under German law, to be allowed to operate at Level 3 above 37 mph, vehicles will have to be able to autonomously reach a 'safe haven' such as a breakdown lane in case of an emergency."[2] Mercedes is not there yet. They claim 10-second handoff, but they really want the human back on line in 3 seconds.[3] Look away from the road for 5 seconds, and the system starts complaining.
[1] https://www.sciencedirect.com/science/article/abs/pii/S00014...
[2] https://www.motortrend.com/reviews/mercedes-benz-drive-pilot...
[3] https://mercedes-world.com/news/mercedes-benz-level-3-drive-...
> You’re allowed to take your hands off the steering wheel, to take your eyes off the road, and even to turn your head to the side. But you can’t take a nap. > > As I discovered in my short closed-course test-drive, some flexibility is built into the system. For instance, as long as you’re belted, it’s OK to look back to a child in the back seat for three to five seconds. > > You can push it too far, though. After consulting with the engineer in the passenger seat, I closed my eyes completely, and just eight or nine seconds later a prompt popped up asking me to confirm I was still alert. I ignored it, which soon started the 10-second countdown toward disengagement.
It's also only available on certain stretches of the Autobahn. If the system allows you to enable it, you can use it. The German law allows autonomous driving up to 130 km/h although Mercedes currently only allows their system to work up to 60 km/h.
This has to be the standard for all but the most egregious examples of other-driver-at-fault crashes. That, together with marketing standards, e.g. you can say FSD but you have to also disclose Level 2, should help clean things up while incentivizing good development.
But companies will offer this only if the downside risk is bearable. Which, (IANAL) is quite certainly much bigger in the US than in Germany.
Is the liability in Germany uncapped?
That's weird; usually laws are behind the tech, not ahead of it.
Mean speed is somewhere at ~110km/h (and you're on the slower end in my experience if you go 110 km/h).
I think the idea here is that typical executive meetings ( which of course Mercedes owners would be) are monotonous and boring and it would not take more than 20% of your attention in any case.
Roughly same for movies as it has to be Hollywood franchise movies which even if one is watching first time, it still feels like seeing same thing thousand times before. And now concentrating on road looks more interesting than movie.
All Mercedes has to do is put `*` on Movies/Meetings and define them appropriately to be on firm legal grounds.
No '*', Mercedes takes full responsibility and there are many seconds before it shifts control (and liability). The system can reliably predict when it will want to give back control.
I have yet to see a good explanation of this.
Maybe this is the case on every forum, simply because that is the prevailing attitude among people in general? During the evolution of mankind, fundamental shifts have probably not happened during a lifetime. So maybe our brains are wired to fight the idea of fundamental change as nonsense or betrayal?
Hate and healthy skepticism are two completely different things.
https://news.ycombinator.com/item?id=34335123
Calling the article "complete horseshit". That sounds hateful to me.
Objectively speaking, the article IS complete horseshit, if it resorts to suggesting "watching a movie" or "participating in a meeting".
Which has got nothing to do with the technology itself, though.
Watching videos is one of the top things I would probably do while level 3 is engaged, and the writing of that sentence seems fine to me.
Even in cars, it's not a new technology. But it's very easy to claim something without backing it up. Which is exactly what tesla has been doing with this "new technology" for the last 10 years. People are cautious about hearing news about self-driving, because in the past there have been some pretty major caveats.
This was certainly not possbile 10 years ago:
https://www.youtube.com/watch?v=5X9XyHHTTQI
How long back in time can we go and still find a Youtube video of a car driving itself as well as this one?
The video you posted is two weeks old.
But I believe the video with the blind guy does have a long form version somewhere, so if you consider that (different vehicle, much older technology) to be "as well as they do now" in the sense you meant then that's still roughly in the six year mark.
xracy talked about autonomous vehicles which will include at least trains. London's Victoria Line underground service build in the 1960s was always ATO, which is closer to L4 in this scheme than L3, under normal operation the original Victoria Line trains basically require the driver to push a button, closing the doors and then once the train concludes all of its doors are safely closed it will leave, drive to the next station, stop and open the doors, then we just rinse and repeat.
Now, trains don't actually use the car terminology, in trains we'd call this GoA2, Grade of Automation 2, with the next steps being GoA3 in which there's a human who knows how to operate the train manually in principle, and they are on the train, but they aren't up front "driving it" unless something went wrong; and then GoA4 in which the train doesn't have anything resembling a human driver it just has passengers, in an emergency some central authority will set out to rescue the passengers from their disabled train. There are a handful of GoA4 train systems, but they're probably only 5-10 years old.
But it's funnier if I don't.
case in point: https://news.ycombinator.com/item?id=34347778
To be fair I don't think this particular "L3" is groundbreaking but what I said still applies.
I try to push back against this attitude by leaving positive comments to the effect of "Hey, this is cool!" for cool things, but it's an eyedropper against the ocean sort of situation.
This may be more a result of news outlets exaggerating claims, than of actual creative entities themselves overpromising.
"We" built a free, open, internet. Which brought the twentiers mostly surveillance capitalism by a few monopolists. "We" built a system in which companies that seed division can generate trillions of revenue from that. Twentiers (tweens?) rightfully are sceptical about the things I am sceptical about today, but which I rooted for (and spent my energy on) in my twenties.
More like the last 200 years will be like the last 200 years. When has there been some truly "giant shift" that swept the world in less than a decade?
The automobile we know today was an outgrowth of the decades of developments of steam engines to farm tractors to passenger vehicles and was only fully enabled once we built the interstate system and has required decades of further development particularly of safety systems to bring us to the comfortable place we are now.
I imagine this technology will eventually be useful, but so many more developments and earned experience needs to be gained before we can build out the rest of the infrastructure that will truly allow this technology to become a ubiquitous improvement over our current state.
It's maybe not that we're negative on the new technology, it's just that we'd like to acknowledge this gap and concentrate on the issues that it creates rather than getting lost in the perennial shine of the new.
When has there been some truly "giant shift"
Cars replaced horses.
Email replaced letters.
Phones replaced desktops.
Phones replaced cameras.
Websites replaced newspapers, TV and radio.The web created a searchable, worldwide instant knowledge base in which mankind shares what it knows.
Mobile makes everyone connected to everybody else, where ever they are.
For all of these examples, the time going from "most people are aware of it" to "most people use it" was less than 20 years.
Example: Most people in developed countries started to become aware of email in the 90s. In the 2010s most people had an email. Probably even in the 2000s already.
Therefore, if you took uber away, they would be worse off, in their own estimation.
You can see significantly more negativity for non-SV tech, especially when they do something outside of the narrative. If it’s an electric car, should be batteries and not fuel cells for example.
This is also one of the reasons why if a company wants to go big needs to move to SV. It looks much better when things happen the correct way, you don’t have to do all that convincing.
I couldn't find the video I was looking for but this is another cool one from the 80s: https://www.youtube.com/watch?v=_HbVWm7wdmE
Anyway, the way I see it is, to get positive reaction you need to drive and follow the narrative. That is, for example, providing arguments about how it should be done as if it is a public debate and then reveal your tech. Then you will have fanboys and and hateboys and people will feel involved and if you do it in SV your ability to do what you say you do would not be questioned as much because magic happens in SV.
You say that your car is basically a computer o wheels and the story gets traction, when a car company makes a self driving car it doesn't fit and will need to do much more convincing because the techies will be digging in to find faults. The computer on wheels might cause crashes but the techies will be much more forgiving and look for excuses.
I feel quite vindicated by this, because I've always been (perhaps irrationally) rather suspicious of Tesla. Software company suddenly decides to make 'luxury cars'? Cool story bro.
My opinion is that the German triad (BMW, VW (and subsidiaries, including Porsche), Merc-Benz) may have lost some ground to Tesla when it comes to EVs and self-driving cars, but they have an immense mind- and market-share and will quickly recover lost ground. They have a pretty good engineering-first mentality, too.
Plus, they have always built generally good cars (late 1990s-early 2000s unreliability notwithstanding).
In fact, this is not surprising at all. Audi claimed in 2017 that its A8 was capable of Level 3 self-driving[1], but the system was subsequently downgraded. In the intervening years, the Germans' research into self-driving cars has drastically expanded.
[1]: https://www.audi-mediacenter.com/en/on-autopilot-into-the-fu...
Americans, SV and HN especially, are also oblivious to this.
The current VW CEO was just sent packing.
Why?
Software quality.
They know it's do or die and the level of investment into it is huge, they've been hiring software devs like crazy, in Germany.
But, like, that's after they were making EVs. The roadster was announced in 2006, and started production in 2008 - 15 years ago.
They went from making EVs, which are cars with some software inherently in them, to making different cars with even more software in them. They're a car company first, who have been increasing the amount of software they've written to put in their cars.
The comment I was replying to was suggesting they were a software company first, which then pivoted to making cars. Which is... just... what?!?
Yes, you technically answered my question, so thanks for taking the time and making the effort and whatever, but just not in a way that is actually helpful in the context that I asked it.
- "Self-driving cars are a suspect and dangerous technology which is unlikely to perform as well as the companies promoting it claim, and may not become broadly useful for decades, if ever" (I broadly fall into this camp) - These people don't like it because it's a 'self-driving' car, FSVO self-driving.
- "Self-driving cars are coming any day now, and will be a transition straight to robo-taxis lead by visionary super-genius Elon Musk" (This has probably become less common as the promises from that camp haven't materialised over the last decade) - These people don't like it because it is insufficiently robo-taxi-y or musky.
There's possibly an argument that the first camp has become too cynical in response to the second, but as it stands there just isn't that much mindshare for this sort of cautious progress.
1. entrepreneur sees an opportunity that no one else sees
2. entrepreneur takes some big risks to take advantage of the opportunity
3. entrepreneur becomes wealthy because they took risks to take advantage of the opportunity
4. other entrepreneurs (or businesses) become interested now that the new opportunity has been proven
5. after 10 or 20 years of hard work, the original entrepreneur wants to shift their attention elsewhere, maybe relax, at the very moment that outside money is pouring into this opportunity
6. original entrepreneur goes bankrupt, as their attention was shifting away from their business just as competition was intensifying.
On a much larger scale, it seems Elon Musk is at risk of exactly this pattern. His attention shifted away from Tesla just as Tesla is facing increasing competition.
Musk's businesses seems so financially diversified that one bombing overnight could just take some cash from the others. He's already leveraged Tesla stock to buy a massive Social Media platform, right? But the profitability of Twitter is certainly another point to be decided.
I'm a Video Gamer, not a Bio-Electrician Researcher aware of the past 3 Decades of Cutting Edge Technologies.
I'm baffled as to how you can say "fyi it's fundamentally broken" without giving more detail.
I agree the work they've done is kind of exciting, but they didn't solve one of the crucial issues that's been blocking the field and technology from flourishing, implantable electrodes have been around for a long time.
Searching 'neuralink "scar"' ('' removed) got me information about the ethical issues of Neuralink@UC Davis, and a 2020 article about potential polymers to use in the wires.
I will continue to wait and see how the tech develops - I certainly was never planning on being one of the first early adopters anyways, but I still see a lot of potential for Paraplegics, and then obviously Consumer usage eventually.
Tell me Elon's the savvy businessman you claim him to be. If Gates were running the company, he would have acquired half the auto industry by now.
Tesla doesn't want to support the supply chain of new and used parts for a legacy car company, they can barely keep up with their own (and they only make 4 models!).
owners and CEOs are powerful but not infallible; Carlos Ghosn got thrown in jail when he got too big for his britches as CEO of Nissan (in Japan Inc.'s view, at least).
Let's drop the BS, please.
Mercedes is a car company. They are not legacy anything. They're the inventor of the car and a myriad other innovations.
They're making a ton of EVs and will sell a ton more next year and the year after that.
They have AR navigation and now are close to L3 self driving.
I’m saying all this to draw a comparison against Tesla. Tesla doesn’t want a dealer network, or a parts counter, or a typical car company valuation. Buying a legacy car company is not in Tesla’s interest, because they don’t want to play by the rules (unwritten or written) that every other auto manufacturer plays by.
99% of the time I see "legacy" used for established car companies is when it's an article or comment about Tesla.
And it is meant as a pejorative.
Maybe you're the exception:-)
I love this. I haven’t heard it put this way before. I’m tempted to agree with you. Gates was one of the most aggressive, cutthroat, successful empire builders of this century.
https://ir.tesla.com/press-release/strategic-partnership-dai...
Tesla could dilute their stock and sell more, but that is quite the dilution. Further, the sell price might not be just the market cap value, particularly for a profitable company. Though, exactly those kind of take overs do happen (and with tesla's numbers, not too hard to imagine)
Market cap has little to do with actual size of a company (employees, manufacturing capacity, ...), nor can a company with a larger market cap automatically buy a smaller one. Especially the latter completely ignores how markets work.
These mistakes are particularly egregious when you bring meme stocks like Tesla into it.
Revenues: Tesla $75B. Mercedes $144B.
Op Income: Tesla $12B. Mercedes $18B.
Assets: Tesla $62B. Mercedes $260B.
But according to this guy, "it could happen tomorrow".
https://gizmodo.com/feds-investigate-elon-musk-tweet-tesla-s...
The competition thought the claims were real and actually made it happen.
It is somewhat predictable that eventually responses would in particular start to focus on the elephant in the room.. fundamentally a parent comment should be about the article
More than zero hands should be on the wheel and your eyes should be on the road.
If you can't be bothered to do this, perhaps find another mode of transport.
If you are responsible for two tons of metal flying down the road at 70 mph you need to be fully engaged in that task.
We can handle being passengers and we can handle being drivers but we can't switch in a split second.
So my contention is that level-X where X is less than "I can fall asleep in the back seat" is a deadly valley between unassisted driving and passive passenger.
The Mercedes system has a top speed of 40 mph.
> We can handle being passengers and we can handle being drivers but we can't switch in a split second.
That's why there is a 10 second handover. If the human fails to take control of the vehicle during the handover, the vehicle will slow to a stop.
That's up to 130 km/h starting 2023
A few examples of activities that could be hard to switch away from in time:
* Eating breakfast with food in my lap and the spoon halfway to my mouth.
* Typing an email.
* Reading a book.
Some of these could allow you to physically have your hands on the wheel in time, but then you still need to context switch your brain and then get up to date on the road.
That's the thing - you're not. Once engaged, Mercedes takes liability, so you're no longer responsible.
Legally (though surely to be tested in court).
Morally though, I'd say the driver is still responsible if they kill someone. It doesn't feel right to say "Oh well, it's Mercedes' fault, shrug" after your car with you in it drives over someone else.
I didn't say you were legally responsible. You are morally responsible.
Like IBM said back in the 70s; A computer can never be held accountable, therefore a computer must never make a management decision.
Well you're letting it make decisions for you in a 2 ton murder weapon.
I feel less safe knowing there will be humans on the roads letting go of their steering wheels and handing over control to an imperfect system.
Some day I'm sure we'll have automated traffic, but the road there will be lined with bodies.