Philosophical question maybe , but is there any practical difference? I'm not being glib but if the human is taking a nap behind the wheel then what use are they?
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Germany to Allow Lv4 autonomy
- Thread starter Edders
- Start date
Philosophical question maybe , but is there any practical difference? I'm not being glib but if the human is taking a nap behind the wheel then what use are they?
Level 4 is "some conditions" - basicaly geofenced, and level 5 is everywhere in all conditions.
Automated Vehicles for Safety
Cardo
Active Member
If the car got stuck, it would come to a safe stop and inform the human it needs to take over.
There’s also the geofencing aspect, as others have mentioned.
If you read different pieces on this, it appears the levels aren’t fully defined, however.
Its often taken as geofenced but it could be much more specific than that, for instance it could be motorways in daylight and not raining. The key difference between Level 3 and 4 as I read it is Level 3 may need to hand back control to the driver at any point but is otherwise driving, whereas level 4 is you have a much more deterministic view on when the driver is required which allows for minds off driving. Level 3 is really a horrendous middle ground unless car manufacturers can come up with a way to keep drivers ready to intervene should the car decide its had enough and wants to fail over. 70mph in the fast lane approaching stand still traffic would not be a good time for the car to bail out and say "over to you sunshine" if you're half way through a crossword. Nor would it be clever for it to stop on a busy A road and wait for you to take over the controls.
The legislation will be really limited in execution. Tesla is not ready for anything above level 2 at the moment and as good as some of the new beta clips appear to be, I've seen far too many "it was ok except for that bit" type comments.
I've thought for a long time level 4 on motorways would be a massive win for tesla and to forget about the urban driving as a priority. I recon a lot of people who drive up and down motorways would be content to do the first and final mile of their journey and let the car do the long haul on the motorway allowing them to have a kip, do some emails, watch TV etc. I used to do 30k miles a year and thats low by some standards but it would have been a major benefit to me. I see virtually no benefit in the urban side unless it can legally bring me back from a pub. Level 3 and having to pay attention is likely to cause more paranoid passenger wondering if its seen something or frustration when every school kid knows they can walk out in front of a Tesla and it will stop.
I don't think the levels of autonomy numbering framework is very well considered, nor the L5 definition even sensible - at least not if interpreted literally. Few, if any, humans can drive absolutely safely in 'any and all situations or road or weather conditions'. I know I can't. Last two days for example, neither human nor robot was going to be driving through the snow and ice choked local roads where I live. Does 'L5' distinguish in its definition the limitations of 'autonomy' as opposed to the physical limitations of moving a car along a road that has essentially disappeared under snow or flood?
In my opinion the big issue for legislation about autonomous car technology is when can it be shown statistically to be demonstrably safer, overall, than the typical human driver, across a range of driving conditions and scenarios?
I really enjoy driving my M3 and like its performance and don't want to rush to use a Robotaxi all of the time. On the other hand, I have experienced drives on dark winter nights with the rain lashing down when I have already found using FSD absolutely brilliant - despite current limitations; its ability to keep safe distancing and automatically lane change to pass large lorries throwing up blinding amounts of spray were really helpful to me and in my opinion a great safety feature. I am very impressed with the capabilities I have seen now in the US on many drives shown on YouTube using the evolving 'FSD beta rewrite'. Still by no means perfect, but many, many human drivers I see every time I take the car out are by no means perfect either. I think fact based demonstration of overall improved safety levels should be what is used to determine if and when to allow autonomous features in cars. rather than demanding unobtainable 'perfection' under all conceivable situations. The fewer humans driving while not concentrating, or talking into their mobile phone, or when they are drunk or tired or otherwise distracted, the better for everyone else using the roads.
A lot of the levels have some obvious issues
L2 has the issue where a human can't possibly be expected to oversee a vehicle that does what the FSD beta does. Every day I'm expecting to read about some accident happening with an FSD beta driver who didn't intervene in time.
L3 is only really practical as a traffic assist package where its limited in speed. If its used at highway speeds the driver could easily nod off, and they won't be able to respond during the hand off.
L4 I like because it makes the car responsible for the drive while allowing restrictions to be placed. Restrictions can come from anything from where the manufacture of the vehicle allows it to drive, the road conditions, or the region accepting self driving cars. It gives room to grow.
L5 has a problem where it needs to distinguish itself from L4, but has to set limits towards expectation. So it uses drivable conditions, but who says what's a drivable condition? Humans drive in adverse conditions all the time, but in doing so we lower the safety threshold. Obviously lowering safety threshold isn't something an autonomous car can do if the expectations are 10x better safety than a human.
My expectation is L4 will be the predominant one, and all the others will fall by the wayside. Eventually L4 will get so good in terms of areas, and conditions that most people won't think about it. At that point it will be effectively L5.
Volvo intend to skip level 3 completely as they don’t consider it safe (here’ their media statement on the subject Volvo Cars CEO urges governments and car industry to share safety-related traffic data). Given how some people already treat autopilot as full self driving despite the warnings with and tragic consequences in a small number of cases, I tend to agree with them.
On the recent earnings call Elon mentioned the FSD subscription would start within 2 months if everything goes well.
Pretty much everything changes if/when Tesla starts the FSD subscription model because it has to do something substantial to get people to pay any amount close to what $10K over a standard car loan would cost.
When/if the FSD subscription model starts is anyone's guess.
I don't expect Tesla to give up on FSD. Instead I expect them to push the limits, and then get held back by regulatory. They'll blame the regulatory for failing to deliver, and then they'll iterate the design with a fresh new promise. When they hit a wall with that they'll do the same thing all over again. Each time we'll get just a bit closer.
They'll lose frustrated FSD customers like me to Rivian/Audi/VW/Porsche, but there are plenty of new customers to take our place.
Ford has also said something similar before. I think they softened the stance a little but I can’t level 3 being insurable given its pitfalls
Do those companies do better driver assist? I can see other advantages, price, quality etc.
TBH the only car i'd consider changing to is a Porsche. Purely for the fancy cabin.
I'd never consider any of them as being better at driver assist.
Edit: Rivian? Only if you want to be an Amazon delivery driver I suppose.
I've a reasonable amount of experience of the latest BMW driver assist (which is mobileye) and the Tesla system and as a driver assist (assist being the important word) the BMW system is far better to engage with as a driver. It also has many of the important bits you need including lane change (but you have to indicate and it has similar restrictions on which road types), but it can also see traffic lights and stop signs. Of course the Tesla system promises much more, but the frustration I find with the Tesla system is the lack of communication on whats happening. As an example the BMW pops up on the head up display if its going to reduce speed ahead due to a sharp corner and then counts you down as you approach, it tells you if the speed limit is going to change and counts you down as you approach it, same with a roundabout. Traffic lights are more temperamental but because you can see if the car has picked it up from the head up display you know whether you need to take over. I've spent too much time in the Tesla thinking "is it or isn't it going to stop?" and so you can end up playing chicken before you step in and take over in some situations.
The other thing I prefer about the BMW system is lets say you're on AP as you approach a roundabout, it will slow down but keep going unless you brake. If you do brake, then manually drive round the roundabout, once established back on the road (so 1 or 2 seconds) it will just signify its taken over again, no need to reengage. You can do the same with an overtake, drive past, pull back over, system picks up again.
The downside is I don't see that system ever getting any better whereas the Tesla system should with, in theory, no limit to how good it can get until its picking you up from the pub if you believe Musk. Its as if the Mobileye system is built with the intention of nailing driver assist, and the Tesla is built as FSD and assumes the driver doesn't need to be kept informed and should just need to turn it on.
Thank you for the informative post. I certainly didnt try the BMW system, being mobileye I presume that is going to be in more than one make of car.
I think this really boils down to FSD beta vs everything else. There is no argument that the current AP (even with the full FSD package) is sub-par compared to Mobileye - at least no argument from me.
And that's essentially the problem, people make decisions based on past performance, but buy a car for 3+ years. If I were to do that I'd not buy for driver assist at all, because it was never "good enough".
So if we were to extrapolate future performance for a product you'll use for 3+ years... The question becomes, will the Mobileye system in that car update during those years? And/Or are you happy with it staying the same for the entire use case duration?
Even going back to past performance, my Telsa AP has improved in a number of ways since purchase (1.2 yr ago). It can now take a turn (slowing down) within regulations. There are fewer errors (phantom braking etc.).
That conclusion appears to be very much against the grain of most of the anecdotes published here.
Ha, No.
A Rivian is for people who want a truly off-road capable SUV. If you want to see the Rivian in action I highly recommend watching a Long Way Up on AppleTV+.
For driver assist a lot of things simply come down to trade offs.
For me I don't mind trading something like auto-lane change for proper blind spot monitoring which pretty much any modern car has these days. I don't mind trading road visualization for night vision assist like Audi E-Tron GT has (the GT is based on the Porsche Taycan). I don't mind trading auto-park for down facing cameras.
The most important thing to me is that I can count on something.
In my 6+ years experience with Tesla vehicles the only time an ADAS system worked 99.99% of the time or more was with Adaptive cruise control on AP1 provided by Mobileye.
With that being said I would love a Supercruise system L2 system. I don't care if its geofenced.
The problem with this approach is you end up creating a car with a lot of misses.
The Model 3 is especially hit hard:
No driver monitoring so it means you have to apply torque to the steering wheel
No blind spot indicator in the mirrors so it effectively means no driver assist blind spot monitoring.
No rain sensor because the AI computer doesn't need it as it can hit the wipers when it needs to
No physical wiper control (for wiper speed)
No downfacing 360 degree camera as apparently that would have been too redundant for Musk
The Model 3 has gone its entire life optimized for something they have yet to achieve with it. The vehicle is a huge sales success not because of it, but despite it. The recent success of European competitors in the Europe is likely at least partially related to the misses.
The other issue with promising things in the future is you end up prolonging the time that the person owns the car basically trapping them in the car while they wait for the features to be released.
Like I'm sidelined waiting for FSD.
Some aspects of the new refreshed Model S would be nice, but there isn't much point looking into that much because Tesla has already promised me the world with FSD.
Last edited:
Mr H
Active Member
Wol747
Active Member
>>Ford has also said something similar before. I think they softened the stance a little but I can’t level 3 being insurable given its pitfalls<<
I think it's more than level 3 - even more so for 4 or 5.
Insurers will be well within their rights refusing to insure the car when it's driving itself legally under 4 and 5. After all, the driver can't be expected to be responsible for an incident caused by his car doing something that can't be addressed in the time available. And numerous studies have shown what is obvious - that one's reactions are considerably slower when monitoring than when in control.
It should be up to the car manufacturer to cover the insurance when their car is doing its thing. Legal minefield here!
From the hundreds of videos on YouTube of trips with the Beta it appears that there are still plenty of interventions required plus, more importantly perhaps, a fair number of sixpence/halfcrown close shaves where the driver has allowed the car to get closer than he is comfortable with but there's no actual collision.
I think it's more than level 3 - even more so for 4 or 5.
Insurers will be well within their rights refusing to insure the car when it's driving itself legally under 4 and 5. After all, the driver can't be expected to be responsible for an incident caused by his car doing something that can't be addressed in the time available. And numerous studies have shown what is obvious - that one's reactions are considerably slower when monitoring than when in control.
It should be up to the car manufacturer to cover the insurance when their car is doing its thing. Legal minefield here!
From the hundreds of videos on YouTube of trips with the Beta it appears that there are still plenty of interventions required plus, more importantly perhaps, a fair number of sixpence/halfcrown close shaves where the driver has allowed the car to get closer than he is comfortable with but there's no actual collision.
What can I say.... My anecdotes are against the grain? Over the one and a half years of owenership and 14k miles I've had about 5 phantom brake episodes. Last one was in September. Back in September it also didn't slow down for corners that it could not take within regulations, now it does.
Not saying it's better that Mobileye, but once you buy the car Mobileye wont get better. The very idea of BMW/MB/VW releasing monthly software updates is hilarious. Last "software update" I had from VW is to reduce engine power due to them cheating on emmisions and the car had to come in to have it done.
So to sum it up, as a consumer buying a car today, you got much better chance of experiencing Level 3+ autonomy with a Tesla.
I think most people here are sour becuase they feel that this option was "miss sold".
Similar threads
