-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Autonomous Car Progress
- Thread starter stopcrazypp
- Start date
-
- Tags
- Autonomous Vehicles
I just read that Benz got level 3 approval. Can my model 2021 model S get there? It says you can legally watch movies and not touch the steering wheel or require eyes on the road (40 mph limit though). Is this a big milestone that Tesla missed or hype? I’m not up on the FSD stuff. Thx.
www.bloomberg.com
Mercedes-Benz Beat Tesla to Cheap, Hands-Free Driving: Test Drive
We tested Mercedes’s new Level 3 autonomous driving system on LA’s snarled interstate.
www.bloomberg.com
KArnold
Active Member
Likely both. Mercedes is good weather only, no lane change, no nav. Tesla does better on limited access highways, but with nags.
And I'm not sure Elon is even interested in level 3, let alone such a crippled version of it. This by itself may not generate many new sales, even if it were available in more than 2 states. Market is very limited - need to be a commuter in a handful of markets to gain much advantage.
On rhe other hand, MB can say they were first un this niche - just not sure if the value of that.
willow_hiller
Well-Known Member
I just read that Benz got level 3 approval. Can my model 2021 model S get there? It says you can legally watch movies and not touch the steering wheel or require eyes on the road (40 mph limit though). Is this a big milestone that Tesla missed or hype? I’m not up on the FSD stuff. Thx.
Mercedes-Benz Beat Tesla to Cheap, Hands-Free Driving: Test Drive
We tested Mercedes’s new Level 3 autonomous driving system on LA’s snarled interstate.
It's so limited, I think it would be difficult to actually shift your attention away from driving. Edmunds broke down all the ODD restrictions:
From the full review:
Tam
Well-Known Member
No. I don't think Tesla will ever pay for damages while using its system.
This is a big improvement.
Hands Off is a reality with GM Cruise since 2017 and recently with others and Ford Blue Cruise.
Eyes Off is a reality with commercial Waymo but this is the first time that feature is transferred to the consumer product.
Of course the Consumer Eye Off has many limitations but this is the first time that there's zero accidents during its operation in Europe and Mercedes is willing to pay for damages while using its system.
Yes, it's only limited to 40 MPH but Tesla wouldn't pay damages for low speed collisions of a few miles per hour in Summon/Smart Sumon. So that's a big deal.
Since it is relatively new transition of commercial Eyes Off to consumer Eyes Off, Mercedes expects to double the speed when there will be enough acceptable data experience.
Autonomous driving: Mercedes is working on Level 3 implementation for 130 km/h
Mercedes-Benz is working on an implementation of the new UN regulation for the Drive Pilot, which allows autonomous driving at 130 km/h.
basic-tutorials.com
There's too much emphasis on L3 and people tend to forget that L2 is not missing just because you add on L3.
It's true that its L3 cannot do Auto Lane change but if you want Auto Lane Change and On-Ramp to Off-Ramp Auto Navigation, you can turn off L3 and use its L2.
Its L2 is Hands On, Eyes On but the steering wheel has a capacitive sensor, so as long as there's human contact with the steering wheel, there will be no nags. No jiggling, no counter-torque needed
Remember, these L2 and L3 are not beta. They have been tested in-house and sell those features to consumers once they are competent enough.
For more features, you can use L2 but Mercedes won't pay for damages. For less features and more limitations, you can use L3 but Mercedes will pay for damages.
Yes, it's new transition from commercial Eyes Off to Consumer Eyes Off with many limitations but the speed limit will be increased beyond 40 MPH in the future. And if you want the machine to drive 111mph, you can use its L2 (Tesla's system is limited to 85 radarless or 90 MPH radar).
Last edited:
diplomat33
Average guy who loves autonomous vehicles
diplomat33
Average guy who loves autonomous vehicles
I see a lot of folks complaining that the ODD for Drive Pilot is too small. True. But I think we need to look at 2 axes. One axis is the ODD. The bigger the ODD, the more useful the automated driving system will be but the higher the risk as well since a bigger ODD means more chance of edge cases and situations that can cause accidents. So companies can limit the ODD as a way of limiting the risk and making the system safer. That is what Mercedes is doing with Drive Pilot. And we see that with robotaxi companies as well like Waymo and Cruise. They geofence as a way of reducing risk by keeping the robotaxis confined to areas where they have tested the system a lot. Bigger ODD will also be harder thoroughly test so it is easier to remove driver supervision in a smaller ODD like Mercedes as done. So Mercedes Drive Pilot has a small ODD but it also allows for less supervision. And I do think Mercedes will expand the ODD over time. Tesla's approach has been to go with a big ODD first but to keep driver supervision.
The other axis is the amount of driver supervision required. Driver supervision can go from full hands-on/eyes-on (highest supervision) all the way to driverless (no supervision at all). As you reduce driver supervision, the risk goes up as there is of a guardrail if the system messes up. So the system needs to be able to handle more cases safely on its own. Supervision is related to safety. The safer the automated driving system is, the less supervision it will require.
Of course, the ultimate goal is to have an automated driving system that has both the biggest ODD (everywhere) and is also safe and driverless. But that would involve huge risk and liability. And it is very difficult to fully validate safety "everywhere". Nobody has achieve safe driverless everywhere. So right now, companies will compromise in one axis or the other. They might reduce the ODD to reduce risk but allow for less or no supervision (Mercedes, Waymo). Or they make the ODD bigger but increase supervision (Tesla).
Since there are two axes, there are two paths you can choose to get to driverless everywhere. You can go down one axis or the other. So you can start with the big ODD first with a lot of supervision and then try to make the system safer and more reliable (reducing supervision as the system gets safer). That's the Tesla approach. Or you can start with a small ODD first, validate safety more quickly, remove driver supervision, and then try to expand the ODD while keeping your system safe enough. That's the Waymo approach. In terms of validation, both approaches are fine. I think it becomes a matter of personal opinion which you prefer. Some will prefer the bigger ODD first because it is more useful. The consumer car approach will likely go that route since consumers want a bigger ODD when driving in their personal car. The robotaxi approach prefers to go driverless first in a "small" ODD since they want to emphasize the rider being a passenger. As a robotaxi, they don't want the rider to have to supervise the car, it would defeat the whole purpose of a robotaxi. And once they have a robotaxi service up and running, they can expand the ODD later.
The other axis is the amount of driver supervision required. Driver supervision can go from full hands-on/eyes-on (highest supervision) all the way to driverless (no supervision at all). As you reduce driver supervision, the risk goes up as there is of a guardrail if the system messes up. So the system needs to be able to handle more cases safely on its own. Supervision is related to safety. The safer the automated driving system is, the less supervision it will require.
Of course, the ultimate goal is to have an automated driving system that has both the biggest ODD (everywhere) and is also safe and driverless. But that would involve huge risk and liability. And it is very difficult to fully validate safety "everywhere". Nobody has achieve safe driverless everywhere. So right now, companies will compromise in one axis or the other. They might reduce the ODD to reduce risk but allow for less or no supervision (Mercedes, Waymo). Or they make the ODD bigger but increase supervision (Tesla).
Since there are two axes, there are two paths you can choose to get to driverless everywhere. You can go down one axis or the other. So you can start with the big ODD first with a lot of supervision and then try to make the system safer and more reliable (reducing supervision as the system gets safer). That's the Tesla approach. Or you can start with a small ODD first, validate safety more quickly, remove driver supervision, and then try to expand the ODD while keeping your system safe enough. That's the Waymo approach. In terms of validation, both approaches are fine. I think it becomes a matter of personal opinion which you prefer. Some will prefer the bigger ODD first because it is more useful. The consumer car approach will likely go that route since consumers want a bigger ODD when driving in their personal car. The robotaxi approach prefers to go driverless first in a "small" ODD since they want to emphasize the rider being a passenger. As a robotaxi, they don't want the rider to have to supervise the car, it would defeat the whole purpose of a robotaxi. And once they have a robotaxi service up and running, they can expand the ODD later.
Last edited:
willow_hiller
Well-Known Member
It's also worth considering that not all ODD are created equal in terms of offering a useful experience for the occupants.
ODD that are predictable and don't change at a moment's notice allow an occupant to plan for when they might need to become a driver. A geofence is limiting, but you're not likely to find your vehicle unexpectedly outside of the geofence (unless it's shaped like a gerrymandered political district). Same thing with only day-time ODD, or only night-time ODD. It's restrictive, but not likely to oscillate unexpectedly.
Meanwhile, ODD like weather conditions, travel speed, lead vehicles, and emergency vehicle presence are likely to change rapidly, and not allow an occupant to fully disconnect from the driving task. The fact that Mercedes Drive Pilot still detects your head position and expects you upright in a driving position is evidence of this.
Drive Pilot may be a L3 system, but those fickle ODD criteria plus the very limited 10 second takeover time mean that in practice, most people will end up not being able to disconnect from the driving task as they would like, and would have to pay just as much attention to the driving task as they would for an L2 system.
diplomat33
Average guy who loves autonomous vehicles
Very correct. Thanks for adding that.
Agreed. And that is why Mobileye considers the Mercedes L3 ODD not to be useful. We see in this slide from the CES 2023 talk, that Mobileye has defined various ODDs. ODD1 (highways only up to 60 kph) is below the usefulness tipping point. Mobileye defines highways-only, 130 kph, and on-ramp to off-ramp as a useful ODD.
Doggydogworld
Active Member
This reviewer used it for 50 minutes in LA traffic without problem. Lots of people spend hundreds of hours a year stuck in highway traffic. Drive Pilot can free up those hours.
It can't make $30k/year for you as a Robotaxi while you sleep, though, as FSD has done since April 2020.
The Level 3 is a Traffic Jam Assist, outside of that it's a Level 2 ADAS like Autopilot. Traffic jams will move <40mph and involve following other cars.
What's more informing to me is the list of restrictions and wondering how long it'll be before Tesla ever takes liability in all these restricted conditions, because all of that will be required to take liability for generalized robotaxis. Imagine that risk across an entire fleet of millions of vehicles, it's likely a long ways off.
willow_hiller
Well-Known Member
What I don't understand is the 10 second handover time. If it's safe for the system to drive outside of the ODD for 10 seconds, why isn't it safe to do it for longer? Or is it expressly not safe during those 10 seconds? Will Mercedes take responsibility for accidents that happen during the handover time?
Doggydogworld
Active Member
It's not like flipping a light switch. When confidence drops from 99.9999% to 99.9998% or whatever they call for handover. They also have margins of safety - e.g. they've tested to a certain level of moisture on the road but call for handover well below that. And some conditions change slowly enough or can be detected far enough in advance that you have 10 seconds warning before you exit the ODD.
Tam
Well-Known Member
If one of the conditions is no longer met, such as the speed no longer 5, 20, 30, but it’s going to go beyond 40mph soon, the 10 second transition alarm will start.
Another example is when the day time transitions to night time. 10 seconds don't mean the driving suddenly becomes more dangerous. Human eyes wouldn't notice day/night transition in that 10 seconds.
To be fair, I think the system needs to be still responsible for that 10 second warning.
These parameters are to make sure the zero rate of accidents will favors the company.
I think once the track record is good, the limitations will gradually taken away.
In order to be level 5 cars need to be able to drive in these conditions.
That's a stretch of "all conditions". It means all conditions that humans can drive in. If the car is physically incapable of driving in a condition that wouldn't limit it from being level 5.
In reality, no car should be operating in those conditions, human or AI.
I'm on vacation in Malaysia, and I can say that none of the current AVs, nor any near future (IMO) AVs or ADAS systems will be able to drive here. The streets and motorways are so poorly planned, poorly executed, and the drivers are insane. Lane lines are just mild suggestions, and turning on your blinker is a signal for everyone in the adjacent lane to form ranks and drive literally 3 feet off the bumper of the lead car. People just start coming over and will hit you if you don't back off, causing honking. Traffic lights are usually honored by most cars, but motorcycles and mopeds usually ignore them. Streets are narrow and people just double and triple park, many times right in the middle of a travel lane, taking a 2 or 3 lane street down to 1 lane which people have to squeeze by.
Similar threads
- Article
