Will Mercedes jump to level 3 before Tesla? Looks like it.

[vapor trail]

I'm not a software engineer or AI expert by any stretch, but I have to wonder if Tesla Vision will ever deliver what we've been promised. Current iteration of autonomous driving beta is still quite wonky. Phantom braking persists, and for me that means leaving autopilot off most of the time. My original 2015 Model S behaved much more predictably and NEVER suffered phantom braking.

Personally, wish Tesla was moving more in the direction of other manufacturers, who embrace radar, Lidar, etc. Why not put all the tools available to use? My sense is that Musk is more prone to doubling-down on decisions he's made out of bullish pride rather than go back.

Mercedes Drive Pilot Beats Tesla Autopilot By Taking Legal Responsibility

Mercedes will accept full legal responsibility for the vehicle whenever Drive Pilot is active. The automaker hopes to offer the system in the U.S. by the end of 2022.

www.roadandtrack.com

I think in the coming years we're going to find Tesla still in beta mode, with us as the guinea pigs (who paid significant $$ to join), while other carmakers quietly produce more mature offerings.

 

[ItsNotAboutTheMoney]

Right now, Drive Pilot can only engage at speeds under 40 mph (60 km/h in Germany) on limited-access divided highways with no stoplights, roundabouts, or other traffic control systems, and no construction zones. Eligible roads must be mapped by Mercedes for Drive Pilot use (similar to GM SuperCruise); the automaker has already mapped every such highway in Germany, and most of those in Nevada and California. The system will only operate during daytime, in reasonably clear weather, without overhead obstructions. Inclement weather, construction zones, tunnels, and emergency vehicles will all trigger a handover warning. And no, you can't close your eyes or go to sleep while it operates.

So there are caveats.

If you find yourself stuck in the supported environments a lot, great, otherwise, not really much help and what you'd want is to focus on is finding the best, reliable Level 2 system instead.

 

[TresLA]

I know many think Musk has the reputation of doubling down for the sake of it, but “the data” doesn’t support this. He doubled down when available data doesn’t prove otherwise. When the data quantifiably shows otherwise, he changes his position.

At this point, it’s still just guessing games what sensor suite is needed to achieve L5 autonomy. Getting vision-only (which actually includes real-time maps) to work well enough is a difficult and currently unsolved problem, but so is sensor fusion (i.e. - when LiDAR, radar, HD maps, etc. and/or vision don’t agree).

Every leader in the space can show specific cases where their approach is superior to others, but as far as I can tell so far Tesla leads in the general driving everywhere. FSD beta for us has been nowhere near as comfortable and anxiety relieving as radar-included production AP (in the same exact car). It’s scary at times, and you learn with experience when approaching a situation it probably can’t handle well. But with many updates it’s pretty amazing as it improves.

Each approach has its advantages and disadvantages, and hopefully more than one approach will eventually get there (L5).

With that said, I don’t see why Musk wouldn’t change his mind if others are shipping L5, he’d go back to LiDAR/radar

 

[EVNow]

Duplicate.

Just below 60 kmph / 37 mph on a freeway. There has been a ton of discussion about this and shows perfectly why "Level 3" label is SO misleading.

Autonomous Car Progress

Because the software is supposed to be safer than a human by the end of the year? Because it will be good enough in April 2020 that you can sleep while it drives. *Checks date*..... oh, oops.

teslamotorsclub.com

 

[Daniel in SD]

Duplicate.

Just below 60 kmph / 37 mph on a freeway. There has been a ton of discussion about this and shows perfectly why "Level 3" label is SO misleading.

Autonomous Car Progress

Because the software is supposed to be safer than a human by the end of the year? Because it will be good enough in April 2020 that you can sleep while it drives. *Checks date*..... oh, oops.

teslamotorsclub.com

How is it misleading? It's literally the example everyone uses for a hypothetical L3 feature...

And the first paragraph of the article seems very easy for people to understand. Clearly you understand it, the writer of the article understands it, so why do think others being mislead? Especially odd since you seem to have no problem with how FSD is marketed even though many people feel like they have been mislead.

Mercedes' new Drive Pilot seems, in operation, like many "traffic jam assistant" technologies already on sale today. On certain highways, below 40 mph, a Drive Pilot-equipped S-Class or EQS will take control of the car's speed, steering, and brakes to move you along in traffic. But there's one key difference: Once you engage Drive Pilot, you are no longer legally liable for the car's operation until it disengages. You can look away, watch a movie, or zone out. If the car crashes while Drive Pilot is operating, that's Mercedes' problem, not yours.

 

[EVNow]

How is it misleading?

Because .... people don't understand they are not comparing apples to apples.

We have debated this a million times.

I think in the coming years we're going to find Tesla still in beta mode, with us as the guinea pigs (who paid significant $$ to join), while other carmakers quietly produce more mature offerings.

 

[Daniel in SD]

Because .... people don't understand they are not comparing apples to apples.

We have debated this a million times.

Ok. I guess we'll have to check the Mercedes forums after this has been released for a while to see if people were mislead.
Though really we'll probably have to wait until it trickles down to higher volume cars.

 

[EVEnergy]

The MB system does nothing that AP/FSD doesn't already do. The only difference is that MB is willing to take responsibility for crashes...and thereby not require the driver to pay attention. Tesla could do the same but chose not to. Given the examples of idiots misusing AP, I can't blame Tesla for not wanting to take that liability.

 

[Daniel in SD]

The MB system does nothing that AP/FSD doesn't already do. The only difference is that MB is willing to take responsibility for crashes...and thereby not require the driver to pay attention. Tesla could do the same but chose not to. Given the examples of idiots misusing AP, I can't blame Tesla for not wanting to take that liability.

What "misuse" is Mercedes taking liability for?

 

[NCC81701]

Personally, wish Tesla was moving more in the direction of other manufacturers, who embrace radar, Lidar, etc. Why not put all the tools available to use? My sense is that Musk is more prone to doubling-down on decisions he's made out of bullish pride rather than go back.

Because throwing everything under the sun at the problem is bad engineering and the end results is a system that may work technically speaking, but it's so inefficient and so unreliable that it is effectively useless. This is the equivalent of filling up the cabin of a Model X full of batteries just to say you have an SUV that has a range of 1500 miles. It works technically speaking but it's utterly useless as an SUV.

With respect to FSD. Tesla's line of thinking is that in order for FSD to truly work on any road and not just highways, the vision problem must be solved. This is because the camera needs to be able to read and interpret road signs, hand signals from a cop, interpret temporary lane markings in road construction situations, be able to distinguish between a garbage bag from a couch on the highway and take appropriate action.

So why wouldn't we still add radar and lidar? Because, they add their own set of problems to be solved in terms of integration, additional clock cycles on the computer and increases energy draw. Both Radar and Lidar are active sensors, and immediately take a relatively large hit in terms of Wh/mi of energy usage. Integration of the respective data streams also increases the computational load which then furthers increases the size of the computer and the amount of energy it requires to run FSD. Adding aerodynamically poor housings for your lidar means now the drag of the car increases and we take further hits on the energy efficiency of the car. By the time we are done, we may have double the energy requirements on an EV to drive autonomously, and an uglier Model Y with 150 mile range with no trunk space isn't that appealing for the mass market even if it can do FSD.

Integrating information from Lidar, radar, and vision can potentially cause its own problems. because which sensor do you trust? Lidar gives you a ton of false positives in the fog while radar sees nothing. Radar might give you a giant return that looks like a wall all across the highway when following a semi-truck, but your lidar gives you the outline of a box that shows the lane next to it is clear, which one do you trust? Seems obvious to us as human with eyes, but finding a general solution for a computer is not a trivial problem. In both of those cases you need the vision camera to distinguish what is happening, but if vision works then why do you need lidar and radar in the first place?

Thus Elon and Tesla's position is that vision has to be solve in order for FSD to work, if the vision problem is solve then radar and lidar are redundant useless appendices on a FSD car that causes their own set of problems. Tesla's proof is humans driving every single day without radar nor lidar. Whether or not they can do that is still an open question.

 

[ZenRockGarden]

One could argue that the vision-only problem must be solvable, because humans can do it.

That doesn't however mean that additional lidar or radar sensors wouldn't be potentially better - just that it can be done with vision.

 

[Twiglett]

yay - another thread of "Tesla need to do FSD the same as everyone else"
Thats the same argument from 10 years ago when EVs were a joke and everyone predicted Tesla would fail
Its also the same argument 10 years ago for reusable rockets when SpaceX was predicted to fail and the incumbents laughed at them.
Will it work? Dunno, but I'm not ready to be quite so convinced of failure as many here seem to be predicting.

Warning
Past performance is not a predictor of future success (but I'm not betting against them on it)

 

[S4WRXTTCS]

Two things.

I see zero evidence that Tesla is even interested in L3. The only people that seem to have any interest in L3 is those of us in environments where traffic assist type L3 would be really beneficial. Like I could probably drive from I5 near me all the way past Olympia, WA without ever exceeding the speed limits of L3 systems posed to come onto the market. That is if I left my house between the hours of 3-6pm.

I don't have a lot of faith that the initial roll out of L3 systems in the US will go well. We're total complete idiots on the road. Maybe we should allow Japan and Germany to have a bit of time with L3 before we get our hopes up of it ever being released in the US. Let the safe drivers use it and see what happens.

 

[Daniel in SD]

I don't have a lot of faith that the initial roll out of L3 systems in the US will go well. We're total complete idiots on the road. Maybe we should allow Japan and Germany to have a bit of time with L3 before we get our hopes up of it ever being released in the US. Let the safe drivers use it and see what happens.

I assume you're talking about the dangerous actions of other drivers on the road that a L3 system will have to deal with. The person in the drivers seat is not driving when the L3 system is engaged.
I'm very curious to see if these systems are actually as safe as human drivers. I have my doubts. If they actually are used for significant mileage I bet we're going to find out about some crazy edge cases.

Fun fact is that the only autonomous miles that Tesla has reported to the DMV were testing of the L3 system for the autonomy day demo video. I agree though that they seem to have no intent of pursuing anything other than robotaxi capability.

 

[S4WRXTTCS]

I assume you're talking about the dangerous actions of other drivers on the road that a L3 system will have to deal with. The person in the drivers seat is not driving when the L3 system is engaged.

dangerous actions of other drivers like lane incursions, and cut offs.
couches, cushions, ladders on the road due to people who couldn't secure their load
tire swallowing pot holes due to incompetence of state DOT's.
The sure amount of oscillation between >50 and <25 that happens due to people following too closely.
Our sue happy nature so if there was an edge case the lawyers will come out of the wood work.


The irony is we need L3 to text, but we can't have L3 to text because everyone is texting while driving.

Sure things won't be perfect in Germany or Japan, but they might at least stand a chance.

 

[Twiglett]

dangerous actions of other drivers like lane incursions, and cut offs.
couches, cushions, ladders on the road due to people who couldn't secure their load
tire swallowing pot holes due to incompetence of state DOT's.
The sure amount of oscillation between >50 and <25 that happens due to people following too closely.
Our sue happy nature so if there was an edge case the lawyers will come out of the wood work.


The irony is we need L3 to text, but we can't have L3 to text because everyone is texting while driving.

Sure things won't be perfect in Germany or Japan, but they might at least stand a chance.

Wonder if the first lawsuit will be in small claims court when one of those nice expensive Merc tires fails to save the nice expensive Merc rims?
If the car wrecks the rim by L3 driving it into a huge pothole, then it must be a warranty claim right?
Oh the fun

 

[Charliek]

How is it misleading? It's literally the example everyone uses for a hypothetical L3 feature...

And the first paragraph of the article seems very easy for people to understand. Clearly you understand it, the writer of the article understands it, so why do think others being mislead? Especially odd since you seem to have no problem with how FSD is marketed even though many people feel like they have been mislead.

View attachment 783971

wow this definition has evolved a bit from previous years . based on this and my experience with the latest FSD beta ( note my further caveat I am not in a city but in the suburbs ) FSD is at Level 3 when I use it today based on these descriptions. Recently I have only had to pop it out of FSD beta under very limited situations ( usually some other idiot driver) , maybe 2-3 more updates and I could honestly say level 3 at 99.9 percent of the time ( probably needs to be 4 to 5 9's) . I wonder what " test" or tests would be to take a claim and certify it as level 3. for me it does this now . and it is definitely not level 4.

 

[Daniel in SD]

wow this definition has evolved a bit from previous years . based on this and my experience with the latest FSD beta ( note my further caveat I am not in a city but in the suburbs ) FSD is at Level 3 when I use it today based on these descriptions. Recently I have only had to pop it out of FSD beta under very limited situations ( usually some other idiot driver) , maybe 2-3 more updates and I could honestly say level 3 at 99.9 percent of the time ( probably needs to be 4 to 5 9's) . I wonder what " test" or tests would be to take a claim and certify it as level 3. for me it does this now . and it is definitely not level 4.

You seem to be talking about reliability but the SAE levels don't actually specify a required reliability. L3 just means that while the system is engaged the driver is not required to monitor it.
How often do you think FSD Beta would have a collision (regardless of fault) with an impact speed of more than 12mph if you were not monitoring it?

Right now there aren't really any tests to certify that a L3-L5 system is safe enough. The manufacturer might report their testing procedure to the DMV (in states like CA where approach is required) but really everyone is hoping that liability is enough to discourage anyone from releasing an unsafe system.

 

[Supcom]

Since Tesla doesn't appear to be interested in getting an L3 certification, there would seem to be no chance of beating Mercedes to L3.

Based on my experience, getting L3 certification for the narrow traffic jam on limited access roads would be within the current capability of a Tesla. My car handles traffic jams exceptionally well and I cannot think of a single time when I had to disengage FSD while driving in one.

But, why should Tesla bother with L3 certification? It would detract from the main FSD development and add liability to the company. For what purpose? To sell cars???

 

[Bouba]

I can understand Mercedes take liability....perhaps the premiums are inbuilt in the price of the car...but it can’t tell you that you have no responsibility....the law is the law and the driver must be in control at all times