Autonomous Car Progress

[diplomat33]

Is Mobileye really promising L4 in consumer cars as opposed to fleets?

Yes they are. It will be called Mobileye Chauffeur. Mobileye is calling it a "turnkey self-driving system designed to turn a consumer car into L4 self-driven vehicle". They are targeting a cost of under $6,000 by 2025, according to the marketing they put out. It is listed as one of their products on their solutions website:

Mobileye Solutions | From Driver Assistance to Self-Driving

Discover our full spectrum of solutions, from the driver-assistance systems in over 100 million vehicles today to the turnkey self-driving systems of tomorrow.

www.mobileye.com

Correct me if I'm wrong, but Baidu, Cruise, and Waymo has zero plans to offer L4 on consumer cars, they are sticking to fleets. Their whole operation model is all based on a fleet with support infrastructure behind the scenes (not on a consumer vehicle that operates largely on its own).

I admit that part is some educated speculation, which is why I said "I think" and "IMO". I know they have not announced any L4 consumer car plans and are focused on robotaxis right now. However, Baidu announced a robotaxi that will only cost $37k and will be available in 10 cities in China next year. It would stand to reason that if all goes well with their robotaxi plans, they could offer it as a consumer car at that price if they wanted to. Cruise is owned by GM. Conceivably, in a few years, if Cruise's L4 is good enough, I think it would make sense for GM to adapt Cruise's L4 to their consumer cars in some way. And as far as Waymo is concerned, they have said they are interested in personal car ownership after they scale the robotaxi ride-hailing. The plan is to eventually let consumers lease a Waymo robotaxi and then after the lease is over, Waymo would add the car onto their ride-hailing network.

Bottom line is that if L4 is solved in 5+ years, I think it would only make sense for these companies to license their L4 to automakers. I think they are doing robotaxis now because it allows them to start using the tech while it is still being developed. It lets them test and offer ride-hailing to the public in a more controlled environment. But once L4 is solved, I don't see why they would continue to restrict themselves to only robotaxis.

 

[Supcom]

They are targeting a cost of under $6,000 by 2025, according to the marketing they put out.

Presumably $6000 when purchased in OEM quantities by a car manufacturer. Which means that the system, once integrated into a production vehicle and priced to the consumer will be at least $12,000. And there may be a monthly cost added to that for software and map updates, for all we know. And someone has to assume the liability for an L4 system. If it's not the consumer, then that risk will be included in the sales price.

And there is the tendency for auto makers to introduce new technology on their highest priced models first, which just adds more to the consumer cost until the technology trickles down to the mainstream products..

 

[drtimhill]

When I see how people drive, I conclude that there are more shortcomings in humans than just an insufficient attention span. They drive too closely behind other cars. They perform unsafe lane changes. They drive too fast in adverse conditions. They lose control in unusual steering situations. They ignore the possibility that there is something in their way that is obscured such that they temporarily cannot see it, i.e. they drive according to the rule, what I don't see, is not there.

To put it simply, many humans have a problem with making proper, logical, safe decisions in the real world. I cannot tell whether they lack the capability to think properly or whether they have that ability and just wilfully or lazily decide not to use it. And this causes practically all accidents.

That is why I cannot accept the argument that

Consequently, the argument that more sensors are needed to make autopilots better than humans is unfounded. To me it is obvious that, once it is properly programmed, a good autopilot with just cameras and perhaps ultrasound sensors will be able to drive much more safely than the average human.

The question remains whether additional sensors may allow an autonomous car to safely drive faster. I think that is currently unimportant and may become important when different autonomous cars compete for speed at high safety levels. As long as they compete with humans, more sensors are not needed. What's needed first is better artificial intelligence. Elon Musk understands this. He is right.

I was giving the benefit of the doubt to humans in terms of the potential to be better than AI. But I agree that in practice laziness and sheer lack of situational awareness leads to many avoidable accidents. And in all these areas that you mention even the simplest AI is better than a typical human (as I have noted in many other posts here). The fact is, most accidents are caused by the exact things that AI is better at .. continuous vigilance, lack of distraction, adherance to safe driving rules etc. Sure, we are not there yet, but we are getting closer every day.

 

[Doggydogworld]

Giving the driver a sudden "red hands on wheel" alert at the last second

Alerts are sudden, by definition. But this is not "last second". The driver has 10 seconds to look around, assess the situation, then take control.

I'm more worried about drivers taking over too quickly than too slowly. Take my example -- car ahead changes lanes due to stopped truck. Drive Pilot will alert you and start braking. If you don't respond immediately it will simply stop. No muss, no fuss. But if you look up and panic you might jerk the wheel to avoid the truck, and in doing so crash into an adjacent car.

The system has to give the driver enough time to re-engage with what is going on.

The system does. The issue is whether drivers will use that time, or will initiate maneuvers before they establish situational awareness. I'm sure Mercedes has studied and tested this, but get enough people using it in enough situations and stuff Mercedes has not tested will happen.

Mobileye is promising L4 consumer cars in 2-3 years.

I think they'll miss, but even if they market Chauffeur to OEMs in 2025 we won't be able to buy it in (high end) consumer cars until 2027. Heck, it's been 5 years since Audi was supposed to start shipping L3 in the A8. It's been 5 years since Waymo started pulling safety drivers in Chandler. It's been 5 years since Elon promised coast-to-coast autonomy (for the 3rd straight year, ha). Real world progress is glacial.

I agree, it's a problem with Mercedes' specific implementation. If anything blocks the eye monitoring camera, it seems to disable the system. This is something that can easily happen if someone was using their personal devices (like their phone in the above example, or a laptop),

It didn't beep at him for using his phone, but for literally holding the phone up directly in front of his face to make a video of traffic. That's hardly a "major use case".

Will it let you hold your phone (or tablet, laptop, book, etc.) normally? It should, but I've read Mercedes won't because "it's illegal". That makes no sense. If the UNECE L3 rule really says you can only use the car's screen, then the whole thing is just a colossal waste of time.

Requiring the user to use the car screen is a non-starter, especially given how poorly most automaker UI is made, and especially with a screen that is "integrated" and is further from the driver (and can't be rotated or moved closer to driver).

Agree 100%. If that's their long term plan they're nuts.

 

[diplomat33]

I think they'll miss, but even if they market Chauffeur to OEMs in 2025 we won't be able to buy it in (high end) consumer cars until 2027. Heck, it's been 5 years since Audi was supposed to start shipping L3 in the A8. It's been 5 years since Waymo started pulling safety drivers in Chandler. It's been 5 years since Elon promised coast-to-coast autonomy (for the 3rd straight year, ha). Real world progress is glacial.

Yes, Mobileye could miss their deadline. That is always possible. My point is just that even if they miss the original deadline of 2025, I am optimistic they will still release in 5+ years from now.

Also, companies missed their deadlines before but that is when their FSD was still very early. Now, we are much closer to wide scale L4 than we were 5 years ago. They are not going to keep missing deadlines by 5 years. At some point, they might only miss by 2 years and then only miss by 1 year and then it will happen.

 

[DanCar]

Yes, Mobileye could miss their deadline. That is always possible.

Very very likely they will miss in the U.S. with a wide deployment in my opinion.

My point is just that even if they miss the original deadline of 2025, I am optimistic they will still release in 5+ years from now.

Some freeways in China in 2025? Release only on a few freeways in Europe in 2027?

Also, companies missed their deadlines before but that is when their FSD was still very early. Now, we are much closer to wide scale L4 than we were 5 years ago. They are not going to keep missing deadlines by 5 years.

I'm not confident of that. Elon keeps missing by 5+ years.

At some point, they might only miss by 2 years and then only miss by 1 year and then it will happen.

Better to keep track of L3. When that happens then we will know we are closer to L4. No L3 means no L4 either. L3 signifies OEMs are ready to take a big step, which is to assume liability.

 

[diplomat33]

Very very likely they will miss in the U.S. with a wide deployment in my opinion.

Some freeways in China in 2025? Release only on a few freeways in Europe in 2027?

I think you are being overly pessimistic. Remember that Mobileye has already mapped major roads in Europe and the US. So why would they only release a super limited L4 that only works on highways? Plus, we know Mobileye is actively testing their FSD in Paris, Munich, Jerusalem, Tel Aviv, Orlando and NYC with L4 ride-hailing pilot expected to start in Tel Aviv and Munich later this year. They are clearly working towards deploying L4 in more than just a few freeways in China or Europe.

And remember too that with Mobileye's approach, the maps are crowdsourced from where ever they have consumer cars collecting data. And their approach is to achieve FSD with vision-only + maps and then just add the lidar-radar redundancy to automatically have enough safety and reliability for L4. So once the vision-only FSD is good enough, Mobileye will have L4 by just adding the lidar-radar redundancy, anywhere that they have crowdsourced maps.

I'm not confident of that. Elon keeps missing by 5+ years.

Because Elon has the wrong approach to FSD, Mobileye does not (IMO).

Better to keep track of L3. When that happens then we will know we are closer to L4. No L3 means no L4 either. L3 signifies OEMs are ready to take a big step, which is to assume liability.

Having L3 is not necessarily a sign of being closer to L4 though. Remember the SAE levels are not in progressive order. You don't need to go L2 -> L3 -> L4 -> L5. But I see your point about liability.

 

[DanCar]

I think you are being overly pessimistic. Remember that Mobileye has already mapped major roads in Europe and the US. So why would they only release a super limited L4 that only works on highways?

In terms of consumers Mobileye isn't going to release anything. OEMs are going to or not release L4. It is the OEMs that decide if they will accept liability. Mobileye can release whatever they want but there is likely a potential of not a single customer in the U.S. early on (years). I'm sure Mobileye will have an excellent L4 system in 2027 in the lab, that no one in the U.S. will be able to buy.

Plus, we know Mobileye is actively testing their FSD in Paris, Munich, Jerusalem, Tel Aviv, Orlando and NYC with L4 ride-hailing pilot expected to start in Tel Aviv and Munich later this year. They are clearly working towards deploying L4 in more than just a few freeways in China or Europe.

Robotaxi is different than selling cars to consumers. With Robotaxi Mobileye can directly release.

And remember too that with Mobileye's approach, the maps are crowdsourced from where ever they have consumer cars collecting data.

Remember that Mobileye promised L4 in 2018? Are you still optimistic about that? After missing that goal they then promised L4 in 2020? Are we waiting for that?

And their approach is to achieve FSD with vision-only + maps and then just add the lidar-radar redundancy to automatically have enough safety and reliability for L4. So once the vision-only FSD is good enough, Mobileye will have L4 by just adding the lidar-radar redundancy, anywhere that they have crowdsourced maps.

Talk is cheap, good enough in 2030?

 

[Goose66]

That is beside the point. Humans cause lots of accidents in spite of the far more advanced capabilities you conjure.

I conclude that there are more shortcomings in humans than just an insufficient attention span... Consequently, the argument that more sensors are needed to make autopilots better than humans is unfounded. To me it is obvious that, once it is properly programmed, a good autopilot with just cameras and perhaps ultrasound sensors will be able to drive much more safely than the average human.

I think your comment about attentiveness is “beside the point.” No one is arguing that an ADAS is not more attentive than a human or that they increase driving safety. The discussion is around visual perception by humans vs. Tesla’s "vision only" FSD. I am not "conjuring" capabilities of humans to make them sound better than camera-based perception - they exist in fact. Human visual processing is far superior to what AI can do today. I don't see how anybody can dispute that.

Common sense suggests (and other posters have commented here) that there are two ways to increase performance in a visual perception system: improve the processing of the system, or increase the inputs. Could a vision-only system, once "properly programmed" beat humans in visual perception? Maybe someday, but that is way down the road and may require computing power order(s) of magnitude greater than what is available in the car today. However, additional sensors beyond cameras are available now and, with cutting-edge work in sensor fusion that has been done by Nvidia, Baidu, and others since Tesla first laid out vision only strategy in 2016, could be integrated into a HW4 platform for Tesla today to yield better autonomous features.

The question remains whether additional sensors may allow an autonomous car to safely drive faster.

Driving faster is also not the point. Driving safer is the point. NNs are all about predictions. Everything is probability. The more sensor input (when properly fused and with a healthy dose of recurrence) the better the predictions. The better the predictions, the more confident and smoother the autonomous system will operate.

As long as they compete with humans, more sensors are not needed. What's needed first is better artificial intelligence. Elon Musk understands this. He is right.

Well, I don't think anybody that has FSD beta is saying it's close to competing with humans on visual perception and driving decisions (attentiveness aside). Again, sure better AI will help, but what does that better AI look like? You think Elon knows? Elon has made it clear over the last 6+ years he has no idea when L5 self-driving will be "solved," much less how it will be solved. IMO, FSD on the current HW3 platform will remain L2 ADAS (maybe L3 on highways after some time). When HW4 or HW5 is released someday for robotaxi, we'll see who was "right" when it comes to the need for more sensors. I'm betting it won't be Elon (at least what he is saying now - I'm sure he will be all about "sensor-fusion" at that point touting how Tesla has developed a "cutting-edge" sensor suite).

 

[DanCar]

Mercedes Preparing for Autonomous Driving in U.S. With Drive Pilot Level 3

We drive, or more correctly are driven, by a Mercedes-Benz EQS in Germany (where it is legal).

www.motortrend.com

Mercedes thinks people care that they fill out paperwork in California for L3.
Quote:
... with Drive Pilot Level 3 able to legally check their email or watch a movie in certain states by mid-2023.
...
In addition to five radars, six cameras, a wetness sensor, 12 ultrasonic sensors, a positioning system, microphones inside the car, this is Mercedes' first use of lidar. There are 35 devices in all.
End quote.

The good thing is that if it becomes popular, Elon will respond.

 

[Goose66]

Mercedes Preparing for Autonomous Driving in U.S. With Drive Pilot Level 3

We drive, or more correctly are driven, by a Mercedes-Benz EQS in Germany (where it is legal).

www.motortrend.com

Mercedes thinks people care that they fill out paperwork in California for L3.

I think this far more significant than another tweet from Elon saying they'll have FSD Capabilities rolled out to the wider fleet by "end of the year." If Mercedes does start shipping cars with true L3 capabilities in the U.S. by mid-2023, they will have beaten Tesla to the punch, which will not be a good look for Tesla, Elon, or the stock.

 

[MP3Mike]

a wetness sensor

Is that so it can tell if the last action it took was scary enough that it caused you to wet yourself?

 

[DanCar]

I think this far more significant than another tweet from Elon saying they'll have FSD Capabilities rolled out to the wider fleet by "end of the year." If Mercedes does start shipping cars with true L3 capabilities in the U.S. by mid-2023, they will have beaten Tesla to the punch, which will not be a good look for Tesla, Elon, or the stock.

I'm not sure people are going to care too much about L3 on freeways less than 37 mph on a $125K vehicle. If I had that money to burn yes I would care. Will they bring the cost down to a $50K vehicle by 2025?

 

[AlanSubie4Life]

I'm not sure people are going to care too much about L3 on freeways less than 37 mph on a $125K vehicle. If I had that money to burn yes I would care.

I think people would be pretty annoyed after all the promises (even though technically they only promised L2 of course) and focus on something other than highway L3 (something people have been clamoring for for years), Tesla was beaten to this by Mercedes. Perhaps not surprised, but there will be annoyance for sure.

 

[DanCar]

Would be nice if Merc published some technical detailed videos, like how pothole detection and avoidance works.

 

[Dewg]

It's funny to see the automotive industry start to look like the computer industry 10-15 years ago. People would brag about their PC rig setup, with the RAM speed, CPU speed, PSU, GPU, etc. "I've got my gaming rig clocked at 6GHz, what about you?" "I'm at 5.5GHz, but who cares about 6GHz - I've got dual water cooled GPUs running SLI, OC'd so I'm getting 120 fps at 6K resolution!" Now we're seeing more and more sensors on cars, and how much of that will just be marketing hype? "Oh, your car only has 12 LIDARS, my new XYZ car has 16 LIDARS!" "Yeah, but my car has 17 RADARS *and* 23 cameras! How many cameras does your car have?"

I get it, more sensing is better, but at some point it's just overkill. How many sensors does it take to get L3, L4 and L5?

 

[diplomat33]

It's funny to see the automotive industry start to look like the computer industry 10-15 years ago. People would brag about their PC rig setup, with the RAM speed, CPU speed, PSU, GPU, etc. "I've got my gaming rig clocked at 6GHz, what about you?" "I'm at 5.5GHz, but who cares about 6GHz - I've got dual water cooled GPUs running SLI, OC'd so I'm getting 120 fps at 6K resolution!" Now we're seeing more and more sensors on cars, and how much of that will just be marketing hype? "Oh, your car only has 12 LIDARS, my new XYZ car has 16 LIDARS!" "Yeah, but my car has 17 RADARS *and* 23 cameras! How many cameras does your car have?"

I think computing power, not number of sensors, is the thing that companies will use for bragging rights. After all, we don't see companies adding 12 or 17 lidars. We do see companies bragging that their computing power is 1200 TOPS. So computer TOPS is the bragging rights.

I get it, more sensing is better, but at some point it's just overkill.

Yes, at some point, more sensors becomes overkill. But companies are not just piling on more and more sensors. That is a strawman. Companies look for the sensor package that they think will provide them with the right coverage and redundancy for safety but also be cost effective. Obviously some companies think they can do that with 20 sensors, others might think they need 30 sensors. But nobody is just adding more sensors because more is better. They add sensors that they think will serve a practical safety purpose.

How many sensors does it take to get L3, L4 and L5?

That's the big question. Safe autonomous driving requires 360 degree perception and sufficient sensor fusion redundancy. But what sensor configuration is best to achieve that is an open question. Every AV company has their own approach. Obviously, since we don't have wide scale L3, L4 or L5 yet, there is no definitive answer yet.

Also, the answer depends on the SAE level and the ODD. For example, you don't need the same sensor suite for highway driving that you need for city driving. You don't need the same sensor suite for L3 that you need for L4 or L5. So it depends on what type of autonomous driving you are trying to do. Obviously, if your goal is L5 then you have to be able to do all types of driving (city, highway etc). So for L5, you will need the most complete sensor suite. If you are only doing L3 highway, you might be able to get by with fewer sensors since the car does not need to handle as many scenarios and has a human as the fall-back. The type of vehicle also affects the sensors you need. For example, trucks need longer range sensors since they require more distance to brake. A sedan car can stop much quicker so it might be able to get by with shorter range sensors than a truck. That's why Aurora developed in-house longer range lidar since their AV trucks need to see further away to be able to brake in time.

 

[Goose66]

I'm not sure people are going to care too much about L3 on freeways less than 37 mph on a $125K vehicle. If I had that money to burn yes I would care. Will they bring the cost down to a $50K vehicle by 2025?

I would be annoyed if limited to 37 mph too. But are we sure that's going to be the case in the U.S.? I read that to be specifically an autobahn thing - it definitely is not a technical limitation of the system. I guess we'll have to wait and see.

As far as the vehicle's price, it is a Mercedes S-class. But I imagine it will be available in E-class and C-class quickly thereafter. BTW, I remind that a Tesla Model 3 LR with FSD Capabilities is currently around $72K - about $4K more than a Mercedes E-450 with all of the ADAS features included.

 

[Doggydogworld]

I would be annoyed if limited to 37 mph too. But are we sure that's going to be the case in the U.S.? I read that to be specifically an autobahn thing - it definitely is not a technical limitation of the system. I guess we'll have to wait and see.

As far as the vehicle's price, it is a Mercedes S-class. But I imagine it will be available in E-class and C-class quickly thereafter. BTW, I remind that a Tesla Model 3 LR with FSD Capabilities is currently around $72K - about $4K more than a Mercedes E-450 with all of the ADAS features included.

60 kph is the current UN ECE regulation. They are raising it to 130 kph in January, but it could take a while for individual countries to adopt the higher speed. I don't think Mercedes plans to take current h/w to 130. Their lidar on these cars is pretty primitive. You need longer range and higher resolution at 130.

60 kph is easier in other ways, e.g.. you always have a lead vehicle. Not so at 130. Liability also goes way up with speed. Serious injury is almost impossible at 60 kph in these high-end cars, but screw up at 130 and death is a real possibility.

I don't think Tesla fans will care one bit if Mercedes "beats" them to Level 3. They're happy to wait another "3 weeks probably, 6 weeks definitely" for full Level 5.

 

[Goose66]

60 kph is the current UN ECE regulation. They are raising it to 130 kph in January, but it could take a while for individual countries to adopt the higher speed. I don't think Mercedes plans to take current h/w to 130. Their lidar on these cars is pretty primitive. You need longer range and higher resolution at 130.

60 kph is easier in other ways, e.g.. you always have a lead vehicle. Not so at 130. Liability also goes way up with speed. Serious injury is almost impossible at 60 kph in these high-end cars, but screw up at 130 and death is a real possibility.

I don't think Tesla fans will care one bit if Mercedes "beats" them to Level 3. They're happy to wait another "3 weeks probably, 6 weeks definitely" for full Level 5.

I assume by "Tesla fans," you mean "Tesla fanbois."