Rate of improvement

[spacecoin]

“Zero chance” is almost as useless a claim as “100% chance” when predicting future innovations.

I stand corrected: Almost zero chance. Level 5 means unlimited ODD. So handle ALL of these - in any weather a human would:

World's greatest driving roads

dangerousroads.org offers more than 12.000 roads online. Discover the most spectacular, scenic and dangerous roads around the world. Drive the most exciting and

www.dangerousroads.org

If it were so clear, we wouldn’t have the myriad of internet arguments about what they do or do not mean. Doing a quick search on just this forum for “SAE levels” and I see you’ve already been a part of them. You can reference those instead of rehashing the same arguments here. Hopefully we can keep this thread on topic.

In my experience, people tend to have a hard time understanding the concept of OEDR, and there is a myriad of clueless influencers like "Dr Know It All", that spread misinformation, sometimes by mistake, because they do not understand the concept either. J3016 is not a document for consumers nor legislators.

In L3 and above the systems perform the full OEDR (object and event detection and response) and in lower levels the system assists the driver with some of the OEDR. In short L3 means "eyes off" (watch a movie) - take over if the car tells you. The car performs the full OEDR and the DDT during the handover procedure. L4 means sleep in the back seat.

This is a good explanation: OEDR: The Key Differentiator Between SAE Level 2 And Level 3 Automated Driving

 

[TresLA]

I stand corrected: Almost zero chance. Level 5 means unlimited ODD. So handle ALL of these - in any weather a human would:

World's greatest driving roads

dangerousroads.org offers more than 12.000 roads online. Discover the most spectacular, scenic and dangerous roads around the world. Drive the most exciting and

www.dangerousroads.org

In my experience, people tend to have a hard time understanding the concept of OEDR, and there is a myriad of clueless influencers like "Dr Know It All", that spread misinformation, sometimes by mistake, because they do not understand the concept either. J3016 is not a document for consumers nor legislators.

In L3 and above the systems perform the full OEDR (object and event detection and response) and in lower levels the system assists the driver with some of the OEDR. In short L3 means "eyes off" (watch a movie) - take over if the car tells you. The car performs the full OEDR and the DDT during the handover procedure. L4 means sleep in the back seat.

This is a good explanation: OEDR: The Key Differentiator Between SAE Level 2 And Level 3 Automated Driving

The danger with oversimplifying the driver’s responsibility to SAE levels (and vague terms like OEDR) is it doesn’t clarify the actual performance/tasks of the vehicle. Mercedes’ L3 system isn’t as advanced as generally communicated (e.g. - media articles, public sentiment amongst those that at least heard about it). It’s only in forums such as these where we have a more detailed understanding of all the caveats and technicalities (37MPH, mapped freeways, etc.) Tesla has only declared L2, most likely to help mitigate litigation. Even amongst AEM or LKA systems the feature set and performance varies. The language is not clear. It’s not for lack of trying, but predicting the future is hard. We in society and the AV industry are still (understandably) discovering and defining language to communicate what a vehicle can or can’t do between advanced LKA and “sleep in the backseat” self driving.

 

[spacecoin]

The danger with oversimplifying the driver’s responsibility to SAE levels (and vague terms like OEDR) is it doesn’t clarify the actual performance/tasks of the vehicle. Mercedes’ L3 system isn’t as advanced as generally communicated (e.g. - media articles, public sentiment amongst those that at least heard about it). It’s only in forums such as these where we have a more detailed understanding of all the caveats and technicalities (37MPH, mapped freeways, etc.) Tesla has only declared L2, most likely to help mitigate litigation. Even amongst AEM or LKA systems the feature set and performance varies. The language is not clear. It’s not for lack of trying, but predicting the future is hard. We in society and the AV industry are still (understandably) discovering and defining language to communicate what a vehicle can or can’t do between advanced LKA and “sleep in the backseat” self driving.

Sure. "Level 3" is not a clear nor complete description of what the product can do, but again, not meant for laymen. Level 3 in the specified ODD xyz in which the manufacturer guarantees function and liability is better. The levels of autonowashing that's been going on with "full self-driving" L2 et.c. doesn't help.

What doesn't help either is the people that just look at the ODD and say "lol, tesla's drive everywhere". Autonomy is a different product than driver assist. Also a form of autonowashing.

Some efforts in clarifying the capabilities for the average consumer, like MobilEye's are a step in the right direction.

Hands Off, Eyes Off: A New Taxonomy for Automated Driving | Mobileye Blog

Instead of autonomous driving levels, our new taxonomy defines assisted and autonomous driving systems by degree of driver involvement.

www.mobileye.com

Finally, Tesla isn't L2 right now for legal reasons. It's because they fail too often. I doubt that will change for existing cars in the near term.

 

[TresLA]

Sure. "Level 3" is not a clear nor complete description of what the product can do, but again, not meant for laymen. Level 3 in the specified ODD xyz in which the manufacturer guarantees function and liability is better. The levels of autonowashing that's been going on with "full self-driving" L2 et.c. doesn't help.

What doesn't help either is the people that just look at the ODD and say "lol, tesla's drive everywhere". Autonomy is a different product than driver assist. Also a form of autonowashing.

Some efforts in clarifying the capabilities for the average consumer, like MobilEye's are a step in the right direction.

Hands Off, Eyes Off: A New Taxonomy for Automated Driving | Mobileye Blog

Instead of autonomous driving levels, our new taxonomy defines assisted and autonomous driving systems by degree of driver involvement.

www.mobileye.com

Finally, Tesla isn't L2 right now for legal reasons. It's because they fail too often. I doubt that will change for existing cars in the near term.

It’s all “autonowashing” (including Mercedes’ DrivePilot) until we have clear & accepted language. Again, many have tried (and are still trying) like MobileEye that you mention, but nothing is yet accepted as the standard by which others will communicate capability. Tesla has stated FSDb is L2, but they haven’t stated ”why”, other than for legal sounding language. Again, the SAE levels are only good for liability language, not for capability language.

e.g. - How many accidents per whatever miles is acceptable to be considered self-driving? Less than human stats? Worldwide? What if it’s really close? What if it handles the vast majority of roads more safely (accidents-wise) but only half of the “worlds most dangerous roads”? What if it could do everything, literally everything, EXCEPT unprotected left turns comfortably for human passengers? The real world doesn’t fit nicely into standards and rules.

 

[zoomer0056]

It’s all “autonowashing”

YUP ^^^
I had to look that up, added it to my vocabulary.

 

[spacecoin]

e.g. - How many accidents per whatever miles is acceptable to be considered self-driving? Less than human stats? Worldwide? What if it’s really close? What if it handles the vast majority of roads more safely (accidents-wise) but only half of the “worlds most dangerous roads”? What if it could do everything, literally everything, EXCEPT unprotected left turns comfortably for human passengers? The real world doesn’t fit nicely into standards and rules.

The average driver has an accident every 30 years on the highway. Liability goes a long way. You go out of business if you launch a shitty product. That, and your customers die.

 

[mjwojtukiewicz]

Let me add an incident that happened today. I was going north on a 2 lane highway in the right lane going 50+. I approached a clover leaf on ramp that had 3 cars waiting to enter. The car slows down, rather than accelerate past. It then tells me it's going to change to the left lane and of course somebody comes up like a bat out of hell. The car goes into a seizure, I had to take over and get out of there. This stuff needs ALOT work before it can be autonomous. Elon is deluding himself and his customers that he's already there. He takes all the radar and sonar out and thinks TV cameras can figure everything out. He says
he uses Lidar on his spacecraft. Yeah, one effup and everyone is dead. So those of us on earth are less important?
And I have my OWN Chuck Cook left that NEVER gets it right. If I have the energy I'll upload the videos. You need nerves of steel to make this left

 

[LadyOnikara]

I'll just wait until automated driving is perfected before going all out and trying it. My Tesla was born last year while I have been on the road for a bit more than 20 years, so I have more experience on the roads around here than he does.

 

[Supcom]

I'll just wait until automated driving is perfected before going all out and trying it. My Tesla was born last year while I have been on the road for a bit more than 20 years, so I have more experience on the roads around here than he does.

That's an interesting statement. From what I have seen in release notes, the software in the car is typically trained using hundreds of thousands of examples for a given maneuver. So, if Tesla feeds 150,000 examples of an unprotected left turn (in reality, it's probably many more than this) that would be like doing 20 ULTs every day for 20 years. That's roughly on par with a human's driving. So, your car may actually have more experience than you. It may not learn as quickly, but it's not for lack of experience.

 

[daniel]

I'll just wait until automated driving is perfected before going all out and trying it. My Tesla was born last year while I have been on the road for a bit more than 20 years, so I have more experience on the roads around here than he does.

Perfection should not be the goal. Better than human should be the goal. I'll buy the first self-driving car that can drive on all my local roads and is better than human. Or, better than me. I'm probably below-average for a human driver. Meanwhile, my Tesla drives better than me on highways. I'm not optimistic about a better-than-human self-driving car that can handle all my local roads in the near future.

 

[spacecoin]

Perfection should not be the goal. Better than human should be the goal. I'll buy the first self-driving car that can drive on all my local roads and is better than human. Or, better than me. I'm probably below-average for a human driver. Meanwhile, my Tesla drives better than me on highways. I'm not optimistic about a better-than-human self-driving car that can handle all my local roads in the near future.

I assume you mean better (equal speed and quality of ride) and safer (less accidents and mistakes) when you say "better"? Currently I don't think one can claim the highway assist is better nor safer than a human if the system was unsupervised. Is it convenient and of some value? Yes.

 

[TarmacSurfer]

Perfection should not be the goal. Better than human should be the goal. I'll buy the first self-driving car that can drive on all my local roads and is better than human. Or, better than me. I'm probably below-average for a human driver. Meanwhile, my Tesla drives better than me on highways. I'm not optimistic about a better-than-human self-driving car that can handle all my local roads in the near future.

Given the lack of skill demonstrated routinely by most drivers, that’s a pretty low bar!

Your point is a good one though - technology and automation absolutely can and should be an improvement simply by virtue that it takes human emotion out of the process (aggressiveness, fear, impatience, ego, machismo, timidness, indecisiveness, etc.) that all contribute to our transportation system being so screwed up and unsafe.

The long-term dream is cars whizzing along at very high speeds more or less bumper-to-bumper and able to coordinate spacing / timing / arrival times at intersections so it eliminates traffic lights and other time-wasting “dumbed down” solutions that have had to be created so that humans do not kill each other (and still manage to).

Doubt we’ll get there in the next 10 years or even 50, but that’s the hope.

The ability of an automated system to obtain, process, and make decisions based on inputs other than simply visual / visible-light-spectrum holds fantastic potential too - particularly in darkness or poor weather. Tesla has completely turned its back on that goal (for now?) though through the elimination of USS and pushing “Tesla Vision”, but maybe it’ll be something that gets layered on later once they figure out the basics in good weather / light conditions first.

 

[zoomer0056]

Given the lack of skill demonstrated routinely by most drivers, that’s a pretty low bar!

Your point is a good one though - technology and automation absolutely can and should be an improvement simply by virtue that it takes human emotion out of the process (aggressiveness, fear, impatience, ego, machismo, timidness, indecisiveness, etc.) that all contribute to our transportation system being so screwed up and unsafe.

The long-term dream is cars whizzing along at very high speeds more or less bumper-to-bumper and able to coordinate spacing / timing / arrival times at intersections so it eliminates traffic lights and other time-wasting “dumbed down” solutions that have had to be created so that humans do not kill each other (and still manage to).

Doubt we’ll get there in the next 10 years or even 50, but that’s the hope.

The ability of an automated system to obtain, process, and make decisions based on inputs other than simply visual / visible-light-spectrum holds fantastic potential too - particularly in darkness or poor weather. Tesla has completely turned its back on that goal (for now?) though through the elimination of USS and pushing “Tesla Vision”, but maybe it’ll be something that gets layered on later once they figure out the basics in good weather / light conditions first.

Emotions are definitely a factor. So too are skill, attentiveness, rules of the road both legal and common sense, respect ... .Taking the human out of the equation is difficult. It will be years.

 

[daniel]

I assume you mean better (equal speed and quality of ride) and safer (less accidents and mistakes) when you say "better"? Currently I don't think one can claim the highway assist is better nor safer than a human if the system was unsupervised. Is it convenient and of some value? Yes.

Sorry. When I said "better" I meant (and should have said) safer.

AP/EAP in my experience (and I only use it on the highway for lane/speed-keeping) when used as intended, which is to say, closely supervised, is safer than when I drive without it.

Given the lack of skill demonstrated routinely by most drivers, that’s a pretty low bar!

Indeed, it is a low bar. And of course it's not the ultimate goal, but rather the first goal. Once an autonomous car is safer than a human driver, it can begin saving lives. And for me, as an old man approaching the age when I'll have to stop driving, it would enable me to have transportation independence when I can no longer drive myself. Taxis and ride-shares are extremely inconvenient.

Sadly, I no longer expect to be able to buy such a car in my lifetime. The problem (fully-autonomous, Level-5, no-human-intervention) is turning out to be much more difficult than I and many others thought. Does anybody really expect Tesla FSD to go from Level 2 to Level 5 in the next 5 or 10 years? Certainly not while Elon is crippling his own team with whimsical technology restrictions, unsustainable work hours, and moving engineers out to work in another company.

 

[spacecoin]

Given the lack of skill demonstrated routinely by most drivers, that’s a pretty low bar!

Not really. A human is involved in an accident 2-3 times per lifetime on average.

Your point is a good one though - technology and automation absolutely can and should be an improvement simply by virtue that it takes human emotion out of the process (aggressiveness, fear, impatience, ego, machismo, timidness, indecisiveness, etc.) that all contribute to our transportation system being so screwed up and unsafe.

The long-term dream is cars whizzing along at very high speeds more or less bumper-to-bumper and able to coordinate spacing / timing / arrival times at intersections so it eliminates traffic lights and other time-wasting “dumbed down” solutions that have had to be created so that humans do not kill each other (and still manage to).

Doubt we’ll get there in the next 10 years or even 50, but that’s the hope.

We're definitely get there in the next 10 years, but in limited ODD:s. We're almost already there (Waymo et al). But not for camera only systems like Teslas. That will take at least 10 years.

The ability of an automated system to obtain, process, and make decisions based on inputs other than simply visual / visible-light-spectrum holds fantastic potential too - particularly in darkness or poor weather. Tesla has completely turned its back on that goal (for now?) though through the elimination of USS and pushing “Tesla Vision”, but maybe it’ll be something that gets layered on later once they figure out the basics in good weather / light conditions first.

This is a solved problem. More modalities such as Lidar and radar are needed given the state of computer vision rn.

 

[daniel]

We're definitely get there in the next 10 years, but in limited ODD:s. We're almost already there (Waymo et al). But not for camera only systems like Teslas. That will take at least 10 years.

Waymo et al are operating in very small areas with ideal streets and perfect weather. Generalizing that to anywhere and under any conditions that a cautious human would drive is a huge problem. Tesla STILL calls my EAP a "beta" feature.

 

[spacecoin]

Waymo et al are operating in very small areas with ideal streets and perfect weather.

This isn't correct. Waymo navigates construction and police signs and drives in fog and heavy rain nowadays. I think they can scale out fast once they're "done".

 

[TresLA]

This isn't correct. Waymo navigates construction and police signs and drives in fog and heavy rain nowadays. I think they can scale out fast once they're "done".

Waymo has hit important milestones. They’re apples and oranges compared to Tesla’s approach and should be evaluated differently, but they’re still far from full autonomy.

 

[spacecoin]

Waymo has hit important milestones. They’re apples and oranges compared to Tesla’s approach and should be evaluated differently, but they’re still far from full autonomy.

It depends on what you mean by "full autonomy". Their goal isn't to drive on any remote farm road, they aim at the dense urban areas and highway.

I found this to be quite impressive. Manual stop sign plus hand signals handled like a pro:

 

[daniel]

It depends on what you mean by "full autonomy". Their goal isn't to drive on any remote farm road, they aim at the dense urban areas and highway.

Full autonomy is the ability to drive without human intervention on any road that would generally be considered reasonable for a human to drive on. No need to be able to drive through a blizzard or off road. But should be able to drive from any postal address to any other.

If Waymo only intends to operate in dense urban areas and highways, then a lot of people won't be able to use such a car as their daily driver. If Waymo intends to remain a taxi service only, this approach is economically feasible, because you know both end points of a ride at the time it is requested. But a consumer car needs more. I want a car that can drive me wherever I want to go. That does not include off-road or roads rated 4-wheel. But it does include South Kihei Road, which sometimes is the best route, and which sometimes gets used as a detour route when Pi'ilani Highway is closed for construction or mishap.