Let's talk Self Driving Cars again!

[Midpack]

When will SDC be ready? Maybe not Tesla, but Waymo already is, I don't know. However, Waymo system is expensive and perhaps they don't even know the real price in production yet.

I thought Waymo was going to reach unrestricted Level 5 (current Tesla FSD is Level 2?), but I’m not so sure anymore. As good as their system is, it relies heavily on almost perfect weather, clearly marked roadways/signs and highly detailed and controlled mapping - that probably won’t be achievable everywhere. Their system may roll out in discrete areas, large metros, and be true Level 5 - a good finite geographic area business but not “anywhere” Level 5 like Tesla, Cruise, Mobileye, Comma.AI and others.

I’m looking forward to Level 5 but there are endless edge cases and exceptions. We’ll get there but the last obstacles, a tiny few noted in the OP, are probably beyond our full comprehension unless you’re an actual AI expert (vs enthusiast). 80% of Level 5 is relatively easy compared to the last 19% - we’ll never have 100% perfect Level 5 with humans...

Just read that Tesla is scrapping the highly touted FSD 8.3, and broader rollout delayed for 9.0 in April?

And to my knowledge, none of the Level 5 developers are confronting severe weather (ice, snow, heavy rain, etc.) - a well known barrier. It would take highly detailed mapping and more consistent road markings to overcome - requiring significantly more capacity and CPU time. Today’s systems would pull the car over when it can’t “see” - acceptable?

I see the exact same issues are surfacing in this thread (years later), so we’re doomed to plow the same ground again. Sigh. How many accidents are OK? Is there a practical driver attention threshold? Is advanced driver assist enough, better than Level 5? Involuntary guinea pigs? Basically arguing Schopenhauer over and over and over...

https://youtu.be/zYZTF-ohQKU

 

[braumeister]

As good as their system is, it relies heavily on almost perfect weather, clearly marked roadways/signs and highly detailed and controlled mapping - that probably won’t be achievable everywhere

They still don't fly, either. Another broken promise!

 

[Aerides]

I think we're a ways off from true auto-pilot driving (10 years?) and along the way there will be lots more stops and starts, and exploding articles and back tracking with every accident.

But I also think by the end of my life it will be the norm.

 

[donheff]

I like my Lexus adaptive cruise control and hope for more and better as the years go by. It's a PITA to keep the car at a steady speed for long periods on the highway. The cruise handles that well although it is easy to lose track of the fact that you have gotten behind a vehicle that is too slow. I assume the Tesla can change lanes of its own volition in that situation. I will be glad to see that capability become ubiquitous. I'm not holding my breath for these systems to handle the types of situations ERD outlined. For many years we will need to be attentive if we want high level performance. If the systems are overly cautious though, they may soon be good enough for unattended safety.

 

[Amethyst]

I agree, the same issues surface, yet to me that is not really bad. Shows we are paying attention; interested in even small advances. I learn something from each thread.

I'm not informed enough to know if SDC are at the 80% point, but it sounds plausible.

The question is, how close to the unattainable 100% would most people willing to accept, and what specific functionalities and levels of performance does that acceptable threshhold entail? Any recent assessments out there?

I

I’m looking forward to Level 5 but there are endless edge cases and exceptions. We’ll get there but the last obstacles, a tiny few noted in the OP, are probably beyond our full comprehension unless you’re an actual AI expert (vs enthusiast). 80% of Level 5 is relatively easy compared to the last 19% - we’ll never have 100% perfect Level 5 with humans...

Just read that Tesla is scrapping the highly touted FSD 8.3, and broader rollout delayed for 9.0 in April?

And to my knowledge, none of the Level 5 developers are confronting severe weather (ice, snow, heavy rain, etc.) - a well known barrier. It would take highly detailed mapping and more consistent road markings to overcome - requiring significantly more capacity and CPU time. Today’s systems would pull the car over when it can’t “see” - acceptable?

I see the exact same issues are surfacing in this thread (years later), so we’re doomed to plow the same ground again. Sigh. How many accidents are OK? Is there a practical driver attention threshold? Is advanced driver assist enough, better than Level 5? Involuntary guinea pigs? Basically arguing Schopenhauer over and over and over...

https://youtu.be/zYZTF-ohQKU

 

[JoeWras]

The question is, how close to the unattainable 100% would most people willing to accept, and what specific functionalities and levels of performance does that acceptable threshhold entail? Any recent assessments out there?

The 100% safety is something that airlines have to fight too. However, airlines have an important advantage: professional, trained, operators and heavy regulation. This helps put the public at ease.

It is my opinion that for the public to really accept SDC, regulations and infrastructure need to be improved. I'm not sure the public will accept an anything-goes attitude from manufacturers. Some standards and testing will have to be created, with the vehicles passing these tests. That's one step. I believe another step toward SDC is improved infrastructure. Start with SDC approved roads that have proper markings and maybe even supplemental guidance systems along the road for tough weather. In other words, per "level 4," we'll start with certain roads or areas.

I also think the industry would be wise to introduce it with professional operators as backup. I'm thinking the trucking industry. This would be much like the air travel industry. Prove it works with trained, professionals as back-up in case manual control is required.

All the dreaming and pontificating is fun right now. Let Elon be Elon. He pushes the limits. But Elon will also have to pass some tests and regulations.

Right now, Cowboy-Wild-West won't instill confidence in the majority of the public. The industry needs to do better now, including realistic expections, else they suffer years or decades of delay due to mistrust.

 

[Amethyst]

Your arguments seem particularly germane to the idea of SDC being a boon to people who have had to give up driving, yet want to stay mobile. They (we?) are bound to be a fearful, cautious group of customers.

Maybe I am just not listening/watching the right things - but I really don't see as much discussion of SDC as I'd expect. If you go to the Tesla forum, you hear from Tesla owners and fans. But in general, the country behaves as if SDC are still science fiction.

Right now, Cowboy-Wild-West won't instill confidence in the majority of the public. The industry needs to do better now, including realistic expections, else they suffer years or decades of delay due to mistrust.

 

[Whisper66]

Thanks for reopening this area for discussion. I occasionally browse discussions like this to gage the status of the technology. The main take away for me has been and continues to be ... When I see a Tesla on the road, I avoid being anywhere near it. I don't want to be near one when a driver, lulled into letting the car manage the trip by "autopilot", allows his/her Tesla to make some erratic move that costs me or my family their lives. One day this technology may be ready for prime time. It doesn't appear to be there yet.

 

[JoeWras]

Maybe I am just not listening/watching the right things - but I really don't see as much discussion of SDC as I'd expect. If you go to the Tesla forum, you hear from Tesla owners and fans. But in general, the country behaves as if SDC are still science fiction.

That's kind of what I'm talking about. The mistrust is already bubbling. There was so much optimism 5 years ago due to lofty claims made by some manufacturers and highly publicized trials like Waymo.

Then... nothing happened. Well, nothing we can see. I'm sure a ton of data has been collected which will move things forward.

But outwardly, it appears promises have already been broken and this ticks people off.

Oh, and of course the media had pandemic and politics they'd rather to discuss because those make for more advertiser engagement.

 

[Midpack]

Thanks for reopening this area for discussion. I occasionally browse discussions like this to gage the status of the technology. The main take away for me has been and continues to be ... When I see a Tesla on the road, I avoid being anywhere near it. I don't want to be near one when a driver, lulled into letting the car manage the trip by "autopilot", allows his/her Tesla to make some erratic move that costs me or my family their lives. One day this technology may be ready for prime time. It doesn't appear to be there yet.

I wonder how many times you’ve been driving right next to a car (or a semi) using Mobileye, Cruise or Comma 2? Or how many times you’ve been right next to a drunk, high, distracted and/or angry driver? And the Tesla autopilot is hardly any different than the driver assistance systems on any newer make/model cars - that you're driving next to.

Tesla FSD is an added cost option, that only some percentage of buyers add, and that system is not a hands free system. The beta FSD package is limited to a very small group of drivers, and it has driver monitoring - Tesla has already taken the beta FSD away from drivers caught not paying sufficient attention.

 

[Midpack]

The question is, how close to the unattainable 100% would most people willing to accept, and what specific functionalities and levels of performance does that acceptable threshhold entail? Any recent assessments out there?

There will have to be substantially lower rates of accidents/fatalities before acceptance is even a possibility. The NHSTA says 93% of accidents stem from human error, other studies in the UK and USA report 90-99% attributed to human error. If autonomous driving can dramatically reduce human error accidents/fatalities - we’ll all want (or be forced) to reconsider eventually provided the systems are affordable.

It’s probably a long way off, but there may come a time where it’s illegal to operate a car manually in some area dominated by autonomous cars because manually operated cars have proven far more dangerous. But for those of you bristle at the idea, don’t worry, that may not be in the lifetime of most members here. If autonomous cars do prove 10x safer for example, insurance for manually operated cars could become prohibitively expensive...

There was a time when landlines were common and smartphones didn’t exist, not anymore. We haven’t always had PCs or thr internet, now they’re ubiquitous. There was a time when horses moved people and goods on land, not anymore. Before the Industrial Revolution 80% of Americans were engaged in farming, now it’s less than 1% and we waste tons of food. Most people vociferously didn’t believe those changes would happen, they refuse to try to see the future...

 

[Whisper66]

I wonder how many times you’ve been driving right next to a car (or a semi) using Mobileye, Cruise or Comma 2? Or how many times you’ve been right next to a drunk, high, distracted and/or angry driver? And the Tesla autopilot is hardly any different than the driver assistance systems on any newer make/model cars - that you're driving next to.

Who knows... but whenever I know I'm driving next to a car / driver I feel is potentially dangerous, I move away from them. Drunk, highly distracted, angry drivers are often easily recognizable examples. Tesla cars are also easily recognizable. When Tesla gets the technology right, I will be comfortable driving around them. That time is not now.

 

[MRG]

I wonder how many times you’ve been driving right next to a car (or a semi) using Mobileye, Cruise or Comma 2? Or how many times you’ve been right next to a drunk, high, distracted and/or angry driver? And the Tesla autopilot is hardly any different than the driver assistance systems on any newer make/model cars - that you're driving next to.

Tesla FSD is an added cost option, that only some percentage of buyers add, and that system is not a hands free system. The beta FSD package is limited to a very small group of drivers, and it has driver monitoring - Tesla has already taken the beta FSD away from drivers caught not paying sufficient attention.

+1

In my logging days I knew a guy who owned a logging company including semi trucks hauling logs. One night he was called by the state police after his truck was pulled over. The officer told him his driver was going 65mph straight down the highway, he was pulled over for an equipment check because he had taillights out. When the officer approached the cab the driver didn't answer, he opened the cab and the driver was passed out drunk.

What's it take, like 5 minutes, for them to approach the vehicle? 65mph to passed out drunk in a couple minutes. This was in the 70s surely nothing like that happens today. Because?

 

[foxfirev5]

Our 2020 Honda CR-V has several Driver Assist features:


Forward Collision Warning
Lane Departure Warning
Collision Mitigation Braking System™
Road Departure Mitigation System
Lane Keeping Assist System
Adaptive Cruise Control


I like that it makes a noise if I am not braking when I get too close to a car in front of me. I like when it gives haptic feedback if I get too close to the edge of the lane I'm in. I've always been impressed with these features. But even with this system, which has never been touted as leading towards autonomous driving, I've had a few surprises.



It is annoying that even when I am changing lanes due to my lane *ending* - if i pass over some stripes the steering wheel sometimes is rather aggressive in trying to keep me back in the lane. On one occasion it applied the brakes enough to pull me out of cruise control! If I use the turn signal, it ignores lane changes (as it should).


I can't imagine a more 'intelligent' aggressive like Tesla's taking over. I don't think I'll still be driving by the time such a thing becomes mainstream.

Just bought a 2021 RAV4 with these features. Last year I rented one in SE Fla and the lane/road departure features drove me nuts. I turned those off. DW said maybe it's because I'm a bad driver. As well, driving here in Michigan with potholes the driver is constantly making adjustments. The beeping is actually a distraction so the lane features are disabled. The adaptive cruise and braking worked good and are adjustable for sensitivity.

At 65 yo this is probably the last of my technological advances in cars. DW's Lexus is 17 yo but will serve only as a backup.

I doubt I'll be around long enough for an autonomous self driving car that does not need a licensed driver. By then it will probably be a pickup service of some kind. I'll let the kids and GK's deal with that.

 

[Midpack]

Maybe I am just not listening/watching the right things - but I really don't see as much discussion of SDC as I'd expect. If you go to the Tesla forum, you hear from Tesla owners and fans. But in general, the country behaves as if SDC are still science fiction.

They’re all over YouTube, that’s where I see a lot of it. I watch way more YouTube more than broadcast TV, but I know lots of folks still think YouTube is just cat videos and other nonsense.

 

[NW-Bound]

For people not following this technology, let's rehearse the different levels of SDC as defined by SAE. The elaboration that I include comes from a Wikipedia article.

Starting from the top ultimate levels:

Level 5: Full Automation

In Level 5, no human intervention is required at all. An example would be a robotic vehicle that works on all kinds of surfaces, all over the world, all year around, in all weather conditions.

Obviously, we are not going to get there any time soon. I did not think so 5 years ago, and have not changed my mind. It is the ultimate goal of every maker, but we are so far from it there's no point to even talk about it now. I don't think Waymo ever claims to have it.


Level 4: High Automation

No driver attention is ever required for safety, e.g. the driver may safely go to sleep or leave the driver's seat. However, self-driving is supported only in limited spatial areas (geofenced) or under special circumstances. Outside of these areas or circumstances, the vehicle must be able to safely abort the trip, e.g. slow down and park the car, if the driver does not retake control. An example would be a robotic taxi or a robotic delivery service that covers selected locations in an area, at a specific time and quantities.

This is something that Waymo has been working towards. Have they achieved it? I am not sure, but the system looks very expensive at $100K or more.


Level 3: Conditional Automation

The driver can safely turn their attention away from the driving tasks, e.g. the driver can text or watch a movie. The vehicle will handle situations that call for an immediate response, like emergency braking. The driver must still be prepared to intervene within some limited time, specified by the manufacturer, when called upon by the vehicle to do so. You can think of the automated system as a co-driver that will alert you in an orderly fashion when it is your turn to drive. An example would be a Traffic Jam Chauffeur, another example would be a car satisfying the international Automated Lane Keeping System (ALKS) regulations.

This Level 3 is still useful, and safe to the user. If he does not take over immediately, he is not going to die.


Level 2: Partial Automation

The automated system takes full control of the vehicle: accelerating, braking, and steering. The driver must monitor the driving and be prepared to intervene immediately at any time if the automated system fails to respond properly. The shorthand "hands off" is not meant to be taken literally – contact between hand and wheel is often mandatory during SAE 2 driving, to confirm that the driver is ready to intervene. The eyes of the driver might be monitored by cameras to confirm that the driver is keeping their attention to traffic.

In Youtube videos of the current Tesla FSD, I saw instances of the car lurching into oncoming traffic in a left turn, or veering towards the curb or a parked car, making the driver crying out "No", and taking over. That's not Level 3. It's only Level 2. And Level 2 is what Tesla says its FSD is in regulatory filings.

In order to be a robotaxi, you don't need to be at Level 5. Level 4 will work. That's what Waymo and AutoX appear to claim at this point.

For a personal car, I would not mind to pay $20K or maybe even more to get Level 3.

Level 2 requires too much attention and too stressful in city driving for me. For highway driving, it may be acceptable.

 

[Midpack]

...

Waymo introduced a fully driverless ride-hailing service to the east side of metro Phoenix. The service was made available to members of Waymo One, the company’s premium membership, on Oct. 8, and it’s now available to anyone in Chandler, southeast Tempe, and southwest Mesa through Waymo’s app.

 

[Zathras]

...

When will SDC be ready? Maybe not Tesla, but Waymo already is, I don't know. However, Waymo system is expensive and perhaps they don't even know the real price in production yet.

One thing about Tesla FSD is that they let outsiders, meaning Youtube beta testers, have access to their system, so that we know about the shortcomings. Waymo is very tight-lipped about their stuff, so I don't know what their real capability is.

Waymo system is not ready for general use.
It is ready in geo-fenced and limited areas. It depends upon extremely detailed mapping of the area.
Tesla's software is learning and is also not readily available. Those who want to test it, and help with its development can.

Those that want to wait until FSD is fully ready and the Beta name is removed, can wait for that.

 

[Ready]

The main take away for me has been and continues to be ... When I see a Tesla on the road, I avoid being anywhere near it. I don't want to be near one when a driver, lulled into letting the car manage the trip by "autopilot", allows his/her Tesla to make some erratic move that costs me or my family their lives. One day this technology may be ready for prime time. It doesn't appear to be there yet.

The autopilot actually does a great job at avoiding other cars. Inattentive drivers might not react quickly enough when the car in front of them slows down abruptly. But autopilot will always handle those situation quite well. I don't like using autopilot because it drives like a little old lady and doesn't want to take any risks. But that actually makes it a very safe car for anyone else on the road driving near it.

 

[NW-Bound]

Waymo system is not ready for general use.
It is ready in geo-fenced and limited areas. It depends upon extremely detailed mapping of the area.
Tesla's software is learning and is also not readily available. Those who want to test it, and help with its development can.

Those that want to wait until FSD is fully ready and the Beta name is removed, can wait for that.

Yes.

I live in the area where Waymo has been testing its SDC for the last 5 or 6 years. Every time I go out on an errand, I see a few of their cars on the road. When Waymo opened the limited service to the public recently, I started a thread on it, but there was little interest at that point. Perhaps people were too busy with politics and the pandemic then.

I never have an interest to register to catch a ride on a Waymo robotaxi. There's not much to learn on a ride, and I can get more info watching the numerous videos on Youtube. Even then, it may not reveal a lot. Here's why.

In order to be a robotaxi, you have to be at least Level 4 SDC (see my earlier post). However, whenever I see a Waymo car, it always has a driver. In fact, I have not seen one with a passenger in the back. Maybe it was because they were reclining. There are plenty of Youtube videos of Waymo robotaxis carrying test passengers, and a couple of videos were even driverless. However, Waymo said that it still had test drivers onboard most of the time even though they were there just for passenger's comfort. Maybe, but I am sure that the requested route was vetted by their system before the passenger was allowed on board.

It is easily seen that Tesla is trying to solve a much tougher problem, hence would show a lot of flaws in the test videos that Youtubers post. Tesla also has no control over the conditions that the beta testers submit its system to. I have to give Tesla credit for that.

I am trying to understand the "smart level" of different systems, despite their differences in operating conditions and resources that they have. All of these systems use AI "deep learning", and I want to know the state of the art in this technology, despite the differences in the implementation of different systems.

Take for example Waymo's detailed mapping of their geofenced system. When they first opened their center here, I saw their cars running around my residential neighborhood several times when I went for a walk, and I did not spend that much time outside. Later, I learned that it was apparently for them to map out the area.

I expect that Waymo would also map out all the parking lots of shopping centers, supermarkets, etc... You need to be able to navigate parking lots to pick up and drop off passengers. Tesla on the other hand is trying to do a lot with the car vision cameras, hence its "smart summon" mode shows very crude performance. However, it does not mean that Waymo SDC is necessarily smarter.

More on the performance comparison of the above systems later.

 

[Animorph]

Animorph, what has been your experience with low-visibility situations, such as a pop-up heavy rainstorm? These can and do occur on freeways, and cannot always be predicted/avoided.

I either put on flashers and inch along, hoping nobody hits me from behind, or (if I can see well enough) inch my way over to the verge, stop the car, put on flashers and hope nobody hits me before the rain thins out. Neither "solution" is ideal...What does a SD car, with its faster reactions, do?

Autopilot generally does a little better than me. It can follow lane lines on a rain covered road better than I can. If the weather gets really bad, or in a few other situations, Autopilot will simply tell you to take over. It does have an intermediate step where it won't change lanes if it thinks its side vision is obstructed. We've had to drive manually through some rainstorms. It's likely that if you would slow because of poor visibility (rain, fog, snow), Autopilot won't want to drive. Snow covering the radar in the front bumper is one cause that doesn't affect humans as much as it does Autopilot. It has been getting better at this (over-the-air updates!), but as a heavily vision-based system I would think it would continue to have human-like limitations.

 

[Car-Guy]

Self driving cars are coming along at about the right time for me... Personally, I've always loved to drive. The more "spirited the driving is", the better I like it. But it's becoming clear to me in another 5 to 10 years (I'm estimating/hope, but it may be sooner) I won't be able to drive like I want anymore. When that day comes, I'll go with a self driving car. They should have most of the bugs worked out by then, so timing looks good for me.

 

[Animorph]

Self driving cars are coming along at about the right time for me... Personally, I've always loved to drive. The more "spirited the driving is", the better I like it. But it's becoming clear to me in another 5 to 10 years (I'm estimating/hope, but it may be sooner) I won't be able to drive like I want anymore. When that day comes, I'll go with a self driving car. They should have most of the bugs worked out by then, so timing looks good for me.

My thought as well, though maybe with robo-taxis rather than ownership. I think that'll be more cost efficient if we're not traveling very far.

 

[Midpack]

Self driving cars are coming along at about the right time for me... Personally, I've always loved to drive. The more "spirited the driving is", the better I like it. But it's becoming clear to me in another 5 to 10 years (I'm estimating/hope, but it may be sooner) I won't be able to drive like I want anymore. When that day comes, I'll go with a self driving car. They should have most of the bugs worked out by then, so timing looks good for me.

I used to think that would be ideal when I get too old to drive, but Uber and the like (or bonafide robotaxis) would be far more cost effective - at least where available like urban/suburban households? Owning autonomous cars may be cost prohibitive for most, remains to be seen. Who knows?

 

[Car-Guy]

I used to think that would be ideal when I get too old to drive, but Uber and the like would be far more cost effective - at least where available like urban/suburban households? Owning autonomous cars may be cost prohibitive for most, remains to be seen. Who knows?

That's something I might consider if I were move back to a "populated area" of the state. But that's not really an option here...