Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register

Tesla years ahead?

This site may earn commission on affiliate links.
Last edited:
Tesla is years ahead in lane keeping perhaps. I think full L5 autonomy is a pipe dream that the regulators will make a nightmare.

Blind spot detection and autopark are horrible with respect to the competition. I don't think AP 2.0 will change that anytime soon.

Tesla's feature set pushes the limits of what the sensors can do. Most of the functionality is in perma-beta and may never get dependable for normal use.

Just my opinion and I am sure I will be crushed with dislikes, but only adaptive cruise control and lane keeping work reliably enough to add value in my AP 1.0 Tesla. Everything else has promise, but is silicon snakeoil at the moment.
While I am certainly not going to give your post a dislike, I do feel that your conclusions are perhaps misinformed.

With the new hardware and the NVIDEA neural learning network the learning (improvement) curve improves exponentially. The combination of cameras, sensors and RADAR eliminate any blindspots. The car's awareness of its location in space imediately improves dramatically. I think the technology will prove to be at least twice as safe as a human within a year and 3-5 times safer in two. But...that's just my opinion.

Dan
 
Last edited:
  • Like
Reactions: dhrivnak and JeffK
While I am certainly not going to give your post a dislike, I do feel that your conclusions are perhaps misinformed.
...

You're missing his first, and I think most important point:

I think full L5 autonomy is a pipe dream that the regulators will make a nightmare.

I also think in reality "full autonomy" is a pipe dream. Sure, you can create the hardware that in theory is safer than a human 95% of the time, but it's that last 5% which is going to kill lots of people.

For the last two months whenever I'm driving on the highway, I am constantly taking note of just how many unpredictable events occur (humans, animals, road debris, construction,etc), as well as events well outside the range of view of sensors, that the car would never see, nor know how to react appropriately, which at the same time are trivial for human drivers. There are A LOT of those. Something no computer can do is think like a human and predict how other drivers are going to react to stimuli in front of them which any autonomous car will never see until it's too late.

One quick example, I was in the middle lane, a car directly to my left, and an 18-wheeler to my right. I could see that another car in front of the 18 wheeler was slowing down too fast to make an off-ramp, forcing the truck to quickly swerve into my lane. The only way for me to avoid a multi-vehicle accident was to slam on my brakes to slow down fast enough to let the truck into my lane. I could anticipate that the truck was going to need my lane before he even started to swerve. And he was depending on me realizing that so he didn't literally run over the little car in front of him. I didn't want to side-swipe the car to my left either. Luckily there was no car behind me so I could slow down fast enough to let the truck in. And I could see it happening ahead of me. All this happened in maybe 2 seconds. Such a simple almost everyday occurrence. No L5 car would know how to handle that with such subtle clues and events happening that far ahead of the L5 car in the other lanes.

Another classic example is an aluminum ladder laying in the middle of the road. It's just 2" off the road surface and essentially the same color as the roadway. Totally invisible to on-board sensors. A human would be able to see that object well in advance and change lanes, OR notice that other cars in front are swerving out of the way of the road hazzard. And L5 car would never see it, and go barrelling over the ladder causing significant property damage and likely losing control of the car and injury to the occupants. An L5 car would just see cars in front changing lanes.
 
You're missing his first, and I think most important point:
I also think in reality "full autonomy" is a pipe dream. Sure, you can create the hardware that in theory is safer than a human 95% of the time, but it's that last 5% which is going to kill lots of people.

For the last two months whenever I'm driving on the highway, I am constantly taking note of just how many unpredictable events occur (humans, animals, road debris, construction,etc), as well as events well outside the range of view of sensors, that the car would never see, nor know how to react appropriately, which at the same time are trivial for human drivers. There are A LOT of those. Something no computer can do is think like a human and predict how other drivers are going to react to stimuli in front of them which any autonomous car will never see until it's too late.

One quick example, I was in the middle lane, a car directly to my left, and an 18-wheeler to my right. I could see that another car in front of the 18 wheeler was slowing down too fast to make an off-ramp, forcing the truck to quickly swerve into my lane. The only way for me to avoid a multi-vehicle accident was to slam on my brakes to slow down fast enough to let the truck into my lane. I could anticipate that the truck was going to need my lane before he even started to swerve. And he was depending on me realizing that so he didn't literally run over the little car in front of him. I didn't want to side-swipe the car to my left either. Luckily there was no car behind me so I could slow down fast enough to let the truck in. And I could see it happening ahead of me. All this happened in maybe 2 seconds. Such a simple almost everyday occurrence. No L5 car would know how to handle that with such subtle clues and events happening that far ahead of the L5 car in the other lanes.

Another classic example is an aluminum ladder laying in the middle of the road. It's just 2" off the road surface and essentially the same color as the roadway. Totally invisible to on-board sensors. A human would be able to see that object well in advance and change lanes, OR notice that other cars in front are swerving out of the way of the road hazzard. And L5 car would never see it, and go barrelling over the ladder causing significant property damage and likely losing control of the car and injury to the occupants. An L5 car would just see cars in front changing lanes.

Can you say 100% of humans would react the way you did?

If the truck swerved into your lane from what you describe, you may have been driving in its blindspot. Would a level 5 car driving in a semi truck's blindspot?

I'd argue when sensing bodies in motion a computer can simulate what will happen far better and faster than a human. It might know there's an impending incident. It's up to manufactures to train the probable outcomes and accident avoidance measures other vehicles may take and how that will affect the driving path. (It's a software issue)

A 2 inch aluminum ladder is large enough to be seen by radar. Although I'm sure it would still hit it. Even so, that wouldn't stop it from being lvl 5. There's probably some poor human driver that's going to hit it as well. Anecdotally, I have several examples where people I know were following someone else too closely and didn't see an obstacle in time.

Lvl 5 doesn't make it accident proof, in addition, both of your examples are examples of human caused incidents.

If you want to prove lvl 5 autonomy isn't possible you'd need a driving situation where 100% of humans will perform it successfully but a computer will never do it correctly.
 
Can you say 100% of humans would react the way you did?

Of course not. That wasn't my point. 100% human drivers is what we have now. Some percentage of them would be able to handle the situation I described. I really think that 0% of L5 cars would be able to visualize and predict what transpired in my first example.

If the truck swerved into your lane from what you describe, you may have been driving in its blindspot. Would a level 5 car driving in a semi truck's blindspot?

I was not in his blind spot. I could see his mirrors and he could see me. He did put his turn signal on as he moved into my lane. For added detail, my car was positioned towards the rear left side of his trailer, well within the view of his side mirrors. Also, I can't say if any L5 cars will be programmed to detect what other vehicles blind spots are, and avoid those.

Lvl 5 doesn't make it accident proof, in addition, both of your examples are examples of human caused incidents.

Yes, actually that is my point. That it's the unpredictability of humans on the road that will make real L5 autonomy realistically impossible. Every time I'm on the highway, I observe several extremely minor events that would be impossible for L5 sensors to predict to avoid accidents or detect hazards (too far ahead, other lanes, obstacles in the way, etc) .

Like I said, you can cover 95% of routine highway driving and make it 500% safer than humans, I'm all for that. But it's that last 5% of human unpredictability on the road that L5 will never be able to handle and will fare far worse than humans in the exact same situation. Would it be a net reduction in the loss of life with L5 cars on the road? Probably. But I don't think we'll ever see the "turn on L5, go to sleep" type autonomy in our lifetime until such a time that 100% of the cars on the road are autonomous, perhaps in dedicated lanes free from humans. Even then there are still hazards like blown tires, big roadkills, that could mess even that up.
 
Yes, actually that is my point. That it's the unpredictability of humans on the road that will make real L5 autonomy realistically impossible. Every time I'm on the highway, I observe several extremely minor events that would be impossible for L5 sensors to predict to avoid accidents or detect hazards (too far ahead, other lanes, obstacles in the way, etc) .

Like I said, you can cover 95% of routine highway driving and make it 500% safer than humans, I'm all for that. But it's that last 5% of human unpredictability on the road that L5 will never be able to handle and will fare far worse than humans in the exact same situation. Would it be a net reduction in the loss of life with L5 cars on the road? Probably. But I don't think we'll ever see the "turn on L5, go to sleep" type autonomy in our lifetime until such a time that 100% of the cars on the road are autonomous, perhaps in dedicated lanes free from humans. Even then there are still hazards like blown tires, big roadkills, that could mess even that up.

I think level 5 self driving cars will continue to improve over time, but even humans being better at some things and computers being better at others, it won't make lvl 5 cars any less fully autonomous (even if they autonomously run into ladders :eek::confused:).

A 16 yr old human is still a fully autonomous driver and yet not nearly as safe as you or I. There are a huge number of adverse events that 16yr olds can't handle. As far as it being a pipe dream though... I think not, I think we'll be ok, not perfect, but "ok".
 
You're missing his first, and I think most important point:



I also think in reality "full autonomy" is a pipe dream. Sure, you can create the hardware that in theory is safer than a human 95% of the time, but it's that last 5% which is going to kill lots of people.

For the last two months whenever I'm driving on the highway, I am constantly taking note of just how many unpredictable events occur (humans, animals, road debris, construction,etc), as well as events well outside the range of view of sensors, that the car would never see, nor know how to react appropriately, which at the same time are trivial for human drivers. There are A LOT of those. Something no computer can do is think like a human and predict how other drivers are going to react to stimuli in front of them which any autonomous car will never see until it's too late.

One quick example, I was in the middle lane, a car directly to my left, and an 18-wheeler to my right. I could see that another car in front of the 18 wheeler was slowing down too fast to make an off-ramp, forcing the truck to quickly swerve into my lane. The only way for me to avoid a multi-vehicle accident was to slam on my brakes to slow down fast enough to let the truck into my lane. I could anticipate that the truck was going to need my lane before he even started to swerve. And he was depending on me realizing that so he didn't literally run over the little car in front of him. I didn't want to side-swipe the car to my left either. Luckily there was no car behind me so I could slow down fast enough to let the truck in. And I could see it happening ahead of me. All this happened in maybe 2 seconds. Such a simple almost everyday occurrence. No L5 car would know how to handle that with such subtle clues and events happening that far ahead of the L5 car in the other lanes.

Another classic example is an aluminum ladder laying in the middle of the road. It's just 2" off the road surface and essentially the same color as the roadway. Totally invisible to on-board sensors. A human would be able to see that object well in advance and change lanes, OR notice that other cars in front are swerving out of the way of the road hazzard. And L5 car would never see it, and go barrelling over the ladder causing significant property damage and likely losing control of the car and injury to the occupants. An L5 car would just see cars in front changing lanes.
Radar would see an aluminum ladder. Software is fully capable of handling the situation you described, I don't think the situation you described will be among the more difficult challenges, I immediately thought of a couple ways that it and similar situations could be successfully dealt with.

Everyone should keep in mind that those best able to answer the full autonomy question are those who are actually working on it. I'm not catching any pessimism from Musk, Sterling Anderson or other sources. I suspect a lot of pessimistic predictors will have egg on their face in the next few years.
 
If you want to prove lvl 5 autonomy isn't possible you'd need a driving situation where 100% of humans will perform it successfully but a computer will never do it correctly.
Excellent point. I cannot imagine such a situation, and given the rapid progress in autonomous driving in just the past few years, in the near future I think it will be clear that there is no such situation. That said, I predict that multiple people will post describing some fantastical one-in-a-trillion scenario that they will say all human drivers can figure out but computers cannot handle. As if computers were not improving by orders of magnitude and humans are never even momentarily distracted.
Like I said, you can cover 95% of routine highway driving and make it 500% safer than humans, I'm all for that. But it's that last 5% of human unpredictability on the road that L5 will never be able to handle and will fare far worse than humans in the exact same situation. Would it be a net reduction in the loss of life with L5 cars on the road? Probably
Every time I drive more than an hour or two I see at least one human driver near me do something highly unpredictable (and astonishingly stupid) that I did not anticipate and react to in an optimal manner because I was either distracted by something else or because -- news flash -- I do not have 360 degree vision or there was sun glare obscuring my vision (and I wear quality polarized sunglasses for daytime driving) or because I couldn't see clearly at night outside of the headlight beams or because I glanced at the driver display screen for a fraction of a second or a dozen other reasons.

The percentage of possible driving scenarios where the computer will be "far worse than humans in the exact same situation" will rapidly decrease to a small fraction of a percentage point over the next few years, at which point the data will clearly show that autonomous driving systems are far safer than humans. Will autonomous cars still have accidents? Of course they will. But their accident rate will be much less than that of human drivers.
 
I could see that another car in front of the 18 wheeler was slowing down too fast to make an off-ramp, forcing the truck to quickly swerve into my lane

Still a problem whilst non-autonomous (or, at least, human-driven but non-communicating) cars are still on the road, but I think this sort of thing will be solved? (or at least "improved"?) by vehicle to vehicle (V2V) comms. There are, I think, a number of opportunities for V2V to be used in this situation:

Car in front "I'm braking very heavily"

Truck "I'm bearing down on car in front and cannot stop in time so will need to swerve"

Another classic example is an aluminum ladder laying in the middle of the road.

I get alerts for that type of thing on Waze. When I first started getting "Car stopped on hard shoulder" (i.e. where there is a physical lane for that purpose, as distinct from "Car stopped on road") I thought it was chaff: that's what that lane is for, and the chance that I might need to take evasive action on such roads, in such a situation, is astonishingly small ... but now I appreciate those warnings as it gives me that tiny bit more Data in case it is ever needed. I never bother to enter car-stopped-on-shoulder into Waze (but I do if the car is stopped on the road, or a policeman is hiding in the hedge :) ), so I assume that the number of people prepared to log that, compared to the number of Waze drivers who pass it, must be small - and the number of Waze drivers compared to the total vehicles on the road must be tiny. But I still get those warnings - its rare to see a vehicle stopped and NOT get a warning.

So that is just for Data Points manually entered by another (Waze-using) driver on the road ... cars themselves detecting such obstructions and announcing them to other cars will be better, and more commonplace.

That doesn't stop the first vehicle ploughing headlong into the ladder though ... unless some other Tech can detect it, and also "announce" it - so we might still have an initial accident under L5? but not a pile-up.

Plus:

If the car-in-front was L5 it would not be having to brake so heavily
If the Truck was L5 it would be following at a safe-stopping distance - or drafting the vehicle in front if it was able to V2V with that vehicle to collaborate on any slowing-down requirements.
 
Don't overthink this. 25 years after consumer introduction OCR and voice recognition they are not completely reliable. A human can scan text accurately as well as listen. Why can't the computers figure it out 100% of the time . . . who knows. Just think of each one of those mistakes as a car accident.

I think driver assist will be perfected with machine learning techniques, but us meatbags will not be obsolete anytime soon.
 
  • Love
Reactions: HankLloydRight
Reality is that Tesla probably has the best path to get there no matter what the timeline is to actually enable L5 from a legal perspective. I can think of no better way to get self driving cars on roads faster than what Tesla is doing right now.

Think about it. Nobody else will have anywhere near the number of sensors on a production vehicle. Nobody else will have the number of miles driven with those sensors.

It is easy to say it won't work - but Tesla has the best realistic path to get there. In my opinion that makes them years ahead. Everyone else will have a fraction of the miles driven fed into their machine learning systems, a fraction of the data to prove to regulators it is safer, etc.

Regarding the system being perfect - looking at other drivers on their phones or paying half attention... yeah, I would rather trust 8 cameras, radar, and 360 degree ultrasonic sensors. Human drivers don't have that kind of visibility. It is about averages too - there are many more simple accidents than complicated ones. Airbags are the best comparison - they are automatic mechanical systems that save lives... but in some situations they cause loss of life. The good outweighs the bad by a high margin so every car has them.

My prediction is self driving features get enabled rapidly over the next 18 months culminating with self driving as long as a driver is behind the wheel to supervise (similar to the most recent video) and be legally responsible. It will stay that way until Tesla can prove to regulators that it is safer. Not sure what laws would prevent them from doing this - no different than autopilot 1.0 but is just handling more scenarios. Who knows when regulators will approve L5, but having that level of autonomy when you are in the driver's seat would be amazing.

It is easy to poke holes, but who else is putting the Nvidia high end GPU hardware, cameras, and investing in fleet machine learning like that? Nobody is close - the data Tesla will collect alone makes them ahead of the competition.

This will all happen faster than everyone realizes and Tesla owners will be a part of that future - helping fund the way towards safer driving.

When the future comes you can help it happen or fight against it. After the FSD video came out I went out and bought a Model X. Vote with your dollars or watch from the sidelines.

Just my two cents though.
 
Don't overthink this. 25 years after consumer introduction OCR and voice recognition they are not completely reliable. A human can scan text accurately as well as listen. Why can't the computers figure it out 100% of the time . . . who knows. Just think of each one of those mistakes as a car accident.
I think driver assist will be perfected with machine learning techniques, but us meatbags will not be obsolete anytime soon.

I believe you're overestimating the capabilities of humans haha. Humans cannot perform this task 100% of the time.
 
I think full L5 autonomy is a pipe dream that the regulators will make a nightmare.

This argument always surprises me. In order to accept this position, we have to assume that the NHTSA is lying to us since publically they are saying the exact opposite:

“Ninety-four percent of crashes on U.S. roadways are caused by a human choice or error,” NHTSA Administrator Mark Rosekind said in a statement. “We are moving forward on the safe deployment of automated technologies because of the enormous promise they hold to address the overwhelming majority of crashes and save lives.”

The NHTSA is pushing automakers to remove humans from driving as much as possible because of this 94%.

And this:

U.S. regulators are eager to accelerate development of robotic driving systems to reduce traffic fatalities, which topped 35,000 in 2015.

35,000 death last year alone! Compare that to US casualties from other causes the vast majority of which we are not on the brink of preventing, or at least significantly reducing, and you claim the NHTSA will make this "a nightmare"? That requires me to believe these regulators are truly evil people, and are conspiring against us, while lying to us with statements like this:

“In the 50 years of the U.S. department of transportation there has never been a moment like this, a moment where we can build a culture of safety as a new transportation technology emerges,” Transportation Secretary Anthony Foxx said at a Washington press conference on Tuesday. “Today we put forward the first federal policy on automated vehicles, the most comprehensive national automated vehicle policy that the world has ever seen. It is a first of its kind, taking us from the horseless carriage to the driverless car.”

The guidelines include a 15-point Safety Assessment governing the "safe design, development, testing and deployment of automated vehicles," the agency said late Monday.


That was 10 weeks ago. I say it's coming within months. It will start with a driver in the seat but the vehicle will do pretty much everything, as Tesla has shown in their videos. The NHTSA will not hold this back -- that makes no sense to me. They want it out and they want Tesla to lead the way (or anyone really but Tesla is the first to have full L5 hardware, including processors, on its vehicles) to show that we have the technology today to prevent a death that is happening right now -- due to human error! Motor vehicle crashes are the leading cause of death for U.S. teens. As a parent with teenage kids, it's my number one fear in life. That says a lot.
 
Last edited:
Don't overthink this. 25 years after consumer introduction OCR and voice recognition they are not completely reliable. A human can scan text accurately as well as listen. Why can't the computers figure it out 100% of the time . . . who knows. Just think of each one of those mistakes as a car accident.

I think driver assist will be perfected with machine learning techniques, but us meatbags will not be obsolete anytime soon.

http://www.nytimes.com/2016/12/14/magazine/the-great-ai-awakening.html?_r=0

With big data sets to train on, deep learning can produce remarkably rapid jumps in performance.

Translation is a bit like driving. Lots of little variations on the problem to learn before things get smooth.

This is the essence of Tesla's advantage. Everybody else is messing around with little test fleets. You can drive 100 cars for 1000 years or 100,000 cars for 1 year. Either approach can train the neural net but only one is a viable business plan.
 
  • Like
Reactions: JeffK and OBX John
In my eyes this is a perfect case of those that con see what things are going to be like within the time frame they will keep a new car and those that can't. I will typically drive a car until the wheels fall off. With the upgradeability of Tesla vehicles through the over the air software advances this becomes even more relevant.

Look at where autonomous driving was five years ago compared to where it is now. What is it going to look like in another five years? Keep in mind these advances are not linear. Like all things electronic these days, the advancements have become exponential. Sure, when I get my full autonomy Model 3 in a year of so the software won't yet be released to operate at full autonomy (probably because of legislative limits rather than product ability). I am totally ok with that. Where will it be over the life of the car though. I like my chances at full autonomy before I need/want another car. Add to that that I don't see anyone else coming close to Tesla, and yes I realize others will disagree, and for me it's a no-brainer.

Call me a fan boy if you must. It's a badge I will wear with pride. ;)

Dan
 
I want to believe . . . but the Koolaid hangover prevents me. That said, I have a couple M3s reserved and will certainly try it out, but I am certain that L5 will not work with the present sensor suite. Driver assist features maybe. Autopark, lane keep, blind spot detection and other things promised for AP 1.0 I would expect, but anything more than that is just fanboi buying silicon snake oil. . . . in my opinion.
 
  • Like
Reactions: HankLloydRight
I want to believe . . . but the Koolaid hangover prevents me. That said, I have a couple M3s reserved and will certainly try it out, but I am certain that L5 will not work with the present sensor suite. Driver assist features maybe. Autopark, lane keep, blind spot detection and other things promised for AP 1.0 I would expect, but anything more than that is just fanboi buying silicon snake oil. . . . in my opinion.
You have seen this, right?

Tesla Autopilot 2 0 Level 5 Autonomy Full Self Driving Hardware official video - YouTube