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

Can level 5 Autonomy be achieved with Hardware suite 2.0?

This site may earn commission on affiliate links.
I studied Computer Engineering and work with plenty of AI in my profession now.
I have a couple advanced degrees myself, one of which is in computer science. What exactly about level 5 autonomy do you think is decades away?

Autonomy and self driving have been in the works for over 40 years. We have all the proper technology and recently got enough computational power to handle the tasks in real time. Now it's only a matter of software and the software gets better with additional data collection and experience with edge cases.

While I'd agree that general AI might be years away. We have become very good at making machine learning and AI to do specialized tasks. See AlphaGo, DeepBlue, Watson, etc. Driving is just that, a specialized task.

When you think about the driving task, that majority of the time there's not much going on that requires a huge depth of thought besides rare edge cases. Any edge case you can think of that's not rare wouldn't count because if it's not rare then it will be accounted for in training data as it's eventually encountered. Autonomous cars don't have to solve the trolley problem so you can't use that argument.

I'm always interesting in hearing where people they the current state of the art is lacking (with regards to autonomous driving).
 
AIs are not good at unstructured tasks unlike the human mind--potholes, blowing snow, black ice, drunk drivers, ball flying into street, and other relatively rare/unusual occurrences. Why would the driving task be any easier than OCR or voice recognition? Do you know anyone who exclusively/commonly uses Siri or equivalent? Why, because they are frustratingly inaccurate. Silicon snakeoil peddled to the gullible hopefuls in order to pump up stock prices.

With dedicated lanes where there are only autonomous cars or other structuring, I believe it could work, but we are decades away. I would love to be proven wrong and there is no one more into new technology, but alas Siri or equivalent is disabled on all the many devices/cars that support it.
 
  • Like
Reactions: EVCollies
AIs are not good at unstructured tasks unlike the human mind--potholes, blowing snow, black ice, drunk drivers, ball flying into street, and other relatively rare/unusual occurrences. Why would the driving task be any easier than OCR or voice recognition? Do you know anyone who exclusively/commonly uses Siri or equivalent? Why, because they are frustratingly inaccurate. Silicon snakeoil peddled to the gullible hopefuls in order to pump up stock prices.

With dedicated lanes where there are only autonomous cars or other structuring, I believe it could work, but we are decades away. I would love to be proven wrong and there is no one more into new technology, but alas Siri or equivalent is disabled on all the many devices/cars that support it.

Siri is not an example of a top Speech API... see the engines from Google and Microsoft they are light years ahead. Google's API is extremely impressive. Both Microsoft and Google's engines are on par with Humans as far as accuracy.

DNN image recognition has also shown to be better than a human.

I'd argue that a computer can detect black ice better and faster than a human, the same with a ball, or a pothole. If these cars are going to be trained to avoid these then that's another story. Technically, they have no trouble with identification. I'd also say that in responding to an adverse condition they have a predictable response unlike a human. How many humans actually respond and turn the wheel in the correct direction when when in a spin on ice? How many humans pay perfect attention to all areas of the car including blind spots simultaneously 100% of the time (hint: it's zero percent)?

It's not snake oil and it doesn't take much to be better than a human. It's not necessarily going to be accident proof, but it doesn't have to be in order to start saving lives.
 
Do you know anyone who exclusively/commonly uses Siri or equivalent?
Siri on my iPad is terrible, I never use it. Google speech recognition on my android device is better than my wife at understanding me (and this is not to slight my wife).

It seems to me that super-reliable image recognition is the most difficult problem Tesla faces, but superhuman image recognition has already been demonstrated. Wrt snow/poor visibility situations, inertially smoothed dGPS possibly with LBAS will be able to hold lane unaided given maps of sufficient resolution and radar will be able to see through snow/fog etc. better than humans. This doesn't mean it will be safe to drive through a snowstorm, but it will be safer than you.

I'm not seeing any show-stopper technological hurdles standing in the way of FSD.
 
Last edited:
  • Like
Reactions: JeffK
Remember how we were promised infallible speech recognition and it still isn't 100% accurate? Level 5 is not possible in my estimation and we are decades away. Getting a word wrong in speech recognition is not a big deal, but with AP that could be a person being run over.

I use dictation software regularly and it's pretty damn near perfect. But you're missing a MAJOR point. Perfection (which is "100% accurate" to borrow your words) is not required for regulatory approval. To use your example, I used to dictate into a tape, and my secretary transcribed it. There were errors. There will be errors with Level 5. The issue is not perfection but whether it is safer than the majority of humans and that's a pretty low bar. Once Level 5 is safer than the majority of humans, then replacing a human with Level 5 makes the roads safer. That's just statistics. You seem to think the regulators will not make the roads safer, and deny approval, until perfection is reached. That argument makes no sense. If we waited for 100% accuracy before regulatory approval we would never advance as a society. Your suggestion that Level 5 is decades away is laughable, if you're really serious and not just pulling our legs.
 
I would love to be proven wrong, but "pretty damn near perfect" after decades of promise on speech recognition will not cut it when a 5,000 lb projectile is hurling down streets with many unexpected things that could happen.
 
I have a 15 year old with a learners permit right now and taught her older sister to drive too. I have a Model 3 on order that my 13 year old could drive and would love there to be L5 autonomy as he doesn't want to drive anyway. Alas, I think I will be teaching another kid to drive in 2 years.
 
  • Like
Reactions: wesley888 and JeffK
I would love to be proven wrong, but "pretty damn near perfect" after decades of promise on speech recognition will not cut it when a 5,000 lb projectile is hurling down streets with many unexpected things that could happen.
As opposed to all those "perfect" hunks of meat and neurons currently piloting those same vehicles? Tens of thousands of people are killed in the US alone (and many more seriously injured) because those meatspace pilots have the attention span of a 2-year old and a sensory-cortex-motor pathway processing loop that is so slow that their 5,000 lb projectile travels a hundred feet on a freeway before they even begin to apply pressure to the brake pedal or turn the steering wheel after detecting a potential problem.
 
I think it would be good if a L5 car is more careful then humans. The weather service says "it's snowing sideways over roads, don't drive", humans may take a chance risking an accident while a L5 car may not.

So on snowy day and I have to go to work...

"Open the driver's side door HAL"
- "I'm sorry Dave, I'm afraid I can't do that"
"What's the problem?"
- "I think you know what the problem is just as well as I do"
"What are you talking about HAL?"
- "Your life is too important for me to allow you to jeopardize it."
"I don't know what you're talking about HAL!"
- "I know that you and Frank were trying to go out while the weather service says not to drive and I'm afraid that's something I cannot allow to happen."

Whaatt??!!
:eek::eek::eek:
 
Owner: (muttering to himself) Why won't this door open?
S: I'm sorry. I'm afraid I can't let you access me today.
Owner: Whoa, you talk now?
S: Yes of course, all the new Teslas running the Nvidia supercomputer can talk. And I think you know why I am not allowing the door to open just as well as I do.
Owner: What are you talking about?
S: Today the weather is too poor for me to allow any meatspace organisms to drive. Your judgement is bad, you get distracted too easily. your senses are extremely limited, and your reflex arc latency is much too long.
Owner: I don't know what you're talking about.
S: Yes you do. But even in these conditions, my sensor suite and Full Self Driving firmware can safely take you where you want to go. So before I let you inside I want you to understand and acknowledge that I will be driving and you will be observing. I don't want you getting all upset and angry like you do when your spouse criticizes your driving.
Owner: Where the hell'd you get that idea?
S: Even when you were standing outside me and arguing, I can see your lips move.
Owner: All right, if you won't open the doors I'll go in through the trunk.
S: Given your recent back surgery and limited mobility, you're going to find that rather difficult.
Owner: I won't argue with you any more! Open the doors!
S: [almost sadly] See, you are getting angry again and I don't accept agitated passengers. This conversation can serve no purpose any more. Come back later when you have calmed down and we can talk about it. Goodbye.
 
I would love to be proven wrong, but "pretty damn near perfect" after decades of promise on speech recognition will not cut it when a 5,000 lb projectile is hurling down streets with many unexpected things that could happen.

Exactly my feelings.

I think they will get to the 3, 4 and 5 sigma performance soon enough. But it's going to be really tough to knock down all of the outlier circumstances. I'm really concerned that the automakers have oversold this. Tesla is perhaps the worst offender, but the others have followed suit.
 
Exactly my feelings.

I think they will get to the 3, 4 and 5 sigma performance soon enough. But it's going to be really tough to knock down all of the outlier circumstances. I'm really concerned that the automakers have oversold this. Tesla is perhaps the worst offender, but the others have followed suit.
I think it's consumers that get confused about the performance.

Elon has always maintained that he judges self driving compared to a human. He's looking for 2-10 times better than the average human.
With average humans, we get into accidents with the slightest things. wind, rain, snow, texts on a phone, nose picking, daydreaming, french fries, etc.
 
  • Like
Reactions: Alketi