tspear
En-Route
- Joined
- Dec 10, 2010
- Messages
- 3,587
- Display Name
Display name:
Timothy
Something seems off on those numbers. Rand was mixing fatal rates and accident rates as one and the same. They are not.
Tim
Sent from my SM-J737T using Tapatalk
Nearly half of Americans don't want a self-driving car
- Thread starter genna
- Start date
Tim
Sent from my SM-J737T using Tapatalk
I want a self washing car.
UH-60V Pilot:
Pardon my skepticism.. I have watched videos of Tesla Autopilot on good roads, and it alternates between granny mode (taking curves at speeds which will get you honked at) and teenager mode (does great 99% of the time, the other 1% you have to take over so it doesn't kill you).
You and me both!
- Joined
- Jul 3, 2012
- Messages
- 15,597
- Display Name
Display name:
Velocity173
UH-60V Pilot:
Coworker flew S92s in the GoM. Pretty much what he did in the left seat. FMS typer / radio operator and rarely touched the controls. Bored out of his mind and didn’t like the schedule so he left.
Juliet Hotel
En-Route
- Joined
- Jul 21, 2018
- Messages
- 2,634
- Display Name
Display name:
Juliet Hotel
So let me get this straight. The claim is we will soon be able to go from point a to point b without having to take over. Not that we can do that now, but that we will be able to do it soon. You're saying we can now go 99% of the way from A to B without having to take over, but 1% of the time, we still have to take over. But you don't believe we're almost to the point where it will be 100%. If 99% isn't almost there, what is?
If that 1% will kill you, you can never leave it alone. If that's the case, why bother?
One percent may not sound high, but it is. It's huge. My commute is short, and I still drive over 3 hours a week just for commuting, even on a week of good traffic. My annual driving is low, but I'm still probably averaging 6 hours a week of driving. That would mean at 1% that the car is trying to kill me for just over 7 minutes every week.
Here and there. You never know when. Thanks, but no thanks. The other drivers are bad enough occasionally trying to swap paint.I'm not a Luddite. I'm in IT (management now, but started doing software development and did a bunch during the DotCom boom).
It would be great to have a self driving car once I'm in my 80's and 90's so I have good transportation. It might be nice now when traffic sucks, so I can let the car drive while I do something other than get frustrated. The rest of the time I don't care as I like driving and the systems just aren't good enough yet.
tspear
En-Route
- Joined
- Dec 10, 2010
- Messages
- 3,587
- Display Name
Display name:
Timothy
I am in IT. In a less complicated transaction software solution, here was some rough budget numbers for a contract a few years ago:
1. 90% up time. Single server, quick/dirty software stack. $2.5 Million software development for the system; $50K HW.
2. 99% up time. Two servers (one application and one database) software stack $3 Million software development; $100K HW.
3. 99.9% up time. Four servers (two application and two database) plus dual load balancers, enterprise grade software stack $5 Million; $400K HW
4. 99.99% up time. Eight servers, two sites (one live one backup), BGP Routing, HA routers, networks, enterprise grade software stack $5 Million; $1 Million HW
5. 99.999% up time. Eight servers, two sites (both live), HA DNS, HA routers, networks, enterprise grade software stack $7.5 Million; $5 Million HW
And that is only getting to five nines; for a distributed transaction system. No AI, no complex rules.... Just complex accounting (synced double entry across multiple journals).1
In software, the 80/20 I think almost becomes logarithmic. The closer you try and push the 80% to 100% the cost goes up exponentially for the next increment in decimal.
Tim
Juliet Hotel
En-Route
- Joined
- Jul 21, 2018
- Messages
- 2,634
- Display Name
Display name:
Juliet Hotel
None of that addresses the comment. We are at 99% right now. You're obviously looking for 100% and you're not alone. We're not there yet so don't buy one yet. Simple.
The comment was regarding the claim that we are now at 99% and will be at 100% soon. Some don't believe that's true but I do. No one is claiming we are at 100% now. No one is claiming being at 99% means you can drive as though its 100%.
Juliet Hotel
En-Route
- Joined
- Jul 21, 2018
- Messages
- 2,634
- Display Name
Display name:
Juliet Hotel
Not apples to apples. More like apples to broccoli. But your point is valid in that we will likely never get to 100%. If we get to 99.99999% and fatal accidents become a fraction of what they are now, isn't that still worth while?
tspear
En-Route
- Joined
- Dec 10, 2010
- Messages
- 3,587
- Display Name
Display name:
Timothy
Getting to 100% is I think impossible. But my point was the cost/complexity to increase the efficiency gets harder the closer you get to 100%.
And that is in a relatively simple transaction system, not something as nearly complicated as an AV.
Last point, I doubt they are even close to 99%.
Tim
Juliet Hotel
En-Route
- Joined
- Jul 21, 2018
- Messages
- 2,634
- Display Name
Display name:
Juliet Hotel
We're not talking about efficiency or up time, we're talking about the machine being able to identify and handle situations that are encountered while driving. You don't need 10 redundant servers for that. I don't want to downplay the size of the task at hand, its huge. But its also 100% doable and I don't think it will take nearly as long as some people claim.
Also as we said, you don't need to get to 100%. In 2016 there were 1.16 deaths per 100M miles traveled. We don't need to get to 0 fatals and probably won't. But if this technology gets us from 1.16 deaths per 100M to .16 per 100M, that sounds like a win to me. The real hard part in all this is the public perception part. Because people are pretty dumb collectively. And if we look at a world where the cars drive themselves and we see .16 deaths per 100M instead of 1.16, it stands to reason that almost all of those .16 deaths will be accidents that would not have happened in a manually driven vehicle i.e. deaths caused by the technology itself. We'll be better off overall but the perception will be that the technology makes it worse.
How many times a week do you hear someone in the office say 'computers were supposed to make things better...' as a complaint because the machine isn't doing exactly what the person expects it to? And yet they're working for a company that is able to more productive and more profitable with far fewer staff doing far less work than was ever imaginable before the PC existed. It'll be the same with these cars. We'll be far safer but most people will swear up and down they're more dangerous and we were safer when we drove ourselves.
Also as we said, you don't need to get to 100%. In 2016 there were 1.16 deaths per 100M miles traveled. We don't need to get to 0 fatals and probably won't. But if this technology gets us from 1.16 deaths per 100M to .16 per 100M, that sounds like a win to me. The real hard part in all this is the public perception part. Because people are pretty dumb collectively. And if we look at a world where the cars drive themselves and we see .16 deaths per 100M instead of 1.16, it stands to reason that almost all of those .16 deaths will be accidents that would not have happened in a manually driven vehicle i.e. deaths caused by the technology itself. We'll be better off overall but the perception will be that the technology makes it worse.
How many times a week do you hear someone in the office say 'computers were supposed to make things better...' as a complaint because the machine isn't doing exactly what the person expects it to? And yet they're working for a company that is able to more productive and more profitable with far fewer staff doing far less work than was ever imaginable before the PC existed. It'll be the same with these cars. We'll be far safer but most people will swear up and down they're more dangerous and we were safer when we drove ourselves.
I think it did, so we're seeing it differently as I was commenting about directly about it. From what I've seen and read we're not even at 99% yet. Certainly not for commercially viable vehicles. That's not far off though.
I never said that I felt one was forced to buy. I wouldn't be surprised if sometime in my lifetime if it because a requirement to have a self driving car; "think of the children!"
I don't think we're at 99%. Maybe 99% for a constrained set of driving, but not all driving. We're not far off, but I can't go down to the store and buy one yet.
No, it's doesn't need to be 100%, and probably never will be. Something will break/go-wrong, but it will be really close to 100%. The big thing is to make sure that whatever is below 100% is minor. Between more cautious driving and safety systems (airbags, crumple zones, etc), if it gets us to a fender-bender when something crazy happens, then it's good. Like a golf instructor said to me one time, "it's not about making your good shots better, but about making your bad shots less bad". If an accident happens, and they will happen, if you can walk away from it it was a good accident. If you can drive away from it it was a great accident.
The near 100% will happen, and sometime soon.
When self-driving cars show a better safety record than human drivers in the wide variety of terrain, weather, road conditions, etc. that human drivers contend with, then I will get excited. Until then, not so much.
Will we be able to do it? Most likely. Soon? I doubt it. Look at some of the visualizations of what the Tesla cameras "see". It is a (relatively) primitive system heavily reliant on painted lines. It has problems with faded lines, rocky roadsides, and sharp curves. At present there is little in the way of big picture thinking which we as humans do all the time. That's why there was the sad case of the Tesla happily slamming into the concrete barrier without braking.
The current Onramp to Offramp system works pretty well, but still has problems knowing when to merge into traffic, camping in the left lane, taking offramps too slow, etc. And this is on highly predictable freeways in good condition. These can be fixed, but it shows how far there is to go. This is why I'm highly skeptical about a Tesla going from my driveway to work anytime soon. I'd be happy to be proven wrong.
He didn’t actually say 99%. He said when it wasn’t failing in a safe but annoying manner (granny mode) it was failing in an unsafe manner 1% of the time. Sure, we’d all agree self driving cars would be safe if they went max 4 mph. If, every car out there was a self driving car.
Watch this video (posted last month) starting at 3:22. Without a takeover, there would have been a dead motorcyclist.
I don't know how long it will be before you don't have to watch the Autopilot like a hawk, but something tells me it will be awhile.
I don't know how long it will be before you don't have to watch the Autopilot like a hawk, but something tells me it will be awhile.
And the closer it gets to working the more deaths due to inattentive drivers.
And looking at photos of the approach to that particular concrete barrier makes it easy to see how the automation could have gotten confused.
azblackbird
Pattern Altitude
Maybe on a closed course they're at 99%. In reality... not even close to 50%.
Juliet Hotel
En-Route
- Joined
- Jul 21, 2018
- Messages
- 2,634
- Display Name
Display name:
Juliet Hotel
We were at 0% of the way there about 10 years ago. We've come a long way since then. I don't think it'll be another 100 years before we get it right. I don't think it'll be 50. My guess is within 10 years every car being built has self driving ability that is significantly more advanced than what is available today. Within 20 years full autonomous is a reality. 20 years isn't that long in the grand scheme. That's about 2 vehicles from now for me.
This sounds like a realistic timeline to me. Tesla talks like it's going to happen in 2 years though, not 20.
Arguably cruise control was > 0% and it's been around since the 60's.
genna
Pattern Altitude
- Joined
- Feb 5, 2015
- Messages
- 1,721
- Display Name
Display name:
ТУ-104
https://gizmodo.com/uber-employee-warned-self-driving-cars-are-routinely-in-1831019048
Uber Employee Warned Self-Driving Cars 'Are Routinely in Accidents' Days Before Fatal Crash: Report
Uber Employee Warned Self-Driving Cars 'Are Routinely in Accidents' Days Before Fatal Crash: Report
Juliet Hotel
En-Route
- Joined
- Jul 21, 2018
- Messages
- 2,634
- Display Name
Display name:
Juliet Hotel
I recall hearing a story about how Google gave several of their top tier execs cars equipped with what was then their currrent level of self-driving technology. All of the execs had to sign agreements saying they understood that the cars were equipped with driver-facing cameras and if they were observed texting or looking at their phone while the car was in self-driving mode, they would be fired. IIRC it took less than 7 days for all of them to be caught using their phones while the car drove itself. I'm guessing Google didn't follow through on the firings but they probably took the cars away.
deonb
Pattern Altitude
- Joined
- Aug 17, 2015
- Messages
- 2,266
- Display Name
Display name:
deonb
We’re at 99% easy. How long does a mistake last? 10 seconds?
In that case 99% is 10 seconds out of 16 minutes. I regularly drive a Model 3 on AutoPilot - It’s better than that already.
I have to take over maybe twice per hour. And I do think that’s nowhere near good enough. But it’s better than 99%.
And it’s getting better all the time. Dramatic difference between the car from a year ago and now. When it works it’s WAY WAY! better than I am to keep the car in the lane and keep follow distances correct. It does screw up, but it is actually predictable where it will screw up. It’s not completely like it will try and kill you at random - you can anticipate what it will do e.g in lanes with faded markings - just takes practice. As long as you don’t think it can do EVERYTHING it’s a great add on. But it’s not autonomous in any sense of the word.
Last edited:
Yep. Me too. We switched seats every day, so everyone got stick time and radar/radio monkey time:
Attachments
- Joined
- Jul 3, 2012
- Messages
- 15,597
- Display Name
Display name:
Velocity173
Lol! Yeah he was PHI as well. Have a few friends that went GoM. Thought about it but ultimately the schedule and over water ops just isn’t for me. Don’t think I’d enjoy the dunker either.
By the way, as I was driving home from the airport today, I noticed that the space between the white lines leading up to the killer concrete barrier has now been painted with white chevrons, presumably to make it easier for Tesla autopilots to tell that it's not a lane.
deonb
Pattern Altitude
- Joined
- Aug 17, 2015
- Messages
- 2,266
- Display Name
Display name:
deonb
Actually that barrier has always been a hotbed for accidents - in fact the reason why the Tesla accident was fatal is because the city didn’t reset the crash barrier from an accident earlier in the week. So makes sense for everyone to make it more clear.
flyingcheesehead
Touchdown! Greaser!
Straight cruise control isn't self driving at all, it's merely holding speed. You could drop a brick on the pedal 100 years ago, but that wasn't self driving either.
Really, the firs iterations of self-driving have been the single axis version of Autopilot, known as adaptive cruise control. Now, that has been around for 20 years, and is still on fewer cars than it's not. Not exactly a shining example of the market moving things forward.
Google's cars didn't have human controls - The human couldn't take over. I can still see them wanting to have the drivers paying attention, but it's not like they could do anything.
They also were a bit too polite - Shortly before they shut the project down, a Google car sat at a four-way stop for a half hour because it was programmed to always let the other guy go first. So, as long as there was another guy, the Google car just sat there.
Was the Tesla car the first to go by before it was reset? Or did a bunch of human drivers miraculously make it also without an accident?
Juliet Hotel
En-Route
- Joined
- Jul 21, 2018
- Messages
- 2,634
- Display Name
Display name:
Juliet Hotel
Better question: Was the accident Tesla the first Tesla to go by it in autopilot mode?
tspear
En-Route
- Joined
- Dec 10, 2010
- Messages
- 3,587
- Display Name
Display name:
Timothy
Missing the point. It was a point of relative complexity. An HA transactional system is pretty straight forward. An AV is not. Getting the last few basis points in performance on what is a fairly simplistic system is incredibly hard. I would not want to even estimate the level of effort for such a complex system as an AV.
Tim
genna
Pattern Altitude
- Joined
- Feb 5, 2015
- Messages
- 1,721
- Display Name
Display name:
ТУ-104
People here keep saying that it doesn't need to be 100%. That a significant improvement over human drivers is enough. I think this is a false premise. When a human makes a mistake, it's understandable. We all do. We know who to blame and the matter is resolved with (relatively) small financial consequences. If Toyota makes a Level 5 autonomous car that eventually gets into a few deadly accidents, it won't matter if a basic Corolla/Human combination kills 10 times more people than a Toyota AV. Passengers are going to imagine that this is a deadly vehicle that can kill them and they cannot stop it from happening. It may be nonsense, but this is how human mind works. This is how mass hysteria works. It's not a numbers game. It's a TRUST game. Most drivers trust themselves(even if it's a misplaced trust). Many people may even trust AV cars. But it doesn't take long for that trust to evaporate with just a few crashes.
We don't really need to wait for AV to see this in action. Toyota was in serious legal trouble for months over perception of unintended acceleration. Audi almost folded for same reasons. Fleets of commercial aircraft have been grounded while investigating some potential flaws that MAY cause an accident.
At the very least, this makes everything involved very expensive.
We don't really need to wait for AV to see this in action. Toyota was in serious legal trouble for months over perception of unintended acceleration. Audi almost folded for same reasons. Fleets of commercial aircraft have been grounded while investigating some potential flaws that MAY cause an accident.
At the very least, this makes everything involved very expensive.
Last edited:
azblackbird
Pattern Altitude
I can't wait for AV's to become more prevalent. With today's "sue happy" society, the money made shorting AV stocks will be like shooting fish in a barrel.
Good point. Hopefully, the chevrons will help.
Another challenge that will need to be met:
Self-driving cars must be experts on ridiculously specific road rules
Self-driving cars must be experts on ridiculously specific road rules