JeffK
Well-Known Member
Wow, you cited a study on self driving car accidents that primarily focused on development cars, because as you mentioned no one has a fully functioning self driving car.... Who funds these studies, I swear?!
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Wow, you cited a study on self driving car accidents that primarily focused on development cars, because as you mentioned no one has a fully functioning self driving car.... Who funds these studies, I swear?!
So someone actually cares about facts enough to follow the math. The figures that i used previously from old stats (about 3 accidents for about 3 trillion miles gives 1/1 million) Latest figures 2015 as you can see is 1 in 500k. I have also mentioned in the previous posts that about half of accidents being not reported. I have also made it known you need 1 in 100k for starters.
P.S
I don't understand why @JeffK would "like" this post.
Gotta love the "I gotta like any post that puts tesla in a good light and dislike any that dont"
Yet this post actually proves my very point.
This is not actually a problem because, well, science. And engineering. And standards.
stopcrazypp
Well-Known Member
I fail to see how it proves your point? You said the right benchmark to take for humans is 1 accident in 1 million (with today's driving trends of average 13k miles annually, that is 77 years of driving). As @MarcusMaximus points out, this does not make intuitive sense and doing the math he comes up with a 1 in 353,275 (27 years), which makes a lot more sense.
I tried to search for recent years where there are on the order of 3 million police-reported accidents per year and didn't find any. As far back as 1996, NHTSA said there were 6.3 million police-reported crashes. This is before factoring in the estimate for crashes not police-reported.
https://ntl.bts.gov/data/letter_am/gpra-96-pln.pdf
The numbers for miles didn't cross the 3 trillion mark until 2006. Also the 3 trillion miles statistic are highway miles only.
Table 1-35: U.S. Vehicle-Miles (Millions) | Bureau of Transportation Statistics
Not seeing how there will be a statistic that supports 1 in 1 million.
JeffK
Well-Known Member
In this case they'd probably do frequency hopping... the radar will sweep through certain frequencies.
For Tesla, the center frequency is 76.65 GHz with a bandwidth of 450 MHz. Natural sources of interference, such as from other cars is unlikely, however you can still jam this range if you wanted to intentionally blind it or even spoof objects which aren't really there. It's not foolproof.
You can do something similar with LiDAR and a raspberry pi
He didn't do that math i did the math. I saw a 3 million stats acouple days ago that i used to calculate with the 2016 miles driven. I went looking for it but i couldn't find the article.
So then i went looking for the latest numbers and supporting articles behind them. I did the math and came up with 1 in 500k. So i just posted the numbers instead of saying its 1 in 500k to see if someone will even attempt to calculate it.
My entire point is that when people throw out 10x, they don't realize what their talking about. and when people throw out 99.9% accuracy they are don't know what their talking about. This includes you.
for Tesla to be 10x, they need to be 1 in 5 million. Yet you ppl talk like that's a piece of cake
Don't try to twist this around. The only thing in question here is your fantasies that Tesla will have a Level 5 car thats 1 in 5 million that you can sleep in, in 2018.
Last edited:
In the FSD demo video on the website, the console display is not running Tesla's navigation system.
Tesla's navigation system can provide some interesting and entertaining routes, adjusting based on perceived traffic conditions, and often takes some creative detours.
The goal for FSD is to tell the car where you want it to go - and it will automatically determine the route and then drive to the destination (and for long distance travel, with the upgraded superchargers - automatically stop at an SC and recharge and resume).
A key milestone for the FSD project will be able to operate off a route generated by Tesla's navigation system. If the coast-to-coast demo uses navigation routing, that would indicate FSD is getting close, because the navigation system would like introduce some unexpected changes during a coast-to-coast route and if the FSD software can drive through those route changes, it would demonstrate the FSD software is able to handle those unexpected routes.
Though it seems more likely the demo will be on a carefully planned route - that has been tested several times (either with coast-to-coast runs, or in segments) before the demo, and when the demo trip is done, Tesla will have high confidence the demo can be successful.
That's typical for major "proof of concept" demos, and would be consistent with Musk's projection that FSD should be ready in 2 years.
Tesla's navigation system can provide some interesting and entertaining routes, adjusting based on perceived traffic conditions, and often takes some creative detours.
The goal for FSD is to tell the car where you want it to go - and it will automatically determine the route and then drive to the destination (and for long distance travel, with the upgraded superchargers - automatically stop at an SC and recharge and resume).
A key milestone for the FSD project will be able to operate off a route generated by Tesla's navigation system. If the coast-to-coast demo uses navigation routing, that would indicate FSD is getting close, because the navigation system would like introduce some unexpected changes during a coast-to-coast route and if the FSD software can drive through those route changes, it would demonstrate the FSD software is able to handle those unexpected routes.
Though it seems more likely the demo will be on a carefully planned route - that has been tested several times (either with coast-to-coast runs, or in segments) before the demo, and when the demo trip is done, Tesla will have high confidence the demo can be successful.
That's typical for major "proof of concept" demos, and would be consistent with Musk's projection that FSD should be ready in 2 years.
JeffK
Well-Known Member
Traffic fatalities in the US in 2016 were the highest in nearly a decade. The WHO reports that if the trend continues, traffic deaths will be the seventh leading cause of death by 2030. A person dies from a traffic accident every 25 seconds and mostly in poorer areas.
The leading causes of these accidents are human error. The goals of the industry, and regulators is to reduce this. In the end it really doesn't matter if Tesla's FSD is 2X or 10X better than a human.... Any better and you'd start saving lives.
The leading causes of these accidents are human error. The goals of the industry, and regulators is to reduce this. In the end it really doesn't matter if Tesla's FSD is 2X or 10X better than a human.... Any better and you'd start saving lives.
Reciprocity
Active Member
FSD will be enabled before regulators approve it and it will be a co-pilot that acts on your behalf in emergency situations like AEB does today. This is already done today in cars that will jerk you back into your lane if you drift. Of course the hard part will be knowing when the car should take control. This will also greatly help Tesla win regulatory approval, which I think will happen much quicker then people think. Because people do die every 25 seconds and this could save a lot of lives. My guess is that it will be enabled as a safety measure shortly after the Nov-Dec cross country demo.
SlyWombat
Member
Musk set expectations that it would be 2019 that you could sleep, 2 years from now "ish". [2017 Ted Talk] He also claimed that the cross country trip will not be canned, it could be be Seattle -> Miami instead of LA->NYC.
Nothing Musk says matters to him. I can see his point there. It's all fluff until it's proven.
stopcrazypp
Well-Known Member
I believe some researchers in China already did radar jamming on a Tesla as a demo. Of course, for the most part this requires expensive equipment, so unlikely to be as practical as an attack other than for very determined attackers.
They also jammed the ultrasonic sensors and that is much cheaper to implement ($40 in hardware).
Hackers Fool Tesla S’s Autopilot to Hide and Spoof Obstacles
JeffK
Well-Known Member
JeffK
Well-Known Member
Total, here's a chart for the US:
Total miles driving only increased 3%
It's 12.4 fatalities per 100,000 people.
Increased vehicle safety initially caused a long term decline, but I'm going to go ahead and take a guess that this rise is due to more distracted driving (smartphones and such).
Last edited:
Thanks. Yes, increasing deaths are bad. But unless it's the death-rate-per-mile-driven number that's increasing, I think it's hard to draw any conclusions. 2014-2016 have also been a time of economic recovery, during which one would presume people are driving more miles - and thus total deaths would logically rise even if the death-rate-per-mile is not rising. Etc. etc. Not debating you but I've seen this thrown around quite a bit - the fact that total highway deaths are rising - but I've never seen it framed as the death rate per mile - which is the important stat (to me).Total, here's a chart for the US:
JeffK
Well-Known Member
Added... since 2015 3% increase in miles driven vs 6% increase in deaths. 2016 was one of the deadliest years in a decade.
stopcrazypp
Well-Known Member
Ok whatever, at least we agree 1 in 1 million is wrong.
I believe I never previously mentioned 99.9% or 10x, so not sure if you confused me with someone else.
I don't believe anyone claims it would be a piece of cake, just that they believe Tesla can eventually get there.
The only reference to 10x I can find is the general goal posted on the Master Plan, but that does not have a specific timeframe. I don't believe Elon ever said they plan to reach 10x in ~2 years, only that they plan to have FSD in consumer hands in around that time. So the FSD at the point could be 2x or 1.5x (roughly 1 accident in 700k/500k miles by math done previously), doesn't have to be 10x.
This is probably looking at it a bit too logically(in practice, humans don't tend to be logical about things like this) but: Hell, even 1.0x would be no worse than having your own personal chauffeur.
Reciprocity
Active Member
Some would have you believe that they will never achieve better then a drunken toddler driving you around. Accidents will happen and validation should show that many accidents will be averted and most will be less damaging because the machines have a much higher reaction time and do not get distracted no matter how complicated the situation. I will take 1x with 0x distractions. I believe that much of the increase in deaths we are seeing recently can be attributable to distracted driving.
Similar threads
