powertoold
Active Member
It looks like the current hardware is capable of at least a geofenced robotaxi service better than average human safety. Tesla has the data and software to geofence routes based on navigation NN confidence.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
FSD rewrite will go out on Oct 20 to limited beta
- Thread starter diplomat33
- Start date
It looks like the current hardware is capable of at least a geofenced robotaxi service better than average human safety. Tesla has the data and software to geofence routes based on navigation NN confidence.
Your last sentence reminded me that only a couple of months ago I had serious doubts that hw3 had the horsepower for fsd. I have slightly less serious doubts now. I still think they could use more cross traffic and rear coverage on their sensor suite, but I guess it generally seems that it works to some minimum level. Would be interesting to know what % of the processing power the current beta is using?
Elon tweeted that 2020.40.8.12 should improve the disengagement rate by up to 33%:
Elon: "We measure this primarily in intervention probability. This update addressed several issues, resulting in perhaps ~1/3 fewer interventions. Many of the improvements consist of fixing silly bugs vs grand eureka moments. True for most beta releases in my experience.”
Musk says that the automaker plans to push new updates every 5 to 10 days leading to the wider release of the Full Self-Driving beta in the US later this year.
The CEO warns that errors will never reach zero, but Tesla’s hope is that they will become low enough to be safer than humans:
Elon: “Faults will never be zero, but at some point the probability of a fault will be far lower than that of the average human.”
Tesla pushes new Full Self-Driving Beta update, Elon Musk says could slash interventions by a third - Electrek
Elon: "We measure this primarily in intervention probability. This update addressed several issues, resulting in perhaps ~1/3 fewer interventions. Many of the improvements consist of fixing silly bugs vs grand eureka moments. True for most beta releases in my experience.”
Musk says that the automaker plans to push new updates every 5 to 10 days leading to the wider release of the Full Self-Driving beta in the US later this year.
The CEO warns that errors will never reach zero, but Tesla’s hope is that they will become low enough to be safer than humans:
Elon: “Faults will never be zero, but at some point the probability of a fault will be far lower than that of the average human.”
Tesla pushes new Full Self-Driving Beta update, Elon Musk says could slash interventions by a third - Electrek
powertoold
Active Member
That's nothing new. Even at 10x human safety, that's still a lot of accidents.
Daniel in SD
(supervised)
I'd be very curious to see what this number is right now. Looks like about 5 miles?
Hopefully an obsessive person with a moderately difficult commute will measure it over time (like this: Navigate On Autopilot: automatic lane change results)
Of course once it gets a little bit better (>1000 miles?) this metric is no longer useful, people will disengage in situations where the car would do fine because they don't want to risk an accident (a good thing!).
Yea I agree.
I think it's because tesla has always secretly planned on a subscription service. They created a bunch of cars that even when sold used they can get more money if someone subscribes. The cameras and sensors prob aren't as expensive as we think.
It even works with the heated back seats. Let's say it costs them $15 in seat parts to put it in every car. They charge people $300 to turn it on. What are the odds that in 15years and a few owners a few people hit that upgrade button.
I wouldn't be surprised if tesla starts turning fsd and other upgrades off when they take their own cars back on trade. They can sell it twice.
I think they already do that. I have no problem with that. Once they buy it they can modify it and resell it anyway they want.
jmho.
I think this is going to be a solution. It's not going to be able to drive everywhere from the start. It's also not going to be legal everywhere.
Last edited:
willow_hiller
Well-Known Member
I could imagine this tying into Tesla Insurance in a funny way. At first FSD will be safer than some drivers, but some good drivers will be safer than FSD. They could programmatically rate your driving ability, and if you're noticeably worse than the computer, an offer could pop up on the screen "Hey, you're not driving so well today. I'll give you a $5 discount off your insurance premium this month if you let me take over."
It makes sense if they are. I think they should do it but I also think they should deactivate all options I purchased when I sell and make them part of my Tesla account. It would force me to be a returning customer forever and I'd happily buy fsd for 10k because I think it will happen... Just not in a useful version while I own this version of model 3.
Elon (for once) is right in the pattern of releases. The first release was date-driven, and no doubt some bugs deemed minor/acceptable didn't make the cut. That's why we saw a very quick turn-around of the first update, it was to get those bug fixes out. After that things will slow a bit as the team take a bit of a break (I bet they were working nights to get the Oct 20th release out). Then the cadence will settle into a pattern of releases, probably one every 2 weeks or so, with less spectacular, but significant, fixes in each release.
The UI update sneaking in was interesting, but as that's probably a different team to the FSD guys, they probably were on a different release cycle.
I'm skeptical about the end of year wider release, I think late Jan or early Feb seems more realistic, though of course that's just a guess.
DanCar
Active Member
Why not release it now? People have the option to enable it or not. It will make people accustom to paying attention. If it gets better, people may trust it too much.
powertoold
Active Member
Car safety is highly regulated, and Tesla can't really make the excuse of "the driver wasn't paying attention" with the FSD beta. With the beta, we've seen that the driver often has to make split second decisions to avoid a dangerous situation. The average joe with FSD isn't going to be as attentive, no matter how many warnings you show.
It would not be safe enough to release it now.
mikes_fsd
Banned
I kind agree with the sentiment.
But I think there is Tesla legal in the way... given how crazy MSM gets with anything "Tesla"
The FSD beta may look ready from every video that has been shared on social media, however it still needs time. Regulators as well as the NHSTA is monitoring the beta closely. If anything happens the United States Government will shut down FSD for a good 20-30 years.
We need close to a hundred million miles driven w/millions of TB of data collected that we can use to show that the beta is secure.
if Tesla releases a new version every week or two weeks, I have no doubt it be released publicly by Christmas.
Another interesting piece of info in this video is that earlier FSD Beta versions would stop near your destination and the driver would need to finish the drive in manual, but in the latest 2020.40.8.12, FSD Beta actually stays on all the way to your destination. Another little hint that Tesla is working on FSD being destination to destination.
DanCar
Active Member
Disclaimer before you enable it: do NOT enable if you don't have cat like reflexes. Average joe should suffer the consequences if they are going to enable software that has very strict warnings.
DanCar
Active Member
When will it be safe enough? As it gets better it gets more dangerous, because people will start to trust it and abuse it.
powertoold
Active Member
Probably somewhere around an accident probability of every 100-200k miles. This would be a ridiculous achievement. I don't know what Tesla's internal goal is.
Similar threads
- Question
