What is up with high beam nonsense talks? I use auto high beam on M3 all the time as well as my Model S. Never had any problems with it. If you overstress high beam use that is on you. If you get a ticket using auto HB then sue Tesla.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Tesla.com - "Transitioning to Tesla Vision"
- Thread starter run-the-joules
- Start date
Radar-less teslas force high beams on auto in order to use Autopilot, no matter what kind of road you are on. That is what is up.
LOL on suing Tesla for getting a ticket while using an L2 system. That's the issue here- if high beams act up, your only solution is to not use autopilot. That you paid for, and when you paid for it, it had radar and didn't required auto high beams.
powertoold
Active Member
I now understand what he meant. He just saw that vision NNs will definitely be good enough. Since driving is based on vision, Elon considered it as solved problem. And he's going to be right (IMO), just very late.
Oh look, another "what Elon meant, not what he clearly said that every normal human would interpret" discussion...
Look at me, I'm a genius and visionary!
Then I consider immortality to be a solved problem. It's clearly physically possible. Just going to be very late for any of us alive today.
Look at me, I'm a genius and visionary!
powertoold
Active Member
Then there's no such thing as a visionary / genius because eventually everything is possible right?
If timeframes mean absolutely nothing then as long as we aren't discussing something physically impossible like time travel, yeah, kinda.
Who was arguing that vision could not be used to drive a car? What is visionary about believing this could eventually become true?
The visionary part would be if he had actually seen things other people didn't, like if he just put a specific team together, that he could actually deliver self driving to his paying customers in just a few years, way ahead of other companies.
Kind of like the founders of Tesla did
Or yes, Elon in many areas of less expensive space vehicles. He just hasn't shown this same vision around self driving. Only time will tell if he really saw something here that nobody else did, but so far the data is pointed more towards fake it until you make it instead of visionary in the area of autonomous driving.I'm waiting with hitting the update button for as long as I can... let others be beta testers when it comes to crucial stuff like AEB and lane departure...
yup/ and current high beam does a bad job.... it shuts off when passing highly reflective traffic signs with no cars around yet sometimes it takes forever to shut off when a car is coming at you and the driver is already annoyed and flashes their headlights at you.... it's great for open stretches of road at night with no oncoming traffic but that's about it. Matrix headlights would be nice if legal here...
Yea haven't seen too many posts or reviews yet from radar model 3/Y's with 4.18.2. I think i will wait as well until it becomes more clear what this changes in cars with radar.
powertoold
Active Member
The Tesla vision story will be similar to when Steve Jobs went to xerox and saw the mouse UI.
Mobileye wanted to do what Tesla is doing now, but they don’t have the car mfging / software vertical integration.
Still, Elon saw the future and went all-in by committing to a sensor suite and built it in all cars. Elon even found the perfect man for the job.
I’d be surprised if V9 doesn’t match or exceed Waymo performance on the same routes.
Daniel in SD
(supervised)
Can you quantify the performance that you think V9 will achieve?
AlanSubie4Life
Efficiency Obsessed Member
I think the problem is that the auto high beams are far inferior to a human neural net operating at maximum attention level. There are so many ways in which they fail it is hard to list them all.
They're better than the AHB on my Highlander, though, which lit up a CHP Charger patrol car from about 50 feet, since the thin rear light bar didn't look like lights to the Toyota. Pretty sure the Tesla would not do that. The officer did not particularly appreciate it, from what I could tell.
powertoold
Active Member
Can't really, no one will be able to verify the data and Tesla won't likely release it, but if someone has V9 in Chandler, we'd be able to assess qualitatively.
In all Waymo videos, we're assessing qualitatively anyway.
Why the US hasn’t allowed active matrix head lights is beyond me. Problem solved. Essentially, your high beams are on all the time, the beam adapts to what in front of it. Europe has had this for years.
Safety and reliability is just as much "performance" in a safety critical, self driving system. Unless Tesla says "you don't need to pay attention," by definition the performance is worse if what we are discussing is who will sell an L4 system first.
Picking and choosing "Well, it stays in the lanes better" or "Sees a cone farther ahead" while the company itself tells you it absolutely is not an L3+ system isn't really proving any superiority or better performance. When it does make mistakes that Waymo doesn't, people will just say "well, Waymo TRAINED on that exact intersection, Tesla is generalized!" The goalposts always move.
Except adaptive lights need sensing too. Right now, Tesla turns off high beams when a speed limit sign reflects light. Their matrix lights should illuminate this sign, but not the oncoming car. If they know one is a sign, and one is a car, they could already do the right thing with high beams.
Deciding if you are blinding someone with your high beams seems pretty simple. There are even laws- no high beams within 500 feet of another car. Supposedly Tesla has awesome vision processing that tells them if something is a car, and how far away. They only need to do this looking ahead, and at night when identifying cars via lights is even easier. High beams seem trivial if you're anywhere near to driving a car around without any human intervention. Yet, here we are...
Daniel in SD
(supervised)
Waymo has published some data. They estimate 47 "contact events" over 6.1 million miles of driving. There were 18 real contact events and 29 simulated (safety driver took over and avoided a collision that occurred in counterfactual simulation). They had only actually done 65k miles without a safety driver at that point.
I think beating that would be a huge step change from the current FSD beta. Obviously it's impossible to know unless beta users never disengage, maybe the car was just about to correct itself!
powertoold
Active Member
Waymo fudges their data by driving easier and easier routes. Any routes that increase risk of disengagement or cause problems will be avoided.
JJRicks’ disengagement rate is generously 1 per 500 miles.
AlanSubie4Life
Efficiency Obsessed Member
Aren't contact events only very loosely correlated with the route chosen, by their very nature?
Called it.
Tesla V9 will be "surprising" if it doesn't do better than Waymo on those routes, but those routes themselves are so simple they are irrelevant, so Waymo's own success there is to be dismissed.
By a continuous shifting of rhetorical focus, the enemies are at the same time too strong and too weak.
Similar threads
- Article
