HankLloydRight
No Roads
Also amazing is how the bird droppings perfectly hit both sensors. That bird has some aim!!
Also if you watch real closely.. The smear self disappears between the fourth and fifth wipe and not as the wiper wipes it. Hmmmmmmm.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Also if you watch real closely.. The smear self disappears between the fourth and fifth wipe and not as the wiper wipes it. Hmmmmmmm.
lunitiks
Cool James & Black Teacher
No that's just some high tech RainX- or SpaceX-stuff.
But if you look really, really hard, it's obvious that this is a beta prototype. One of the water jets doesn't hit the sensor but the back of the enclosure. (Btw I hope that dirty water doesn't hit the driver in the head. Wait, there is no driver...)
Still amazing, though
Last edited:
Have any of you read this article? Is Tesla Building A Moat With HD Maps? - Tesla Motors (NASDAQ:TSLA) | Seeking Alpha
It seems to me that the key to self driving is mapping. With essentially real time good HD mapping, it comes down to avoiding random events and Tesla is already pretty good at that. Cloud/big data on steroids with AI, Musk's prediction of a FSD drive across country looks more probable than many think.
It seems to me that the key to self driving is mapping. With essentially real time good HD mapping, it comes down to avoiding random events and Tesla is already pretty good at that. Cloud/big data on steroids with AI, Musk's prediction of a FSD drive across country looks more probable than many think.
Maximilien
Member
By the end of 2017!
I believe in Elon. I guess I am drinking too much Tesla koolaid.
I believe in Elon. I guess I am drinking too much Tesla koolaid.
By the end of 2017!
I believe in Elon. I guess I am drinking too much Tesla koolaid.
Elon time may be inflated and certainly unrealistic/off, but do we really expect his estimate to be off on the scale of a couple years? lol. 2017 seems realistic to me, if not for the most basic of self driving features.
I don't think High resolution mappings is the answer for self driving. Road works occur all the time changing the position of lines, cars break down blocking lanes, etc. There was a major accident near me and the police ended up opening the central reservation and making drivers cross the centre which is normally Armco and drive on the wrong side, a situation a high res map would say "No! don't do it!". If the car needs to deal with new road layouts not yet mapped, it can deal with any road.
It's the things that hardly ever happen that's the challenge for self driving whereas it's the things that happen all the time that's important for AutoPilot, and Tesla is a long way from being good and dealing with the unexpected. Driving 100 miles on a freeway (or motorway as we call them) is very predictable compared to navigating some workman with a manual stop/go board near parked cars, outside a school - a scenario you may not have encountered yet in life but could easily happen tomorrow. You need a human driver to take over and as soon as you accept that may happen, you've just got a higher degree if AutoPilot in your car.
So for me, AutoPilot was over sold as it's really just active cruise with lane keeping assist, as good as both of those are.
FSD is what people thought AutoPilot was going to be and will make some decisions along the way
Real Full self driving is not going to happen on the current cars.
I think full self driving everywhere is a long ways away. It will probably take at least several years. There is also the possibility that it will never be fully implemented due to technical or social challenges.
Partial functionality can be rolled out sooner, but the special or difficult cases will remain problematic for a very long time.
Partial functionality can be rolled out sooner, but the special or difficult cases will remain problematic for a very long time.
lunitiks
Cool James & Black Teacher
HabitualUser
Member
Disagree, though I used to think this was true. Apparently the only necessities are cameras and software.
To me the interesting developments will be in dealing with the unique moments of street driving, like when driving behind a bus that makes a stop, will the car notice and change lanes? Or patiently wait for the bus to get going again.
Will the car try to race through a yellow light, or stop immediately? Sometimes its safer to drive through it.
Will the car try to avoid potholes/dips? That sometimes requires drifting into another lane
I agree, and will add that this will continue to be the answer for the next decade or two.
Coast to Coast Autonomous Driving Demo ---- Still on Track for End of 2017
(but this means almost nothing)
Level 4 on freeways (Better than human performance on freeways) ----- E) First Half of 2020
However, I am not sure Tesla will roll-out L4 in sections. Like I do not feel Tesla cars will allow drivers to go to sleep on freeways, but also be fully autonomous in urban areas, but require hands by wheel and eyes on road. So I do not think Tesla will enable any kind L4(allowing driver to do something else) until they can enable it for all operational domains at once.
Level 4 in urban areas (Better than human performance in cities) ---- G) 2021 and Beyond...
Really not sure about this but we will see, I guess it could be around 2021.
Level 2 Fully autonomous on most roads (Can drive from A to B with disengagements required) ---- B) Second Half of 2018
What I mean by this is Tesla AP2 will have like "intersection autonomy" and in like easy suburban cities. A Tesla could drive fully autonomously say from my house to the grocery store and back. Can handle merging, exiting, stoplight intersections, stop sign intersections. And in perfect conditions the Tesla may drive with no disengagements. However, the Autopilot will definitely not be close to human level performance, and will require human driver watching at all times ready to take over, i.e. L2. I actually feel the frequency of disengagements would be pretty high: like 1 disengagement for every 20-50 miles. (depending on the area and types of roads)
@Bladerskb
I am interested in your opinion on Tesla's progress towards 'Level 2 full autonomy'
freeewilly
Member
My guess it will reach 99% level 5 by 2020, and beyond 2021 for 100%, that means it can drive anywhere in any condition, no road marking, weather, road hazard, animal jump in front of the car...etc.
I see L5 capable on interstates and some divided highways by mid 2018. Similar to speed limits some roads will be L5 some L4 depending on the markings, quality and difficulty level. I think roads with speed limits of 25-50 mph are going to be the toughest.
In Nebraska they were going to pass a law to allow L5 but the autonomous companies were neutral for it. They wanted national rules instead of each state being different so Nebraska decided to wait.
In Nebraska they were going to pass a law to allow L5 but the autonomous companies were neutral for it. They wanted national rules instead of each state being different so Nebraska decided to wait.
I suspect that navigation/user interface ends up being one of the harder parts of FULLY self driving vehicles.
Fast food drive thru. You have to stop to order, stop to pay and stop to get food at many of them. Sometimes you also have to pull ahead and wait for food. Assuming the car can figure out where to order and where the windows are, it is still going to need to be told when to pull to the next station and if a stop at the wait for slow food spot is needed. (Carwash, atm, mailbox, ...)
Super Target parking on the grocery side near a cart corral. There are tons of where to park sorts of questions. I'm visiting a friends's house... do I park in the driveway or on the street? do I know not to block certain garage doors?
Driving down the street and you see a sign for a garage sale you want to stop at. "Hey car, park near that big bunch of balloons."
All sorts of drive past x on the way to y.
This is just a couple minutes of thinking. I suspect there are at least 100s of these sorts of special cases that have to be handled more or less one at a time before one would never want/need to do any driving.
Fast food drive thru. You have to stop to order, stop to pay and stop to get food at many of them. Sometimes you also have to pull ahead and wait for food. Assuming the car can figure out where to order and where the windows are, it is still going to need to be told when to pull to the next station and if a stop at the wait for slow food spot is needed. (Carwash, atm, mailbox, ...)
Super Target parking on the grocery side near a cart corral. There are tons of where to park sorts of questions. I'm visiting a friends's house... do I park in the driveway or on the street? do I know not to block certain garage doors?
Driving down the street and you see a sign for a garage sale you want to stop at. "Hey car, park near that big bunch of balloons."
All sorts of drive past x on the way to y.
This is just a couple minutes of thinking. I suspect there are at least 100s of these sorts of special cases that have to be handled more or less one at a time before one would never want/need to do any driving.
True, but that doesn't make the problem less daunting. With the right software the car could write a Shakespeare play and then drive you to the theater to see it. Driving requires a lot of unstructured behavior that we haven't yet demonstrated we know how to program. At a chess tournament there are the grandmasters, and then there are the janitors who clean up afterwards. Computers can win the tournament, but can't do the janitor's work at all.
lunitiks
Cool James & Black Teacher
Never going to happen. Forget L5. Fast food joints will end the self driving fantasy
Various views expressed in this thread suggest that FSD is years and years away.
Elon has promised a coast to coast FSD demo drive within the next 7 months.
So those who believe FSD is a long, long way away must think that either:
Regarding 2 - it's not clear to me exactly what game he might be trying to play. Until some reasonable "game-playing" theory emerges, I'm forced to also discount 2.
(by game-playing theory, I'm talking about an explanation of exactly what short term goals Elon is seeking to achieve by making knowingly false optimistic claims about FSD and how the value of these goals exceeds the longer term price he will pay when they prove false)
We may come back to 3. depending on what responses I get to this ramble...
Therefore if Elon says "it's coming sooner than you think" and repeatedly promises to demonstrate it before the end of 2017, I'm inclined to take that at face value.
By the way, comparing FSD with current autopilot is a chalk vs.cheese exercise. Current autopilot is designed for highways only, and is using a combination active cruise and highway lane keeping vision which are both quite old technologies now; I was driving cars 8 years ago with both of these facilities. Having said that, I like autopilot a lot. I have used it over thousands of miles in my AP2 Model S.
FSD, in comparison, will use technology that was not available to anyone 8 years ago. AI became better than people at vision only about 5 years ago, and the even bigger achievements in deep learning that are making headlines at the moment (e.g. DeepMind/AlphaGo) only became possible far more recently than that. We're on an AI roll at the moment, and Elon Musk is an insider in that process. His claims are made based on stuff that he knows, that we don't, and stuff he has seen, that we haven't. And I repeat, he's not stupid.
There. Said it now
Elon has promised a coast to coast FSD demo drive within the next 7 months.
So those who believe FSD is a long, long way away must think that either:
- Elon Musk is some kind of idiot who is entirely ignorant of difficult FSD scenarios that quickly occur to you and me
- Elon Musk is somehow trying to play a cynical game. He knows that FSD is years away but it somehow serves his purpose to say it's far closer
- Or some other theory - suggestions welcome
Regarding 2 - it's not clear to me exactly what game he might be trying to play. Until some reasonable "game-playing" theory emerges, I'm forced to also discount 2.
(by game-playing theory, I'm talking about an explanation of exactly what short term goals Elon is seeking to achieve by making knowingly false optimistic claims about FSD and how the value of these goals exceeds the longer term price he will pay when they prove false)
We may come back to 3. depending on what responses I get to this ramble...
Therefore if Elon says "it's coming sooner than you think" and repeatedly promises to demonstrate it before the end of 2017, I'm inclined to take that at face value.
By the way, comparing FSD with current autopilot is a chalk vs.cheese exercise. Current autopilot is designed for highways only, and is using a combination active cruise and highway lane keeping vision which are both quite old technologies now; I was driving cars 8 years ago with both of these facilities. Having said that, I like autopilot a lot. I have used it over thousands of miles in my AP2 Model S.
FSD, in comparison, will use technology that was not available to anyone 8 years ago. AI became better than people at vision only about 5 years ago, and the even bigger achievements in deep learning that are making headlines at the moment (e.g. DeepMind/AlphaGo) only became possible far more recently than that. We're on an AI roll at the moment, and Elon Musk is an insider in that process. His claims are made based on stuff that he knows, that we don't, and stuff he has seen, that we haven't. And I repeat, he's not stupid.
There. Said it now
Various views expressed in this thread suggest that FSD is years and years away.
Elon has promised a coast to coast FSD demo drive within the next 7 months.
So those who believe FSD is a long, long way away must think that either:
I think it's fair, based on the guy's reputation and achievements, to eliminate option 1. above. He's not an idiot; quite the opposite.
- Elon Musk is some kind of idiot who is entirely ignorant of difficult FSD scenarios that quickly occur to you and me
- Elon Musk is somehow trying to play a cynical game. He knows that FSD is years away but it somehow serves his purpose to say it's far closer
- Or some other theory - suggestions welcome
Regarding 2 - it's not clear to me exactly what game he might be trying to play. Until some reasonable "game-playing" theory emerges, I'm forced to also discount 2.
(by game-playing theory, I'm talking about an explanation of exactly what short term goals Elon is seeking to achieve by making knowingly false optimistic claims about FSD and how the value of these goals exceeds the longer term price he will pay when they prove false)
We may come back to 3. depending on what responses I get to this ramble...
Therefore if Elon says "it's coming sooner than you think" and repeatedly promises to demonstrate it before the end of 2017, I'm inclined to take that at face value.
By the way, comparing FSD with current autopilot is a chalk vs.cheese exercise. Current autopilot is designed for highways only, and is using a combination active cruise and highway lane keeping vision which are both quite old technologies now; I was driving cars 8 years ago with both of these facilities. Having said that, I like autopilot a lot. I have used it over thousands of miles in my AP2 Model S.
FSD, in comparison, will use technology that was not available to anyone 8 years ago. AI became better than people at vision only about 5 years ago, and the even bigger achievements in deep learning that are making headlines at the moment (e.g. DeepMind/AlphaGo) only became possible far more recently than that. We're on an AI roll at the moment, and Elon Musk is an insider in that process. His claims are made based on stuff that he knows, that we don't, and stuff he has seen, that we haven't. And I repeat, he's not stupid.
There. Said it now
I think Tesla will demonstrate a Coast to Coast Demo at the end of 2017.
But that does not mean that any form of FSD is ready to be released or even close.
When you say "at or better than humans" leaves a LOT of room. If, for instance, you take the driving habits of the person who hit me in March, then I'd say AP2 is about the same level (if not better) already. That said, I don't believe FSD will be fully worked out when it does get released as it is virtually impossible to encounter all possible scenarios without "experience". The miles gained on AP1/AP2 and an early FSD will hopefully lead to systemic learning as time goes on, which in turn, makes a truly dependable FSD a moving target.
Basically, I agree with what you're saying and I think you're somewhere in the right ballpark.
Similar threads
