Tesla says that "All Tesla Cars Being Produced Now Have Full Self-Driving Hardware" - I do not understand something here. Tesla has only a forward-facing radar, so how to make turn left/right correctly at night? I know.. Tesla has ultrasonics around the car, but its working only for 8 meters. For eg. Google self driving car has long distance, 360 degree lidar and it makes sense. Do tesla notice fast moving motorcyclist when will turn left (at night)?
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Is the Forward-facing radar be sufficient to self-driving car?
- Thread starter Wojtek
- Start date
-
- Tags
- Tesla Inc.
People do it without radar. And people get hit occasionally by cars with their lights out.
Admittedly, I have the same concern.
regardless of what is being put out, and I don't think the hardware currently in use is capable of true full autonomous driving, we will not be seeing a mass adoption of FULLY autonomous vehicles in the main stream for a long time. Initially they can be used in limited settings such as truck transfer yards, ports and the like and as they prove to be viable then they will expand the scope of usage like in airports from terminals to off site rental agencies.
I see the fully autonomous car being ready for mass usage as being at least a decade if not more away.
I see the fully autonomous car being ready for mass usage as being at least a decade if not more away.
James Anders
Member
dhcp
Member
I think the regulations alone and FUD by the general public is going to keep it from becoming a reality for a while, even if the technology is there.
The first time a car running full autonomous takes out little Billy riding his bicycle there is going to be major litigation and people going after Tesla. It will be a mess for a long time and states will freak out and ban it until "more evaluation can be done."
We live in such a litigious country, this won't be easy.
The first time a car running full autonomous takes out little Billy riding his bicycle there is going to be major litigation and people going after Tesla. It will be a mess for a long time and states will freak out and ban it until "more evaluation can be done."
We live in such a litigious country, this won't be easy.
The new hardware has significantly expanded the range. No longer limited to 8 meters.
From the Tesla web site:
"Eight surround cameras provide 360 degrees of visibility around the car at up to 250 meters of range. Twelve updated ultrasonic sensors complement this vision, allowing for detection of both hard and soft objects at nearly twice the distance of the prior system. A forward-facing radar with enhanced processing provides additional data about the world on a redundant wavelength that is able to see through heavy rain, fog, dust and even the car ahead."
Autopilot
Volvo and Audi/VW I believe are headed toward 360 camera coverage, 360 radar coverage, 360 ultrasonics, as well as LIDAR in the front.
Tesla's suite certainly is behind all that, but from a customer point of view will probably make up at least some of that with their more aggressive software update based progress. So while an AP2 Tesla possibly will never drive without a driver, it will still quite probably be one of the first things one can buy that will end up driving mostly by itself - especially with a driver present.
Elon's comment to an AP1 owner asking about AP2 upgrade recently talks of major revisions every 12-18 months. I would not be surprised to see an AP3 with upgraded computer (maybe Tesla's own silicon?) and added 360 radar coverage in 12-18 months. LIDAR, now, that is more iffy because we know Elon has been a bit against it for automotive use publicly.
Agreed on the concern. One comment on your first part: People do drive without radars, but they do have moving heads and the ability to clear windshields, step out of the car if need be etc. Tesla in AP2 has a camera based system where everything apart from the front vision (which seems fairly robust with radar, three cameras behind heaters and windscreen wipers) is based on small fixed lenses with nothing but heating to keep them seeing things. Without radar backup, that vision is quite easily clouded I would expect if there is no driver to maintain it.
Tesla's suite certainly is behind all that, but from a customer point of view will probably make up at least some of that with their more aggressive software update based progress. So while an AP2 Tesla possibly will never drive without a driver, it will still quite probably be one of the first things one can buy that will end up driving mostly by itself - especially with a driver present.
Elon's comment to an AP1 owner asking about AP2 upgrade recently talks of major revisions every 12-18 months. I would not be surprised to see an AP3 with upgraded computer (maybe Tesla's own silicon?) and added 360 radar coverage in 12-18 months. LIDAR, now, that is more iffy because we know Elon has been a bit against it for automotive use publicly.
Agreed on the concern. One comment on your first part: People do drive without radars, but they do have moving heads and the ability to clear windshields, step out of the car if need be etc. Tesla in AP2 has a camera based system where everything apart from the front vision (which seems fairly robust with radar, three cameras behind heaters and windscreen wipers) is based on small fixed lenses with nothing but heating to keep them seeing things. Without radar backup, that vision is quite easily clouded I would expect if there is no driver to maintain it.
Last edited:
That is a mistaken reading. The 250 meter visibility is for the cameras and the radar. The ultrasonics have been upgraded, but they definitely don't see up to 250 meters. Nowhere near that. 10 meters maybe?
Also, ultrasonics are slow.
The new hardware has significantly expanded the range. No longer limited to 8 meters.
From the Tesla web site:
"Eight surround cameras provide 360 degrees of visibility around the car at up to 250 meters of range. Twelve updated ultrasonic sensors complement this vision, allowing for detection of both hard and soft objects at nearly twice the distance of the prior system. A forward-facing radar with enhanced processing provides additional data about the world on a redundant wavelength that is able to see through heavy rain, fog, dust and even the car ahead."
From the Tesla web site:
"Eight surround cameras provide 360 degrees of visibility around the car at up to 250 meters of range. Twelve updated ultrasonic sensors complement this vision, allowing for detection of both hard and soft objects at nearly twice the distance of the prior system. A forward-facing radar with enhanced processing provides additional data about the world on a redundant wavelength that is able to see through heavy rain, fog, dust and even the car ahead."
Tesla's web site just claims twice the distance for the updated ultrasonic sensors not 250 meters. My only reason for posting was to provide additional information for the OPS whose initial posting appeared to be based on the initial AP hardware.
Last edited:
It seems that Volvo and Audi/VW current in-vehicle suite is behind all that as well. Isn't it true that by definition any current suite will be behind the future ones?
disagree
hos epi to polu
Tesla asserts that the hardware suite makes it possible for the cars to be FSD. Of course, they don't control regulations. What is worrisome is that Tesla states some weird "if we can get our software to work" condition that is pretty odd since it entirely undermines the initial claim of knowing that the car has all the hardware needed to make it work.
So is the suite sufficient? Wait to see. Is the FSD video completely misleading and perhaps fraudulent? Wait to see. Exciting times!
So is the suite sufficient? Wait to see. Is the FSD video completely misleading and perhaps fraudulent? Wait to see. Exciting times!
I'm not sure what Volvo ships today but the Germans definitely have been shipping far larger radar and camera coverage already for years than Tesla. And will deliver Level 3 solutions this year with again more sensors than Tesla in AP2. And their Level 4/5 self-driving prototypes use again much more sensors than Tesla (as does Volvo's self-driving suite). On the other hand, they are very conservative and slow with their software, so far updating software only through model changes and even there very conservative steps...
So a different approach. The competition will be using more sensors from the get go than AP2, but still AP2 may quite possibly be the first that can self-drive in a significant fashion on the market, due to the more aggressive software policy. It may not be sufficient for driverless self-driving (at least in a very wide fashion, maybe in California
), for that perhaps AP3 will be needed.I'm not saying AP2 can't get to self-driving. Maybe it can. Maybe even in many markets. But I can see how it would be problematic in snowy countries with these sensors. Redundancy through 360 radars would do wonders to my confidence. So I would expect the self-driving to turn off quite easily. We shall see.
I still think AP2 is a very, very nice feature, so don't get me wrong. But Tesla is using quite a bit less sensors than the competition - always has...
Tesla asserts that the hardware suite makes it possible for the cars to be FSD. Of course, they don't control regulations. What is worrisome is that Tesla states some weird "if we can get our software to work" condition that is pretty odd since it entirely undermines the initial claim of knowing that the car has all the hardware needed to make it work.
So is the suite sufficient? Wait to see. Is the FSD video completely misleading and perhaps fraudulent? Wait to see. Exciting times!
I am 100% confident AP2 can self-drive and that Tesla will release the software to do so eventually. There is a small chance Tesla needs to ugprade the CPU for FSD, but probably not.
What I am not as confident is whether or not they will get wide regulatory approval without 360 radars - and whether or not 360 cameras will be sufficient for operation in more challenging weather.
The idea of your car coming to pick you up driverless might still elude this car generation. But sure, drive around in sunny California in 99% of the situations with a passive driver there, I am confident AP2 will be able to do it eventually...
I agree with @AnxietyRanger . I'm confident they can do good weather, sitting at the helm ready to take control.
But several times a day, picking up my kids from activities, I have to turn left across two lanes of traffic (it's a 4 lane, 2 each way with a center turn/suicide lane). This seams very hard to do automatically. I have read the white papers on approximating distance/speed using a single camera. That should work in the daytime. But it just seams very difficult at night on a dark road.
But several times a day, picking up my kids from activities, I have to turn left across two lanes of traffic (it's a 4 lane, 2 each way with a center turn/suicide lane). This seams very hard to do automatically. I have read the white papers on approximating distance/speed using a single camera. That should work in the daytime. But it just seams very difficult at night on a dark road.
I don't see how cameras can do the FSD job, especially at night. It would be like expecting humans to drive at night with only their eyes to see.
Point. Gloriously. Missed.
Humans have hands to clean their lenses and ability to move their lenses around vision obstacles.
Most humans have two lenses for depth vision with the ability point to sides, the rear.
The target here - right - is a car that can come pick you up by itself? At night in bad weather too?
It seems like all it takes is one side camera blocked by an unlucky spot of snow or dirt and that's gone.
I am thinking FSD suites will probably have to be more robust than this for totally driverless driving. 360 radars to compliment cameras would seem like a good addition. Jury is still out on LIDAR I guess. Volvo for example is doing 360 cameras, 360 radar, 360 ultrasonics and LIDAR. That is quite a bit more comprehensive.
That said, surely FSD is possible with AP2 with a passive driver present and in many weather scenarios. And I fully expect it to be the first on market to do so. Still a great accomplishment.
Last edited:
Point. Gloriously. Missed.
Humans have hands to clean their lenses and ability to move their lenses around vision obstacles.
Most humans have two lenses for depth vision with the ability point to sides, the rear.
The target here - right - is a car that can come pick you up by itself? At night in bad weather too?
It seems like all it takes is one side camera blocked by an unlucky spot of snow or dirt and that's gone.
I am thinking FSD suites will probably have to be more robust than this for totally driverless driving. 360 radars to compliment cameras would seem like a good addition. Jury is still out on LIDAR I guess.
That said, surely FSD is possible with AP2 with a passive driver present and in many weather scenarios.
FSD is not equivalent to unsupervised driving, and is not designed for it. There is driver in the car who can take over, or clear obstruction.
The point is that FSD will handle (insert your number) percent of driving, including driver pickup, reducing driving workload, fatigue and improving safety.
Unsupervised fully autonomous driving will require redundant sensors/processors.
Obsession with hardware is theoretical and missing the point that while Tesla suite of sensors is less robust, it has functionally the best driving assist system bar none, confirmed by multiple car magazine comparisons.
Neural net training using existing fleet is the differentiator that will make it very difficult for others to catch up.
In my opinion being "conservative" is just an excuse used by other automakers to cover unpleasant fact that they are behind (robustness of sensor suite notwithstanding). Reminds me of Mobileye excuses.
Similar threads
