Are Tesla's Data-Gathering Cars Secretly Improving Autopilot's Algorithms? (cringely.com) 119
"When the history of autonomous cars is written, the winner will be Tesla," speculates long-time technology pundit Robert Cringely. "Heck, I think they've already won."
But his article includes a disclaimer that it's "based pretty much on logic, not knowledge, which is to say I might again be too frigging stupid to read, much less write." Tesla has more than a million data-gathering devices on the roads. We call them cars. Tesla cars have no LIDAR but they have eight cameras and RADAR. Every night all those cars wirelessly report their driving data back to Tesla. I would love to know how Tesla decided what to put in those reports. Given the limited bandwidth LTE connection involved, it can't be a complete data dump. They have to pick and choose what to report. And what does Tesla do with the reports? I think it comes down to algorithms, mapping, and exceptions. They are logically trying to improve their algorithms, improve their maps, but mainly — after having already parsed billions of miles of driving data — they are looking for exceptional events that are testing their algorithms in ways never seen before...
Tesla has a dual processor system in their cars — two completely distinct computers. Why...? Because every night is an A-B test for Tesla — a test that is running on your car. One processor is driving the car (or following the driver's actions if Autopilot isn't being used, which is most of the time) with production software while the second processor is running beta software, simulating the drive, and noting discrepancies between the two software versions. Multiply this times a million cars per night. Whether Autopilot is used or not doesn't matter: the evolution of the software continues. And it's finished when the beta software stops improving and the outcome shows the only difference between human and Autopilot driving is that Autopilot does it better. Continue for another month or year or decade just to confirm your results, then announce that full autonomous mode is available. That is exactly where I believe Tesla has been heading for as long as those two-processor cars have been on the road.
Tesla's autonomous driving software could be ready right now for all we know. Elon certainly hints at this from time to time in his tweets. And THAT's why I believe Tesla has already won the autonomous driving war, because they have real cars facing real exceptions that you won't find in a simulation, and their dual processor system knows what it knows.
Yes, I reached out to Tesla about this last week. They still haven't replied.
Again, Cringely wants that this is "based pretty much on logic, not knowledge, which is to say I might again be too frigging stupid to read, much less write."
But his article includes a disclaimer that it's "based pretty much on logic, not knowledge, which is to say I might again be too frigging stupid to read, much less write." Tesla has more than a million data-gathering devices on the roads. We call them cars. Tesla cars have no LIDAR but they have eight cameras and RADAR. Every night all those cars wirelessly report their driving data back to Tesla. I would love to know how Tesla decided what to put in those reports. Given the limited bandwidth LTE connection involved, it can't be a complete data dump. They have to pick and choose what to report. And what does Tesla do with the reports? I think it comes down to algorithms, mapping, and exceptions. They are logically trying to improve their algorithms, improve their maps, but mainly — after having already parsed billions of miles of driving data — they are looking for exceptional events that are testing their algorithms in ways never seen before...
Tesla has a dual processor system in their cars — two completely distinct computers. Why...? Because every night is an A-B test for Tesla — a test that is running on your car. One processor is driving the car (or following the driver's actions if Autopilot isn't being used, which is most of the time) with production software while the second processor is running beta software, simulating the drive, and noting discrepancies between the two software versions. Multiply this times a million cars per night. Whether Autopilot is used or not doesn't matter: the evolution of the software continues. And it's finished when the beta software stops improving and the outcome shows the only difference between human and Autopilot driving is that Autopilot does it better. Continue for another month or year or decade just to confirm your results, then announce that full autonomous mode is available. That is exactly where I believe Tesla has been heading for as long as those two-processor cars have been on the road.
Tesla's autonomous driving software could be ready right now for all we know. Elon certainly hints at this from time to time in his tweets. And THAT's why I believe Tesla has already won the autonomous driving war, because they have real cars facing real exceptions that you won't find in a simulation, and their dual processor system knows what it knows.
Yes, I reached out to Tesla about this last week. They still haven't replied.
Again, Cringely wants that this is "based pretty much on logic, not knowledge, which is to say I might again be too frigging stupid to read, much less write."
Re:Doesn't matter (Score:5, Informative)
technology pundit Robert Cringely
No. Just Fucking No.
A Brief History of Robert X. Cringely
In 1986, Mark Stephens was hired as a writer by Inforworld magazine. Writing under the name "Robert X. Cringely", he began his career as a professional bullshitter.
When he was fired a few years later, Infoworld sued him over his continued use of the Cringely name. Eventually they reached an agreement where he was allowed to continue using the name Robert X. Cringely as long as he wasn't working for a competitor of Infoworld.
After Infoworld went out of business, Stephens began claiming that he is "the original Robert X. Cringely". He isn't. Before he was hired by Infoworld there were at least two other people who wrote columns using the the Cringely pseudonym.
At various points in his career, he has also claimed that he was employee number 12 at Apple, he helped them move out of Steve Jobs' garage, and he designed the original Mac trash can icon. None of this is true. Apple employee number 12 was actually Daniel Kottke. The trash can icon was created by Dan Smith and Bill Atkinson and there is no credible evidence that Mark Stephens ever worked at Apple.
In 2015 Cringely announced 'The Mineserver Project' on Kickstarter. These miniature Minecraft servers would be small, inexpensive ARM-based boards, running Linux, slightly more powerful than a Raspberry Pi and selling for $99. The project raised a total of $35,452 from 388 people and the finished boards were supposed to ship in December 2015. They didn't.
For the next two years Cringely was asked about the Minseserver boards and repeatedly claimed that they were finished and ready to ship, but there was always "one more little thing" that needed to be fixed. Then in October 2017 Cringely claimed that his house burned down and all the Mineserver boards were destroyed. Just like his tenure at Apple, there is no credible evidence that any of this is actually true.
It has now been 5 years since Cringely took $35,000 from people and delivered nothing. This isn't particularly interesting or unusual since a lot of Kickstarter projects fail. It happens. But people keep asking him about it and Cringely keeps lying. Blatant, obvious lying. One lie after another, each one bigger and more ridiculous than the last.
Cringely's latest and biggest whopper hit his blog in January 2020 when he announced his new business venture, Eldorado Space. This would be a company using F-104 jets to launch satellites. Cringely says revenue from this business will fund his retirement (he's 67 now) and give him enough money to finally deliver those Minecraft servers he's been promising for the last 5 years.
He also claimed that the business is guaranteed to succeed because his new company has bought all the F-104s in existence, so he won't have any competition. To prove this is all real and legitimate, Cringely found a picture of an F-104 on the Internet and photoshopped the word Eldorado onto it.
The truth and Robert X. Cringely are not well acquainted.
Re: (Score:2)
I remember feeding "Cringely" leads he later turned into articles, but never met him personally. Unless it was at some trade show or conference and I didn't recognize him. However it never sunk into me before that there was more than one Cringely and I wonder if it was Stephens or someone else that I was corresponding with. Not that it matters.
I did know Steve Gibson pretty well although we have fallen out of touch. I'm sure if I called him he would know.
Re: (Score:1)
Beep beep-Excessive Bass Line-Rims in Your Lane (Score:1)
Tesla autopilot: "Detecting abnormally loud bass lines" ...
Tesla autopilot: "Detecting rims extending into adjacent lanes"
Tesla autopilot: "Taking evasive action captain"
No (Score:4, Insightful)
Tesla is nowhere near having fully self driving cars. No amount if data will help, the problem is the limits of their vision system.
Even if they somehow get it to work they don't have wipers for most of the cameras so they can't deliver on the system they have been selling since 2016 anyway.
Humans have a pretty limited vision system (Score:2)
Re:Humans have a pretty limited vision system (Score:4, Interesting)
Unless you both were arguing about something different, humans also have eyelids, i.e. wipers for their cameras.
Re: (Score:2)
Humans also have a much better understanding of the world. Computers have to make do without the ability to infer or anticipate nearly so much.
Every other company has demonstrated a 3D model of the environment around the car. Tesla has not. Turning an arbitrary set of images into a 3D model is really hard. Maybe impossible with the kind of computer Tesla has been fitting.
Re: (Score:2)
Uh, Tesla has. And turning stereo images into a 3D model isn't very hard, it's fairly simple. Simpler than building a LIDAR device. I've got a $100 Intel RealSense camera that does it, in real time.
Re: (Score:2)
The weights in a neural net *are* the model of the world. Just because it's hard to pull them out separate from the rest of the program doesn't mean it isn't there.
That said, they aren't necessarily either a good or sufficient model of the world, and they *are* limited by data availability. So I don't think a strong opinion either way is valid. Bumble bees have a pretty good model of the world on a rather limited computer, even though nobody could pull that "world model" out as a separate thing.
That said
Re: (Score:2)
To give you an example consider a pedestrian crossing. The car is obliged to stop if someone is waiting to cross the road, that's the rule.
So first the car has to identify people waiting, including children, people in wheelchairs, on bikes etc. Then it has to identify the pedestrian crossing. It has to locate them all in 3D space so that it can calculate how far they are from each other and if it is likely to need to stop by the time they reach the crossing. That means it also has to understand the route th
Re: (Score:3)
And yet, Tesla consistently outperforms all of them. How can this be?
But, we don't have eyes in the back of our heads, or vision on either side, simultaneously feeding into our visual perception from the front. Teslas do. Yet, you doubt they would be capable of making decisions based on what's going on around them.
It's always surprising to see technophile technol
Re: (Score:2)
And yet, Tesla consistently outperforms all of them. How can this be?
It does not. Waymo is operating a driverless taxi service today. No safety driver at all. Even Tesla is only claiming that they will reach that level by the end of this year, which is a looking increasingly unlikely.
Re: (Score:2)
Consistently? 16 year-old new drivers "consistently" crash into things, Teslas on Autopilot have a much better record than the average 16 year-old. Yes, one recently ran into a jackknifed truck, but that was enough of an anomaly that it made national news. The driver survived, in part because once it realized there was a problem the software reacted far faster than a human could to reduce the impact. There were another 11 (daily average) drivers nationwide who didn't have the same luck and died that day
Tesla demonstrated it on April 22, 2019 (Score:2)
However based to your use of insults .. (Score:2)
.. you do not seem to use much of it.
Re: (Score:2)
Even if they somehow get it to work they don't have wipers for most of the cameras so they can't deliver on the system they have been selling since 2016 anyway.
They only really need wipers on the front. The other cameras only really matter in a very broad, general sense. The side cameras basically only matter if you're changing lanes, and the computer should already know if there's a car beside you based on previous frames, plus the car has the ultrasonics as a secondary source of information. The back camera basically only matters in parking lots and driveways, and even then, distorted images are unlikely to prevent you from seeing an obstacle.
Re: (Score:2)
If they didn't need them they wouldn't have fitted them.
Re: (Score:2)
If they didn't need them they wouldn't have fitted them.
Needing them to exist is not the same as needing their images to be perfect. In perfect weather, you could ostensibly get by with only four cameras. The Tesla has eight cameras for redundancy, with maximum redundancy in the front, and minimum redundancy straight back.
When you're driving, the forward cameras need to be pretty close to perfect, because you're moving in that direction at a rapid speed. That's why there are three cameras pointed in that direction — a normal camera, a wide camera, and a
Re: (Score:3)
Okay let's just review what Tesla have been selling people since 2016. Full Self Driving, the ability to summon your car from thousands of miles away.
So the car needs to be self cleaning. If a bird craps on the camera or it gets covered in snow or splashed with mud it needs to self-clean. Same goes for the radar, when you have snow it gets blocked by build-up as you drive along and autopilot becomes unavailable (this affects all cars with radar).
It also needs to self charge. Tesla needs to upgrade its charg
Re: (Score:2)
Okay let's just review what Tesla have been selling people since 2016. Full Self Driving, the ability to summon your car from thousands of miles away.
So the car needs to be self cleaning. If a bird craps on the camera or it gets covered in snow or splashed with mud it needs to self-clean.
Like I said, that's what redundancy is for. Anything close enough to matter should be captured in more than one camera, which gives you one-camera fail handling, at least long enough to get to the nearest supercharger, where the car wash attendant can take care of it.
Same goes for the radar, when you have snow it gets blocked by build-up as you drive along and autopilot becomes unavailable (this affects all cars with radar).
The snow and RADAR thing is why the RADAR unit has heaters. Just give it long enough, and the snow should melt. Also, the RADAR is itself an entirely redundant safety system, at least in theory.
Re:No (Score:5, Interesting)
Re: (Score:2)
They need to be able to build a 3D model of the world in order to navigate through it safely and anticipate the actions of other vehicles, pedestrians and more. So far they have not been able to do that, nothing even close.
That in turn limits their ability to learn. All they can do at the moment is object recognition and looking for cases where the current driving aids consistently fail.
Also as I said they can't deliver on things like robotaxi or summoning from far away because they have no way to clean the
Cameras are cheap, as is stereo vission (Score:2)
A basic extra wide angle camera would cost less than $100, even with a small wiper. So I doubt if the fields of view would be an issue.
Stereo vision works pretty well at producing a 3D map of objects. I am not sure why they even bother with radar, I suppose the backup is useful in bad weather.
Interpreting what it sees and making sensible and totally reliable decisions. Now that is very challenging.
But self driving is not all or none. Driving down the freeway and well marked main roads is much easier tha
Re: (Score:2)
Basic one is no good, needs to be automotive grade so it survives the harsh environment and high dynamic range so it can operate in various conditions.
They don't have stereo vision so can't use that for 3D mapping. They could potentially use temporal data but it requires a lot more processing power.
Radar is required because the front cameras can't estimate depth reliably and not crashing into the back of things is extremely important.
Also note that they are claiming their current cars are capable of full se
Re: (Score:2)
Why wouldn't they use stereo vision? Relatively simple. Do you know they don't or are you just guessing?
Cameras are typically behind the windscreen (under the rear view mirror), so not that hostile an environment.
Radar is a good backup, but no substitute for vision. Remember that car that smashed into the underside of a truck. Or, alternatively, won't go under a bridge.
Re: (Score:2)
I don't know, why don't you ask Tesla why they only fitted one camera instead of two for stereo vision? I guess they don't think they need it.
Realistically it probably wouldn't work very well as the kinds of distances they need anyway. They cameras would need to be fair far apart and have very high resolution, meaning smaller pixels in the sensor and worse low-light performance and dynamic range.
Even the inside of a car is fairly hostile in terms of temperatures, it gets baking hot in the sun and freezing c
Re: (Score:1)
The actionable field is too small from the front camera
There are three front-facing cameras above the rear-view mirror. The one recorded by the dashcam software is the near-field camera. But one of the other cameras records about a 160 degree view.
Re: (Score:2)
No wipers? Wow. The fact that their current model does not have wipers clearly means that their future cars will totally suck. No way they can overcome that permanent disability.
Yeah, read the freakin' post, let alone the article. The article is claiming that they have an unfair advantage in building future vehicles due to their data. You ignored their argument and said they were wrong because their current models have a minor limitation.
Try again, saying something remotely convincing. I do think t
Re: (Score:2)
They claim the cars they have been selling since 2016, all current models, will be fully self driving and operating a robotaxi service by the end of the year. In the next 3 months. With no wipers.
Re: (Score:2)
All cameras on Tesla have impervious spell cast on them. Much better than wipers.
I see nothing secret about this. (Score:5, Interesting)
They still haven't replied.
I am shocked, shocked! that they don't respond to a has-been.
-I'll never pay for a smartphone on wheels, where are the damn buttons?
Anti-secret, Tesla gave a whole presentation (Score:4, Interesting)
We know that is how they are operating. Isn't that the point?
Yes in fact, Tesla gave a huge presentation [youtube.com] on self-driving where they explained in great detail about how the Tesla fleet helps train!
So many people would be so much less confused if they simply spent four hours to watch that video.
Yep .. it's Captian .. (Score:2)
.. Obvious day.
Re: (Score:2)
I'm more interested in results than in a video about the theory of how they imagine their system may someday work.
Last month Musk repeated his claim that they will have self driving robotaxis by the end of the year. That's a clear goal we can measure their progress against over the next three months.
Given their history of missing self driving goals I'm not holding my breath.
Re: (Score:2)
Yeah, but it's Cringely, research is below him. He gets paid to pull specious blather out of his nether regions, doing some actual research would put him in the neighborhood of those lowly people who actually qualify for the title "journalist", and we can't have that now, can we?
Re: (Score:2)
It's not a secret. Cringely is on the bad crack again. Musk has specifically pointed to their deployed fleet and its data gathering abilities as their biggest strategic advantage in self driving software.
Re: (Score:3)
As I mentioned in my other comment here. It is not even hidden, go into the menu and it says so right there:
https://youtu.be/vWo_nImCWaY?t... [youtu.be]
Driveways? (Score:2)
Re: (Score:3)
Why every driveway? Once you drive down enough of them to train a robust algorithm, that should be good enough. Humans don't have to drive down every driveway.
Re: (Score:3)
Re: Driveways? (Score:2)
Re: (Score:2)
It cannot work outside of it's training set.
So classification systems can't work with images they've never seen? Train one with images of cats (for example) and it can only recognize cat images it has already seen? I think the technology is well beyond that and has been for some time.
Re: Driveways? (Score:2)
Re: Driveways? (Score:5, Interesting)
By the second DARPA autonomous vehicle challenge [wikipedia.org], the ability to select which parts of the terrain were driveable was basically solved. And we've come a long way from that, where video image processing is able to replace LIDAR. Picking a roadway out of a forest image isn't that difficult. And the military has systems in work that can even find driveable terrain between the trees that fall within a vehicle's capabilities. They don't use LIDAR, as that is an advertisement to get shot when used near a battle zone.
Re: (Score:2)
Re: (Score:2)
Cameras are getting there.Actually, the cameras are ready now. We're just waiting on the software.
Re: (Score:1)
We're just waiting on the software.
Famous last words.
Re: (Score:3)
They use equipment which, if made in bulk for a wide consumer market, wouldn't add much to the price of a car, yes.
At the moment, they are made one-by-one, using discrete logic, top end CPUs, GPUs and expensive FPGAs, but once the design is pinned down and they start being made from ASICs on full production lines, there's nothing there that will be really expensive.
Now, LIDAR is something that might never become cheap enough for consumer autos, but even there there is work happening on solid state devices t
Re: (Score:2)
LIDAR is something that might never become cheap enough for consumer autos
Oops, it already is. https://www.spar3d.com/news/li... [spar3d.com]
Velodyne’s latest sensor was engineered to be an optimal automotive-grade lidar solution for Advanced Driver Assistance Systems (ADAS) and autonomous vehicles, and is priced at only $100.
Re: (Score:2)
Re: (Score:2)
Re: (Score:2)
$100 is stratospheric?
Re: (Score:2)
Velodyne has a lidar that wholesales for under $200.
Cringely, go away (Score:2)
I mean it.
A better question (Score:2)
The thing that isn't explicitly mentioned here is that Tesla must know that IF their FSD had been deployed and used they would know to a pretty high level of precision how many people would have been killed had they been using it. They certainly would know how many accidents would have happened.
That is true of past versions and future versions.
So the question is -- what does their shadow fatality number have to be before they release it?
I would like to say that number is zero and that actually happen
Re: (Score:3)
So have they made any statements as to what their fail number has to be before they ship?
Good question. I'd be more interested in seeing if the NTSB is involved in self driving systems design (probably not much) and what their criteria for certification is. The FAA (for example) have certification criteria for failure analyses and the resulting system effects that result.
Other than that, I suspect that the numbers would be proprietary for reasons of liability. Work out how many lives could be lost per year (which many industries do) and if someone doesn't like the numbers used it's "See you in
Re: (Score:2)
Ask their insurers.
Because this technology isn't going to happen until the burden of insurance moves from the driver to the car manufacturer. At that point, their insurers will actually determine the real risk because they have to pay for its mistakes, rather than lumping all responsibility on the driver.
Imagine, for example, an insurance company that would insure a single person for a single, fixed price, no matter how many accidents they have. Literally "Joe Bloggs" is insured no matter what happens, wh
Re: (Score:2)
You raise an interesting question but I don't think it will happen that way.
Automobile insurance insures the property, not the driver. I get the same coverage if someone steals my car and damages someone else's property with it. You don't go to an insurance company and say "I want to buy a policy that insures me no matter what car I drive." Nobody has a policy for that.
Also, Tesla has its own insurance company now. My car is insured with Tesla insurance. This insurance issue with regard to Autop
Re: (Score:2)
You obviously only have lived in one particular country.
That's exactly how insurance works in many countries.
Re: (Score:2)
zero isn't the correct cutoff--we shouldn't be looking *solely* at the deaths that *would have* occurred but also at those that *would have been avoided*.
Subtract the second from the first, and we get the net loss of lives from the hypothetical self driving. Once this is negative (i.e., net lives saved), it's time to. consider allowing it to happen.
As for insurance . . . it can be split between while self driving, and while under manual control.
Ultimately, it would seem to make more sense to put insurance
Seriously?!?!?! (Score:1)
Cringley states that Tesla runs 2 different versions of software on each car: Production and Test (I have no idea if this is true or not), and that Tesla is running A/B testing. He also states that LTE bandwidth is too low to do a complete data dump.
Does this guy not know/understand what Report by Exception is??!?!?
If Production and Test are performing in the same manner, then all you need to report back is "Yeah, we good". If there is a difference then you just need to know what that difference is. And
Re: (Score:2)
What you wrote is the same as what he wrote. I'm sure you think you're blowing his mind, but you're phrasing his point slightly differently.
How the fuck is that /. material? (Score:2)
Re: (Score:2)
Tesla can retrieve video from their cars, over the air. Musk has said that they can do things like ask for particular things (cars with bikes on the back was the example he gave) and Teslas will send in their videos. I suspect the cars also send back video clips of their WTF moments. Tesla uses those capabilities to train their automated driving software. It's not a secret, they brag about it.
Re: (Score:2)
It's a secret to Cringely, who flaunts his lack of research capabilities at every opportunity. I don't think the guy has ever even heard of a search engine.
Didn't Read... (Score:2)
Moderation needed (Score:2)
The level of insults and profanity on Slashdot is reminiscent of middle school. There are some very good discussion sites I visit where adults have reasonable and civil debates. But that doesn't happen here.
The fact that it does happen elsewhere tells me that whoever runs this site is either incompetent, or just doesn't care. That's why I spend less and less time here. It's just not interesting watching a schoolyard fight.
Re: (Score:2)
What are they?
Slashdot got invaded by trolls and crazies, as the registered users suddenly jumped up over six million.
Privacy (Score:2)
Re: Privacy (Score:2)
Re: (Score:2)
You've mentioned this before, and I'm still puzzled over why you care. You're not parking your car in your bedroom (I assume).
Tesla isn't even remotely close (Score:1)
I own a Tesla, it's a brand new one, I really like the car as a car that a human being drives, I am not a short seller, I am not a troll, I am not a hater. Its beyond absurd that I need to say these things, and post anon, to prevent harassment but this is where we are.
Tesla is nowhere remotely close to full self driving. Their telemetry reports or whatever they are each night are in the hundreds of KB and if they include driving data, it's very very little. I know this because I have no mobile phone coverag
Re: Tesla isn't even remotely close (Score:1)
I am not a short seller,
By trying too hard to convince that you own a Tesla, you've inadvertently revealed that you do not, in fact, own a Tesla.
Re: (Score:2)
I am not a short seller,
By trying too hard to convince that you own a Tesla, you've inadvertently revealed that you do not, in fact, own a Tesla.
As a Tesla owner I can attest that his descriptions and observations are pretty much spot on and it is very easy to believe that he is actually an owner (or at least a regular driver) as well.
But his conclusion that LIDAR is needed is not necessarily correct. I suspect that Tesla does in fact build remarkably high-fidelity 3D models but they don't show them. The car's display of that model makes it look crappy as all hell but it is clearly highly abridged. Anyone going by what that display shows to det
I'm still waiting for RoboTaxis. (Score:3)
Each Tesla Model 3 will make its owners $30,000 a year!
If it was on schedule to be "functionally" ready in 2020 like Musk said, then I'll be ready for the demonstration on Dec 31st. Not.
https://www.usatoday.com/story... [usatoday.com] https://www.cnet.com/roadshow/... [cnet.com]
No! (Score:2)
Not secretly.
Secret? It Openly Stated! (Score:2)
The car tells people that it's gathering data. You can tell it whether or not to send the data back. Musk has talked publicly about this. He mentioned that if they're having trouble with a particular type of situation, they'll have the cars send back video clips so that they can have more data to train for those situations.
The cars are awesome--far better than anything else on the road.
Do they have sufficient vision to drive themselves as robotaxis as hyped? Maybe. There would be an easy way to check: R
Re: (Score:2)
The people who say LIDAR is necessary don't have a clue. LIDAR gives you back a point cloud, possibly with some relationships between the points. It's straightforward to get the same thing from stereo cameras, with higher frame rate.
As you said, the real question is whether we can make software that's reliable and flexible enough to drive as well as a human. That question has nothing to do with LIDAR. Basic sampling theory, a century of experience developing computer algorithms, and most of the objections o
Bob isn't an idiot (Score:3)
But google maps does the same. Sort of. Since long.
It understood from me and my neighbours about a temporary change on a one way street and started suggesting it for directions. Once the change was reverted back it stopped within one day to suggest it.
Whether the two CPUs are used like that, I really dunno, but the data gathering sounds reasonable. A part of the "night" guess which could be not correct: why not using it during the day when the car is on the road and with full 3g/4g coverage?
Shipping a few MB every 10 or 15 minutes looks more reasonable, doesn't it?
Re: (Score:2)
I've seen Google Maps do that. There is a shortcut near us that it did not know about until we started driving it. And now it recommends it to others.
A bit spooky!
Re: (Score:2)
That is very likely " anonymized data usage" as they have no use in identifying the actual users taking that new route.
And it looks like is done accordingly to the EULA. Being that lawful is a different story.
Nonetheless, Google/Tesla service is being improved automagically with users data. This is the one of actual points from Bob X.
It's not a secret (Score:2)
I mean, which makes sense. If any of us were on that dev team it's what we'd do. The stupid headline is just clickbait.
Secretly? They outright tell us they are! (Score:1)
This one's easy (Score:2)
"Every night all those cars wirelessly report their driving data back to Tesla. I would love to know how Tesla decided what to put in those reports."
The answer to "how they decided" was to for them to say, "put all of it in".
Every single bit of data will be packaged up and sent back in one form or another. It's like gold to them.
Interviewed an outgoing Tesla data scientist (Score:1)
He is right Tesla uses real world data to improve the deep learning computer vision algorithms, but wrong basically about everything else.
"AI". Which in this case means ma
Noooo! (Score:2)
Oh great, thanks a lot. Now that Creepy Cringely has made a prediction it's bound to turn out false. I mean the guy only makes "predictions" of totally obvious bound-to-happen well known things, but then somehow even those don't happen. If there were a rock falling down to the Earth, when Cringely sees it he will no doubt declare himself a genius and predict that the rock will hit the ground. But then somehow gravity itself will fail and the rock will disappear into a wormhole.
Yes they are (Score:2)
You can opt out of it in the software.
I took a quick test looking at who it talked to when connected to WiFi at home, shortly after I got the car.
And it send data to some servers in the Amazon cloud. And then there's a bunch of other things that looks like it is because I am using Spotify, Tunein, maps etc.
https://www.youtube.com/watch?... [youtube.com]
RADAR (Score:2)
What is RADAR? I only know of radar.
Re: (Score:2)
Radar is, was and always has been an acronym.
"Radio detection and ranging"
Don't worry, it's only been like that since its invention in 1940, so it's not like you're out of date or anything.
P.S. It's like laser... which is also an acronym and hasn't ever been anything else. It's just that people get bored of typing L.A.S.E.R. if they type it enough, so the acronym gets subsumed into the language as an actual word.
Sigh (Score:2)
How to demonstrate a complete misunderstanding of machine learning, including completely ignoring the fact that all AI plateaus no matter how much training data you have, and that for each new piece of training data you have to "untrain" or "retrain" millions of previous results in context for it to have any significant effect on the outcome.
If throwing data at it made AI intelligent, then Google would own the entire AI market, I-Robot-esque.
That's not how it works. It's a statistical model that - after in
no shit Sherlock... (Score:2)
Massive data advantage vs. Subaru (or others) (Score:2)
AFAICT, Tesla's current self-drive mode is more or less distance sensing cruise control combined with lane centering.
How do we know their data gathering is so much better, than, say Subaru, who has been using stereo cameras in their Eyesight system for several years and who have the same basic self-drive setup (distance sensing cruise/crash avoidance for years, lane keep for at least a couple, and lane centering recently)?
Subaru has the advantage that they sold twice as many cars in 2019 as Tesla. They ma
They'd be stupid if they weren't. (Score:2)
Driving is a rote, repetitive task that takes a fair amount of multi-input data to derive a result, interrupted by highly unpredictable, sudden, highly variable emergencies.
An 8 year old can drive a car down the highway (some have) on a clear dry day on an empty road, it's the "driving on an icy road at night in traffic full of people going too fast when the semi coming the other way jackknifes across the median at you" that training and experience will HOPEFULLY equip a driver with the reflexes that will i
Yeah, good luck (Score:3)
I got my Model 3 at the end of July. I do one long drive to the cottage, 250 km one way, so I use Autopilot -- not "self-driving", just the fancy cruise control.
AP hits the brakes at random times. Hard. Most of the time you can figure out why, even though it's clearly wrong. For instance, there's a turn on one highway that's sharp enough that they put those "sharp turn" indicators on the top of the don't-drive-into-the-ditch poles-n-wires. The system sees these signs as traffic cones and hits the brakes for the second while they are in front, and then hits the accelerator just as hard when you turn a bit.
But there's other times where I have zero idea why it does something. I was driving down the 401 in low traffic with the nearest car at least 20 car-lengths in front when it suddenly hit the brakes for a second. In another case, it hit them when it saw a flashing amber light above an intersection. And it hits them hard enough that anyone following close behind me will hit me.
Even when everything is working fine, its still crap as a pure cruise control system. If the car in front of you changes lanes, AP will stomp the accelerator to bring you up to the next car in front. If there's any variance in traffic at all it accelerates and slows back and forth. It's *way* less smooth than my right foot on the accelerator and gets way less energy efficiency. AP gets at least 5% less miles on the highway than I do.
When I sit in the passenger seat, I get to watch the system look at the traffic. It's downright scary. For one thing, when you are following a curve, even a relatively mild one you might see on an interstate, all of the traffic on the outside of the curve disappears. You know, like the oncoming traffic. When it does see that traffic, it often gets confused and displays a car that just drove past going in the opposite direction as sitting still beside you going in the same direction trying to pass. Crossing traffic in front only shows up maybe, *maybe*, 25% of the time, and it rarely if ever sees pedestrians.
So don't give me this about all the data, its crap. There's no way this is anywhere remotely ready for FSD.
Mootish (Score:1)
Tesla is not really pushing Autopilot for cars.
Musk knows the big money is in replacing truck drivers so they can reduce costs in the supply chain by laying off a bunch of people.
The secret question is: (Score:2)
Do Slashdot editors secretly use the word secret to increase clicks when there's nothing secretly happening?
Tesla is already vertically integrated to a fault. (Score:3)
Tesla makes lots of their own stuff - more than pretty much any other auto maker. Go to your GM, toyota, et.al car and count off how much of it is made by Bosch.
Unlike almost any other vendor, Tesla makes their own driver assist hardware. Using Telsa designed silicon.
Tesla doesn't use LIDAR. So I don't know where you are getting your information from.
It makes sense for them to outsource their radar sensor hardware. Those things are off-the-shelf hardware, but Tesla is doing things with them that others are