Slashdot Mirror
Tesla Shifts the Goalposts For 'Full Self-Driving' Technology (arstechnica.com)
AmiMoJo writes: Tesla has been selling "full self-driving" capability since 2016, promising that "you will be able to summon your Tesla from pretty much anywhere," and that "once it picks you up, you will be able to sleep, read or do anything else en route [sic] to your destination." Last week Tesla shifted the goalposts, redefining "full self-driving" as a number of Level 2 driver assistance features that were already available, and a few new tricks to be delivered later. All will require a qualified driver behind the wheel, paying attention at all times and ready to take over if the car can't handle the situation. Worse, owners who bought the previous full self-driving feature paid $8,000 for it. Tesla is now offering owners who bought their cars prior to the change the same package for $5,000. Owners who paid the $3,000 higher price are unsure if the previously promised technology has been abandoned and Level 2 is now the most they can expect.
What's wrong with "en route"? Don't tell me - a cretinous AMERICAN didn't understand the language. What's new?
Aye, it may be impossible the way Tesla is trying to do it. Their original plan was for a coast-to-coast demo in 2017, which obviously failed.
Other self driving systems like Google/Waymo's one use lidar, cameras, radar and ultrasonic sensors. They are anticipating the cost/size of lidar systems to reduce rapidly in the next few years.
If Tesla had managed to use just cameras, radar and ultrasonics. It would have been a huge coup if it had worked.
Their problem is twofold. First they underestimated the processing power needed to do handle images from the cameras. They use neural nets to process them and on the original hardware they shipped (known as AP2) it just wasn't powerful enough, they couldn't even get it to compare consecutive images (which helps when you don't have stereo vision). They went to AP2.5 and now AP3, but it's not clear if even that is fast enough for what they want to do.
The second problem is that it's just really, really hard to use neural nets to do everything they need. Not just recognizing objects like cars, signs and traffic lights. It has to see road markings, it has to see traffic police and understand their gestures, it has to understand complex 3D spaces with no/poor road markings like car parks and private driveways. It has to be able to recognize small objects that the radar/ultrasonics close to the ground won't pick up, like toll barriers and the over-hanging rear ends of trucks.
To give you some idea of how far away they are, even the current driver assist parking isn't good enough for full self driving. Sometimes it ends up a metre away from the kerb. The human driver can fix that, but for full self driving they have to get the camera to recognize the kerb, indistinct as it may be, and get close to it. Worse still, the current side facing cameras don't point far enough down to actually see it close to the car, so it has to see it from a distance, make a 3D model of the parking spot and navigate into it from memory.
const int one = 65536; (Silvermoon, Texture.cs)
SJW, n: "Someone I don't like, and by the way I'm a fuckwit" - AC
Musk has been very successful in getting Tesla treated like Kickstarter - people paying money, $8,000 for this software, thousands to reserve a car, for things that did not exist at the time. Usually using similar motivations as kickstarter - preordering because they like the company and want it to exist even more than because they want the product. Man, I wish I had that salesmanship.
Your ad here. Ask me how!
These are referring to autonomy levels, not versions. They are defined by the federal government (at least in the US). Level 5 is what all non-tech people imagine. "Car, take me to work. I'm going to sleep now". Level 0 tops out at something like ABS. Level 1 is something like cruise control or lane assist (but not both). Level 2 is both, or Tesla's autopilot. The car can maintain speed and steer, but the driver must be ready to take control back at any time. Level 3 is the car drives itself and asks for help when it needs you to take over (if traffic is crazy or the rain is messing with LIDAR), so you can read a book til then and not pay attention. Level 4 is fully autonomous but it has limitations known at purchase time. And Level 5 drives as well as you.
So, yeah, it's meaningful.
Your ad here. Ask me how!
I know as a techie that humans tend to be really bad at statistics and often over estimate the abilities of technology and underestimate the abilities of humans. Techies are especially bad about this.
You seem to forget that self driving cars "freak up" about once every 600 miles right now, humans "freak up" about once every 150,000. And that human number includes all the very worst drivers driving in all the very worst conditions. That self driving car number is them operating only in the best conditions. Fact is, the average human driver will only be in a handful of accidents in their life time and will never be in a severe injury or fatal accident.
Humans tend to be very bad about understanding rare occurrences in large populations. Yes, somebody dies in a car wreck every day. The chances of you dying in a car wreck ever are very small.
I've posted this a few times: I never understood why Tesla pursued self driving so vigorously. In my mind, a really nice electric car was groundbreaking enough that I didn't see the need, and I saw a lot of downsides.
One downside is certainly that I didn't think they could pull off FSD ever. When I got my Model 3 last October and saw how poorly Autopilot worked, I couldn't believe Tesla ever believed they could improve it enough to FSD. They need many orders of magnitude improvement before they'll be able to turn it loose on city streets by itself. Waymo seems to have the strongest story, and I think they're still 15-20 years away from a coast to coast drive without intervention.
Another huge downside is that FSD is a bet your company proposal. First there are all the lawsuits if you can't make it work... But even worse is the liability. And the more cars on the road, the worse the liability gets. Every time a pedestrian gets hit, there goes millions of dollars. Every time the car runs itself into a truck and kills the occupants, more millions of dollars. Aviation went through a phase where half the cost of a GA aircraft was for the liability insurance. I could see that happening for automobiles as well.
I don't see that they have any choice but to immediately refund everybody who paid for FSD. It'll cost them a lot more if they have to be sued for it. And they'll still get sued... they might end up having to buy back some cars from people who claim they wouldn't have bought the car if it wasn't for the FSD promises. Cheaper to buy the car than go to court.
Right now seems to be one of the more difficult times for Tesla. Certainly their announcement of closing all their stores worries me. And I really like Elon (being an engineer myself I appreciate his humor and way of looking at things). But I have to say, I think it was a huge mistake for him to have gone down the FSD pathway. He should have partnered with Waymo with no promises of the technology ever making it into a Tesla... It's one thing to overpromise a bit on schedules to push the workforce... that's pretty common in high tech. But overpromising stuff like FSD just gets you sued. I hope Tesla survives.