Tesla's Elon Musk Promises Full Self-Driving Autopilot Beta In 'A Month Or So' (cnet.com) 94
"I think we'll hopefully release a private beta of Autopilot — the full self-driving version of autopilot — in, I think a month or so?" CEO Elon Musk said this week at Tesla's annual shareholder meeting/Battery Day event. "And then people will really understand the magnitude of the change," said Musk adding, "It's profound. You'll see what it's like, it's amazing."
CNET reports that attendees then showed their approval "by honking the horns of their safety bubbles." "It's kind of hard for people to judge the progress of Autopilot," Musk told a crowd of shareholders present at the event, each social distancing in their own Tesla Model 3, drive-in style. "I'm driving a bleeding edge, alpha build of Autopilot, so I sort of have insight into what is going on."
Musk went on to explain how Tesla's engineers recently had to overhaul major parts of the Autopilot, including a rethinking of how the system sees the world. "We had to do a fundamental rewrite of the entire Autopilot software stack... We're now labeling 3D video, which is hugely different from when we were previously labeling single 2D images," Musk explained, referring to the way the Autopilot software understands what the objects it sees with its eight cameras are, and how it should react to them. "We're now labeling entire video segments, taking all cameras simultaneously and labeling that. The sophistication of the neural net of the car and the overall logic of the car is improved dramatically."
CNET reports that attendees then showed their approval "by honking the horns of their safety bubbles." "It's kind of hard for people to judge the progress of Autopilot," Musk told a crowd of shareholders present at the event, each social distancing in their own Tesla Model 3, drive-in style. "I'm driving a bleeding edge, alpha build of Autopilot, so I sort of have insight into what is going on."
Musk went on to explain how Tesla's engineers recently had to overhaul major parts of the Autopilot, including a rethinking of how the system sees the world. "We had to do a fundamental rewrite of the entire Autopilot software stack... We're now labeling 3D video, which is hugely different from when we were previously labeling single 2D images," Musk explained, referring to the way the Autopilot software understands what the objects it sees with its eight cameras are, and how it should react to them. "We're now labeling entire video segments, taking all cameras simultaneously and labeling that. The sophistication of the neural net of the car and the overall logic of the car is improved dramatically."
Really soon, promise (Score:5, Insightful)
Tesla's Elon Musk Promises Full Self-Driving Autopilot Beta In 'A Month Or So'
Heard that before. Guessing the timeline will be like Trump's ACA replacement. Just sayin' ...
Re:Really soon, promise (Score:4, Funny)
Keep shorting their stock, you're sure to win eventually...
Re: Really soon, promise (Score:4, Interesting)
Transfer autopilot to a new vehicle (Score:3)
Re: Transfer autopilot to a new vehicle (Score:5, Informative)
Re: Transfer autopilot to a new vehicle (Score:4, Informative)
I'm fine with what I got for 2015 (although I hoped for a much better summoning experience). But for my 2017 model I was promised a full self-drive capability and right now it's not even close.
Re: (Score:3)
Navigate on autopilot was an FSD feature.
Depends on when you bought your car. Prior to 2019, it was actually an Enhanced Autopilot feature, not an FSD feature. It became an FSD feature in 2019, when they split up the EAP features, moving some of them up to FSD and the rest down into the base AP configuration.
The only thing pre-2019 purchasers have gotten so far from their FSD purchase is the HW3 upgrade, if that.
Re: (Score:3)
The original Full Self Driving they were selling back in 2016 included the following features:
- Summon the car from the other side of the country, it will drive 1000s of miles by itself, charging and cleaning itself as needed, to your location.
- Drive you to work while you read the paper.
- Go off and find a parking space by itself, then come back to collect you later.
More recently Musk added:
- Robotaxi service, your car will earn you money while you are not using it and become an appreciating asset, worth m
Re:Transfer autopilot to a new vehicle (Score:4, Informative)
My spouse bought her Model-S back in 2015, so we are already past the 5-year point.
She paid extra for Autopilot, but never got FSD. However, she still got her money's worth. Even normal Autopilot is a big win for highway driving.
Re: (Score:2)
Re: (Score:3)
Well, you know what they (should) say about AI with human lives on the line: the last 1% of the development takes 99% of the time.
Re: Really soon, promise (Score:2)
Alas, thatâ(TM)s not how it works. Either the model will work or they will have to rewrite and retrain it again. So itâ(TM)s more like to 80%, then back to 0%, then to 85% then to 0%, rinse repeat.
Re: (Score:2)
True, but AI development is a lot more than training - it's also all the learning done by the design team in developing the training regime and underlying neural architecture.
Re: (Score:3, Insightful)
It's also a problem gaining acceptance that it's not perfect- just much better than most human drivers.
Re: (Score:2)
Re: (Score:3)
Re: (Score:2)
>There is no last 1% because there is no state of completion
Neither is there for virtually any software - but there's still a finish line of "good enough to ship".
Unfortunately for Tesla, a full self-driving autopilot won't be able to pull the *very* common software dodge of "not warranted as suitable for any purpose, including those for which it was marketed"
Re: (Score:1)
Re: (Score:2)
Re: (Score:2)
I don't know about you, but I do not drive safely while I'm napping.
Re:Really soon, promise (Score:5, Interesting)
It's been a frothy stock in recent weeks. I just trade the bounces these days rather than holding. The number of people making the optimum or worst case bet must be minuscule compared to the misguided shorters who have been shorting for the past few years and the many happy longs who got it back in the 200-400 range.
Tesla stock has paid for my model 3 and my cybertruck order - I'm keeping those gains in cash since they are technically spent.
I goes up and down like any stock, it's just had a good run recently based on a compelling growth story.
Re: (Score:3)
"I think we'll hopefully...." (Score:3)
Re:"I think we'll hopefully...." (Score:4, Insightful)
Doesn't exactly sound like a "promise."
(Consumers reacting to a CEO statement from a PRIVATE company): "Wow, that sounds pretty cool."
(Consumers reacting to a CEO statement from a PUBLIC company): "Yeah whatever bitch. Where's my money?"
Sometimes I wonder just how much greed, taints people's reactions.
Self-driving cars have been in our dreams since flying cars. Don't suppose you could put aside the pointless vitriol and just admire the progress, or would you prefer to keep worrying about your loved ones being killed by a fucking social media junkie who just had to answer that text message?
Be patient. Intelligent humans are trying to solve for Mass Ignorance. That ain't fuckin' easy. Humans are really good at it.
Re: (Score:3, Interesting)
He didn't define "full self driving". Originally it was level 5, go anywhere without a driver, charge itself etc. At other times he has suggested that at first it will just be level 3.
This seems like an admission that the robotaxi service won't launch this year as promised.
Re: (Score:2)
Those inbetween levels should be banned, either the driver should drive the car or the computer should drive the car otherwise the human will fall asleep, text, watch videos etc. It's already happened and people have already died from not understanding the inbetween levels.
And self-driving cars should be tested to show themself to be proficient at understanding every single aspect of every single (UK) highway code. Or else they don't pass the test and they don't get to drive until fixed.
Re:"I think we'll hopefully...." (Score:5, Insightful)
Nope.
They just have to be safer than existing human drivers. Perfection is the enemy of the good.
Also some situations different from others. I'd say driving down freeways in good weather they could be 100%.
But yes, the in between can be very dangerous.
Re:"I think we'll hopefully...." (Score:5, Insightful)
They need to be damn close to perfect if they're to be trusted and insured.
The programming company should become the driver. If the programming company accepts liability, people will trust it. Currently that liability is distributed among millions upon millions of drivers. Even if US deaths are dropped by 90%, that's 4,000 deaths per year centralized on a couple companies.
If the programming company defers liability to the non-driving human, people will simply not trust the vehicle.
These are the non-programming issues that still aren't sorted that can destroy companies and sub-industries of their not sorted prior to level 5 AVs hitting the road.
Re: (Score:2, Insightful)
The programming company should become the driver. If the programming company accepts liability, people will trust it.
That shell company in the Caymans? Not sure many will trust it.
Re: (Score:3, Insightful)
If a car can't follow the highway code then it is questionable as to whether it can drive safely and in a manner predictable to other drivers, riders, cyclists and pedestrians.
The standard of safety shouldn't simply be human drivers, self-driving vehicles should be as safe as patient drivers who are not tired and not under the influence of drink or drugs.
Re: (Score:1)
Nope.
They just have to be safer than existing human drivers. Perfection is the enemy of the good.
Also some situations different from others. I'd say driving down freeways in good weather they could be 100%.
But yes, the in between can be very dangerous.
No, that's not how legal liability works. The courts look at each accident in isolation to work out who was in the wrong. The fact that a driver is, statistically speaking, a safer than average driver doesn't matter when it comes to legal liability.
Societies will have to decide how to handle liability for autonomous software. Imagine your loved one is killed by an autonomous vehicle. Do you just shrug your shoulders and say "Shit happens"?
Re: (Score:3)
But that's not "full self driving", that's just level 2 again. Level 2 requires a human driver ready to take over, if say the weather changes or there is some other event that the car can't handle. Tesla specifically sold level 5, they made that extremely clear so must now deliver it or have broken their contract.
I imagine investors will be rather upset when the promised robotaxi revenues fail to materialize and someone else beats them to it, e.g. Waymo.
There are already cars with better level 2 than Tesla,
Re: (Score:2)
I imagine investors will be rather upset when the promised robotaxi revenues fail to materialize and someone else beats them to it, e.g. Waymo.
Waymo was already running minimally supervised robotaxis six months before Musk announced that plan, and fully driverless robotaxis almost a year ago. Don't you think that ship has already sailed? I mean okay, maybe not the revenues, but that's more a matter of scale rather than readiness.
Re: (Score:2)
I mean first to market with a widely deployed non-beta product.
Re: (Score:3)
Those inbetween levels should be banned
Having used the current Autopilot for a while I disagree, but even ignoring that, I don't think you can argue that level 4 isn't perfectly safe. Level 4 is complete autonomy -- driver doesn't have to be paying attention, or be awake, or even at the wheel -- but not in all conditions/environments. A level 4 vehicle has to be able to recognize when it's entering a situation it can't handle and if the driver can't be called to come take over, it has to be able to bring itself to a safe stop out of traffic.
I'
Just don't call it autopilot (Score:2)
And honestly, even a freeway-only level 4 AI would be a *huge* boon to a lot of people by turning much of a frustrating commute into unstructured "me time".
I just really hope they come up with a new name for full self driving mode - "autopilot" is already confusing enough people without calling the version that *can* drive on its own the same thing.
Re: (Score:2)
Re: (Score:2)
I'm not holding my breath, but I also wouldn't be shocked if Tesla is only a few months away from a freeway-driving-only level 4.
I remember a way old TV show - think it was a pilot, maybe for a show that didn’t get picked up - where the main character was a LA private detective in the somewhat near future (think this was the 80s, so “near future” was probably right about now). I remember a scene where the dude was cruising down the freeway, dictating a letter or maybe in a call, leaning back with his feet up near the left-side view mirror because the car was doing everything. Then at one point the car announced
Re: (Score:3)
Freeway-only autonomy would solve a LOT of problems.
Heh. It will be very convenient, and I'll be happy to have it, but I expect it to create a lot of problems. People will commute farther, and more. People will probably fly less, too, moving more medium-distance transit to the highways. I think I would drive instead of fly when I go to California for work: I'd get in my car at about 9 PM, lay the seat back and wake up in the bay area in time to get a shower and have breakfast, then start my day (we need fully-automated charging stations to completely realiz
Re: (Score:2)
Potentially, but I don't think we'll see huge changes in carrying capacity until the human drivers are mostly or entirely removed.
No argument there. I was thinking about the situation once pretty much all vehicles have the technology. And, even then, there would have to be an actual requirement for its use if people want to see the biggest benefits - which I imagine will be hard-to-impossible to legislate.
Re: (Score:2)
It *can't* be a promise. Even if it were technically available, laws would need to be changed.
Re: (Score:2)
Re: (Score:1)
Re: (Score:1)
Re: (Score:2)
Police can tell people to stop and pull over right now. Try not doing it, and see what happens.
Lots of bad things. [google.com]
Re: (Score:2)
Police can tell people to stop and pull over right now. Try not doing it, and see what happens.
Facts. I highly doubt Tesla is teaching their autopilot how to avoid a PIT maneuver.
And that $300 police cruiser bumper is going to destroy a $30,000 Tesla bumper. You'd have to be stupid and rich.
Unsettling... (Score:5, Insightful)
There's something unsettling about the words "self-driving" and "beta" together in the same sentence...
Re:Unsettling... (Score:4, Interesting)
Well, at least from the news reports... It seems like when a Tesla autopilot fails, it’s the Tesla driver who pays the price. So you could argue any danger is “opt in”.
Re:Unsettling... (Score:4, Insightful)
That is, if you can find anyone actually on the Vegas Strip during COVID hysteria.
Yeah, hysteria. Over 200,000 dead in six months. The third most lethal medical reason people die in this country, and it's communicable to others.
Re: Unsettling... (Score:3)
It is though? I own two Teslaâ(TM)s and although I love LOVE the cars, including Autopilot, I feel like itâ(TM)s always been BETA. You always have to keep an eye on it because although it drives great most of the time it gets confused by many lane highways and exists with odd markings.
Re: (Score:2)
More like, every time a Tesla autopilot fails, the company says it's the drivers fault since they didn't babysit the car at all times. Typical Silicon Valley mentality. If the car doesn't drive itself properly, you're driving it wrong.
But then, I'm not part of Tesla's audience. I don't want a car to turn on the headlights or windshield wipers automatically, let alone drive itself.
Re: (Score:3)
I have a fully equipped Model 3 and Autopilot really is improving rapidly.
Re: (Score:3)
have a fully equipped Model 3 and Autopilot really is improving rapidly.
Same here. It's good. But it really does need a re-write to improve its awareness. I only trust it not to make poor decisions on relatively clear highways.
Far more unsettling (Score:1)
There's something unsettling about the words "self-driving" and "beta" together in the same sentence...
Have you ever met a 17-year old? They are able to drive legally in all states as far as I am aware....
A 17-year old is VERY much a beta human, and we let all of them drive today - by the millions. They simply have somewhat higher insurance rates.
Imagine five years after full self driving is released. What will be more expensive to insure... the FSD car, or a 17 year old? Or a *60* year old, for that ma
Re: Far more unsettling (Score:2)
You can drive a car at 15 in the US with someone in the car with you, alone at 16.
Re: (Score:2)
A 17-year old is VERY much a beta human
Beta? Those are the mature few. Most are more like early alpha (and, looking back - that probably included me).
Re: Far more unsettling (Score:2)
Nope, inaccurate. (Score:1)
At least a 17 yo can see a big white truck overturned on the highway.
That happens way more often [google.com] with teenage drivers than With Teslas...
Tesla still beats humans there,
Re: (Score:2)
Re: (Score:2)
Even after betas, I don't feel comfortable with this feature. It's not stable as KITT and KARR. :P
Re: (Score:2)
It amazes me that our government actually allows these things on the road with other people.
There are so many points of failure with the hardware, software, image recognition as well as our completely unpredictable road markings and characteristics.
Re: Unsettling... (Score:2)
It amazes me that the governments allows teenage drivers on the road at 15/16 years of age honestly but we survive them and weâ(TM)ll survive this.
Re: (Score:2)
Re: (Score:2)
Just make the self-driving beta wear the "student driver" warning
"fundamental rewrite" (Score:3)
Wow
That sounds scary to me - don't get me wrong, I love the idea of a self driving car and I know Tesla's are, in many respects, excellent cars, but at least autopilot worked pretty well generally
The fact that much of the underlying code was rewritten for this version scares me that new unforeseen issues will arise when it's released into the real world
I don't like having QA done by the end user - that's like putting Windows 10 on the freeway
Re: (Score:2)
Re: (Score:2)
I hope the data they have is fully testable though
I mean, I suppose they would theoretically need the actual video from the cameras on those cars for millions of miles as opposed to what the cameras interpreted in those miles since the new software interprets things differently
I doubt they are retaining all of the video from those cars?
That said, this is all really just BS as I don't know the details of their internal processes, they probably have some reasonable process to test this against at least some r
Re: (Score:2)
Each Tesla has two AI chips [cnet.com].
It was my understanding that Tesla was able to load new algorithms into one of them and have it "play along" while a driver was operating the car. They could even do this while Autopilot is not engaged (since the new alg plays along without adjusting any physical inputs of the car).
So it's entirely possible for them to test the software in existing cars without endangering anybody. They know the new algorithm beats the old one when it causes/avoids fewer crashes as it's playing
Probably semantics (Score:2)
My guess is this beta will look a lot different from what most people are expecting. You won't be able to hop in your car at home and pull up in a parking spot at your destination. There will probably be a number of minor things that still aren't accounted for as well. "Full self driving in a month or so" could simply mean that AP can now attempt to make turns at lights and stop & go through stop signs.
This is similar to the AP investor day stuff he talked about. He said that "AP will be feature complet
Right (Score:5, Interesting)
It really has to be a massive improvement. Today autopilot(just using cruise control) gets scared and brakes when it sees a car parked by the side of the road(outside the lines) and in a lot of places it feels annoying and I just switch it off. It works keeping distance to the car in front of me on the freeway but it’s behaviours doesn’t really inspire me to pay $8000 for the upgrade to full autopilot, I get a smoother ride just controlling the speed myself.
It does seem to work well when a faster car drives past and pulls into the right lane too close in front of me, where it does not break to increase the distance. Well I don’t drive much on freeways so it isn’t a huge issue for me.
Re: (Score:2)
That kind of thing seems like it would need car-to-car communication to really fix. As things are, the system can't tell if that car in the shoulder has the motor on, ready to dart out in front of you, or if it's disabled and in park. I slow down at least a bit in that scenario anyway. As of a couple years ago, it's actually state law in Texas to slow down by 20 MPH when passing a vehicle on the shoulder. (But only if it's a police vehicle. Blue lives matter, I suppose.)
You'd need to start removing "normal"
And (Score:1)
Even more idiot Tesla owners will die.
Re: (Score:1)
the current version *saves* lives.
https://cleantechnica.com/2020... [cleantechnica.com]
Re:And (Score:4)
Fun little fact left out is the number Tesla's on the road (just over 500k) vs the number of total cars on the roads (273.6 million) in the US. This is the same fallacy all self-driving proponents gloss over when they site safety statistics. The number of "autopilot" miles is dwarfed by the number of miles by human drivers per year that it is statistically insignificant.
Re: (Score:2, Flamebait)
that's not what "statistically (in)significant" means, shit-for-brains.
Re: (Score:2)
Lol OK buddy.
Re: (Score:1)
Sensors (Score:2)
I think what htey are doing is great, but I feel like the side facing cameras being on the B-pillars are too far back to detect front cross traffic early. When pulling into an intersection, especially in many urban areas the car would have to jut out too far into the intersection to get a proper view of cross traffic. They need an extra pair of cameras that are placed more forward .. maybe just behind the headlights or in that groove in the front bumper under where the front license plate is. Rear cross tra
Re: (Score:2)
people are doing fine with just 2 low resolution vision devices.
Re: Sensors (Score:3)
Actually eyes are much higher resolution with 100x more dynamic range and have better placement than the 8 cameras on a Tesla when it comes to seeing forward cross traffic. Especially when you are at an intersection of a one way road with traffic coming from the right. The B-pillar on the right is much further back than your eyes are when youre sitting in the driver seat.
WHere are the regulators...? (Score:1)
Well if Boeing can fly the max without regulator . (Score:1)
White trucks (Score:2)
Hopefully, maybe, in a month or so, I think. Beta (Score:1)
In other words nothing will happen within an unknown timeframe
Dogfooding? What's he mean 'driving'? (Score:2)
"I'm driving a bleeding edge, alpha build of Autopilot, so I sort of have insight into what is going on."
Does that mean Musk is actually driving around in a Tesla with an autopilot like that? Or just testing it on a closed course? The difference in required intestinal fortitude is massive...
This missing factor (Score:3)
So this article has about 80 comments so far and although there are many opinions on how "safe" Tesla FSD might be none of them mention what the data Tesla has actually shows.
Their shadow mode tells them pretty accurately just how many accidents their FSD would have been involved in had it been in control. From this they will also likely tell how serious those accidents might be and how many of them if they would have caused injury or maybe fatality.
Real world data, real world conditions. All the "aha I got you" arguments about wipers, LIDAR, 3D object modeling, AI, Musk on weed, yadda yadda is pretty much useless against hard data. They know. Musk knows.
I can understand why they don't publish that data. But sooner or later they will release this software for end-user use because their data shows them that the risk is acceptable. No doubt some would argue that the FSD risk is better than a human driver risk. Personally I don't argue that because it requires too many assumptions.
What I do believe is that also sooner or later if they release FSD that data will be forced out into the open. There will be some accident, a lawsuit, and the plaintiffs will argue for discovery that the data must be turned over. That will be appealed and the defendants will lose. Then Tesla is going to have a "some of you are going to die, but that's a sacrifice I'm willing to make" type of moment.
This isn't a legal jeopardy situation unique to Tesla or this situation. There are other FSD competitors out there, one with a fatality already on the books. All vehicles have potentially fatal engineering compromises like this. Just not as visible or dramatic.
Re: (Score:3)
The problem with real world data is that there is no way to know whether its coverage is adequate. It's entirely possible that somewhere there is a road where every single Tesla that drives it goes straight over a cliff. Even if that means only 5 out of 50 billion trips goes over, it's going to make big news.
Re: (Score:1)
Death and destruction (Score:3)
This sounds promising, but .... (Score:2)
As someone who has now owned 3 different (used) Teslas ... a Model S, Model X and now a Model 3 Performance? I've had the chance to use AP1, AP 2 and the latest/greatest FSD software and hardware.
I feel like really, AP1 (Intel's MobilEye that Tesla just licensed, back then) was an amazingly good, capable product given what it had to work with. Tesla spent a lot of time and money just trying to re-invent a parallel system when they decided they wanted to end the Intel licensing deal and do it themselves. An