Tesla autopilot HW3

[daniel]

IMHO...

If you are going to own a Tesla and be happy, you have to reset your time scale to a couple of years at a time, a time scale of a year or even months won't work out so well for you. You'll find a way to be perpetually miserable.

I bought my Model 3 with EAP for what it could do when I bought it, not for what it might be able to do in the future. Now it's even better, and I love it for what it does now. Maybe it will get even better and I'll like it even more. But I love it for what it does now. Yesterday driving home from Wailuku with EAP engaged, I had a big smile on my face because it was so much easier and nicer making that drive with EAP than it was with the Prius.

It's not a self-driving car and probably won't ever be. But it is self-driving enough for me, for now. On narrow city streets such as South Kihei Rd. I often disengage AP to avoid an obstacle or give more space to a bicyclist. But it's still better than the dinosaur of a Prius with its 14-year-old technology, which I sold once the 3 got here.

My time scale is today, and this car definitely does not make me miserable.

 

[CarlK]

True. I was watching the PBS News Hour the other day and one guest was saying that 5G networks would be necessary for autonomous vehicles. I'm hoping that he meant vehicle-to-vehicle communications, and not some sort of "cloud-autonomy" where a car is just equipped with cameras and a 5G antenna, and all the decision-making processing is done in the cloud.

Neural net training may be done at the server level but neural net does reside in the car. So are decisions and driving instructions to be made in real time. Whatever that person meant he was just cluless.

There is also no such thing as V2V communications for self driving technology even some cluless people seem to be advocating it. You can't rely on that to make your technology to work. It will collapse if only one car at the stop sign that could not commnuncate with you electronically. It's even worse than to rely on HD map. Tesla, as Karpathy has described, does use deep learning to try predict other car's intentions from slight movements they make. This is just how we do when we're driving and the only way it could work.

 

[mongo]

2. Also, the new SoC was billed as built in conformity to "AEC Q100", where does that fit in?

It's a quality/ records/ tractability standard. Nothing FSD specific.

 

[favo]

Neural net training may be done at the server level but neural net does reside in the car. So are decisions and driving instructions to be made in real time. Whatever that person meant he was just cluless.

There is also no such thing as V2V communications for self driving technology even some cluless people seem to be advocating it. You can't rely on that to make your technology to work. It will collapse if only one car at the stop sign that could not commnuncate with you electronically. It's even worse than to rely on HD map. Tesla, as Karpathy has described, does use deep learning to try predict other car's intentions from slight movements they make. This is just how we do when we're driving and the only way it could work.

There was a video with Geohot a couple of years ago where he was very negative about V2V and said it couldn't be used because it isn't reliable enough and wouldn't be any time soon (he may have said never, but never is a long time). I imagine at some point, cars will report obstacles, wrecks, potholes, and other anomalies, much they way "real-time" traffic is reported now. Not sure if that needs to be V2V or could remain going from vehicle to server to vehicle. People always bring up vehicle platooning (or trains) where the vehicles use V2V to coordinate. (Tesla mentioned 3-truck platoons with a single lead human driver and two driverless followers at the Semi presentation.) This might make sense, but obviously you don't want crashes to result if communications drop out for a second (or less).

 

[dgatwood]

At some point, vehicle to vehicle communications will be a big deal. That will be yet another HW update, but I don't think we have to put that on our 'radar screen' until 3-5 yrs from now, but it is coming.

Vehicle-to-vehicle comms have an inherent trust problem. Accordingly, my Tesla reports itself as being an 18-wheeler towing a space shuttle, going the opposite direction at 825 MPH with its left flasher permanently stuck on.

Not sure if that needs to be V2V or could remain going from vehicle to server to vehicle. People always bring up vehicle platooning (or trains) where the vehicles use V2V to coordinate.

A server-based approach is more robust against attack, because each device has its own identifier, and reports can be correlated across multiple vehicles at that location and validated. A V2V approach is just asking for someone to install a base station that reports the speed limit as 125 MPH through a school zone, or reports that a vehicle is stopped, and to brake as hard as you can, immediately, in the middle of an empty freeway.

 

[CarlK]

There is this fundamental problem which is there will always be cars on the road that do not have the capability even if you do as some suggested to have an universal standard and make it mandatory in every new car. Your system would still need to be able to handle any car that you can't communicate with. You can't say my car will be safe at intersections if all other cars are equipped with V2V and not so safe if some of them aren't. It's like belling the cat thing who is going to put bells a ALL cats? And if your system can handle exceptions why you need the capability at all? It's pretty similar to the case against HD map. In some special situations like platooning or shuttles in a closed community where only one type of vehicles can run it will still have its use. Not for general autonomy though.

 

[tmoz]

There is this fundamental problem which is there will always be cars on the road that do not have the capability even if you do as some suggested to have an universal standard and make it mandatory in every new car. Your system would still need to be able to handle any car that you can't communicate with. You can't say my car will be safe at intersections if all other cars are equipped with V2V and not so safe if some of them aren't. It's like belling the cat thing who is going to put bells a ALL cats? And if your system can handle exceptions why you need the capability at all? It's pretty similar to the case against HD map. In some special situations like platooning or shuttles in a closed community where only one type of vehicles can run it will still have its use. Not for general autonomy though.

I would imagine that you don't rely upon V2V exclusively, but is an enhancement that might provide favorable cooperation.

 

[mongo]

I would imagine that you don't rely upon V2V exclusively, but is an enhancement that might provide favorable cooperation.

If it is an enhancement to FSD, it is not needed and therefore is a waste of resources to add as FSD must/ does function fine without it.

At the vehicle to vehicle level case, turn signals are adequate at communicating intentions.

If you are talking about traffic grid level smart traffic routing/ signal control, that is a different subject where knowing vehicle destinations is helpful (vehicle to infrastructure).

 

[R1Bill]

I noticed that FSD is now available as a $6k upgrade on my account. Does anyone know what it entails? Hardware etc? Did anyone order it and get a timeline?

 

[diplomat33]

I noticed that FSD is now available as a $6k upgrade on my account. Does anyone know what it entails? Hardware etc? Did anyone order it and get a timeline?

What car do you have? If you have a Tesla with AP2 or AP2.5, you will get the AP3 upgrade + FSD software whenever it is released. If you already have a new Tesla with AP3, then you will get the FSD software whenever it is released. Also, if you have EAP, you already have some of the FSD features like Nav on AP, Auto Park, and Summon. So purchasing FSD now will only get you the future FSD features listed as coming later this year, whenever they are released. Essentially, purchasing FSD now won't get you anything right now, but will get you the AP3 upgrade and a future software update later. However, If you only have the new basic AP, then purchasing FSD will get you Nav on AP, Auto lane Change, Auto park and Summon right now + the future FSD update later when it comes out.

 

[R1Bill]

Mine's an early 2017. Don't know what AP is in it. How do I find out? I have summon and Nav on AP currently. Don't know about driver assist or auto lane change. Having some sensor issues so it currently doesn't work. Hopefully they'll fix it this time on my Tuesday SC appointment.

 

[diplomat33]

Mine's an early 2017. Don't know what AP is in it. How do I find out?

You should have EAP then. Do you have Nav on AP? If so, then you have EAP. And early 2017 should be an AP2 car.

 

[R1Bill]

That's what I thought. Then why does it show up in my account as a $6000 upgrade? (Just bought the car a few weeks ago so still full of questions) I'm able to select NAV on AP from the touchscreen.

 

[Az_Rael]

That's what I thought. Then why does it show up in my account as a $6000 upgrade? (Just bought the car a few weeks ago so still full of questions) I'm able to select NAV on AP from the touchscreen.

You are seeing the original FSD upgrade before they re-defined the packages. In 2017, there were 2 options, EAP for $5000 and FSD for $3000. EAP included TACC, summon, autopark, autosteer, auto lane change, and eventually NoA. FSD was just that "you will get in your Tesla and it will drive to your destination with no interaction from you"

Currently the 2017 FSD package doesnt have any released software functions. It will include the HDW3 upgrade and eventually city driving. For $6k I would take my chances on a future "sale" and watch Twitter. It was briefly offered for $2k in March.

 

[R1Bill]

You are seeing the original FSD upgrade before they re-defined the packages. In 2017, there were 2 options, EAP for $5000 and FSD for $3000. EAP included TACC, summon, autopark, autosteer, auto lane change, and eventually NoA. FSD was just that "you will get in your Tesla and it will drive to your destination with no interaction from you"

Currently the 2017 FSD package doesnt have any released software functions. It will include the HDW3 upgrade and eventually city driving. For $6k I would take my chances on a future "sale" and watch Twitter. It was briefly offered for $2k in March.

Thanks! What's the twitter channel I should monitor?

 

[wcorey]

At some point, vehicle to vehicle communications will be a big deal. That will be yet another HW update, but I don't think we have to put that on our 'radar screen' until 3-5 yrs from now, but it is coming.

I don't see that happening for three primary reasons. Cost, Varied manufacturers, a caboose is not in digital contact with the engine yet it moves within seconds of the engine, slows, accelerates in unison. What I do see is certain roads, perhaps interstates being restricted to Autonomous cars only. In the train example the caboose is mechanically attached to the engine. If all cars had radar they would be able to be virtually inches away from another car as they would react virtually instantaneously to an acceleration (deceleration is not actually a word in physics) change as the radar would be digitally connected to the speed control. In this scenario interstates could be akin to the AutoBahn with respect to no or extremely high speed limits. I, further, think we are a mere couple of years away from insurance companies and regulators mandating some level of autonomy. If every car was controlled via AI there would be no rush hour traffic, there'd be no accidents beyond the level of airline accidents. If so much as one person were driving their car, it would be mayhem.

 

[wcorey]

Currently the 2017 FSD package doesnt have any released software functions. It will include the HDW3 upgrade and eventually city driving. For $6k I would take my chances on a future "sale" and watch Twitter. It was briefly offered for $2k in March.

But it will this year. Also, Tesla has said the price now will only go up. When it gets regulatory approval it may well be $12,000. My 2012 Prius Plugin Advanced had Adaptive Cruise Control. It was ~ $4500 extra. Earlier cruise control at all was an optional $$$ extra. Now I suspect all cars have some version of CC, even Teslas have CC for no extra money.

 

[willow_hiller]

I don't see that happening for three primary reasons. Cost, Varied manufacturers, a caboose is not in digital contact with the engine yet it moves within seconds of the engine, slows, accelerates in unison. What I do see is certain roads, perhaps interstates being restricted to Autonomous cars only. In the train example the caboose is mechanically attached to the engine. If all cars had radar they would be able to be virtually inches away from another car as they would react virtually instantaneously to an acceleration (deceleration is not actually a word in physics) change as the radar would be digitally connected to the speed control. In this scenario interstates could be akin to the AutoBahn with respect to no or extremely high speed limits. I, further, think we are a mere couple of years away from insurance companies and regulators mandating some level of autonomy. If every car was controlled via AI there would be no rush hour traffic, there'd be no accidents beyond the level of airline accidents. If so much as one person were driving their car, it would be mayhem.

I think about this every time I'm stuck behind a long line of cars at a left turn light. Human reaction speeds are slow, it takes every person in the chain 5-10 seconds start moving, and by the time I even get to pull up the light is red again...

 

[Az_Rael]

Thanks! What's the twitter channel I should monitor?

Elon’s is the best bet.

But it will this year. Also, Tesla has said the price now will only go up. When it gets regulatory approval it may well be $12,000. My 2012 Prius Plugin Advanced had Adaptive Cruise Control. It was ~ $4500 extra. Earlier cruise control at all was an optional $$$ extra. Now I suspect all cars have some version of CC, even Teslas have CC for no extra money.

Yeah, Tesla told me that the FSD price would only go up in 2017. So I bought back then. Then the price dropped in early 2019. And we know that they offer AP/ FSD for a discount with the free trials on a regular basis. Once FSD is out, they will probably continue the same tactic.

 

[willow_hiller]

But it will this year. Also, Tesla has said the price now will only go up. When it gets regulatory approval it may well be $12,000. My 2012 Prius Plugin Advanced had Adaptive Cruise Control. It was ~ $4500 extra. Earlier cruise control at all was an optional $$$ extra. Now I suspect all cars have some version of CC, even Teslas have CC for no extra money.

Not doubting you, but could you find where it was said that the price will only increase from here on out? I'm really hoping for a Q2 sale, because otherwise the increasing cost of FSD might outpace my rate of savings.