Post Elon Update Poll on FSD

[nvx1977]

I'm optimistic that Tesla can get FSD to a really good state. But really hard to imagine it happens at the end of this year. Before beta testing, I would have guessed 2 years, based on the AI Day presentation. After using FSD beta and seeing the rate of improvement (which I consider slow), it feels more like 2-5 years to me.

While that might seem like a long time, it doesn't feel so long ago that Tesla posted the infamous 2016 self driving video. Can't believe my Model 3 is almost 4 years old. Barring some disastrous accident, I feel pretty confident my car will see really good FSD before I sell it. Not going to call it an SAE level because it's never been clear to me how those levels determine actual proficiency with self-driving (in terms of miles per crash rate or something).

 

[EVNow]

People insist on using levels because the Society of Automotive Engineers has official definitions for levels 0 through 5 and has had for years, and these are the levels the National Highway Traffic Safety Administration uses. The official definition for Level 4 is:

"High Automation: The vehicle is capable of performing all driving functions under certain conditions. The driver may have the option to control the vehicle."

You could argue this isn't a terrible detailed definition-- "certain conditions" sure has a lot of leeway-- but it's there.

It is a useless definition. As George Hotz would say something made up by business people, not engineers. Not sure whether any of the engineers at Society of Automotive Engineers were involved

Having worked on various engineering standards, I can say it is one of the worst I've seen.

ODD is not the only issue, it is also about error rates (what Mobileye calls MTBS - I'd call it MDBF, Mean Distance Before Failure). If a vehicle generally performs "all driving functions under certain conditions" (i.e. for a given ODD) - but fails in that ODD at a particular instance - is it at L4 or not ? How many times can it fail before we say its not "performing all driving functions" ?

ps : ANyway, my last post here about levels - there is a separate thread on it we can continue on. But my main point stands - the question had something about Levels, but Elon's answer had nothing about levels, only about error rates - which as we have seen here levels definitely do not talk about.

 

[Goose66]

Elon stated (on the record) that before end of this year, FSD will be "much safer than a human driving" and that he "would be shocked" if FSD does not reach Level 4 autonomous driving is not reached by Dec 31 2022. What say you?

I think we will find this prediction to be about as accurate as all his other predictions for FSD since 2015.

As I have said before, I see two likely outcomes to the whole FSD Beta thing: 1) Tesla gets it to some state, probably somewhere between L2 and L3 with the current hardware and sensors, and they pop champagne and call it done, or 2) or they continue to release new beta version for some period of time until they come out and say that regulators have shut them down and they have to indefinitely postpone rollout. In the case of #2, I predict Elon will say they were so close and that the next version would have "blown our minds!"

 

[Daniel in SD]

@EVNow just needs to add a browser extension that converts every mention of L4/L5 to robotaxi and everyone will be happy.
That's clearly how Elon interprets it.

 

[2101Guy]

Damn. Jalopnik used to LOVE elon. What happened?

Elon Musk Promises Full Self-Driving 'Next Year' For The Ninth Year In A Row

When he gets to 10, all Teslas will magically come to life and make sweet, sweet love to everyone who believed in them all along

jalopnik.com

 

[Sporty]

L4 implies a limited ODD and the manufacturer can make the ODD whatever they want and still be L4. So depending on the ODD, sure, Tesla could achieve L4 this year. For example, they could pick a 10 sq mi geofenced area in LA where FSD is able to drive with no human intervention and that would technically be L4. Or Tesla could try FSD that works basically everywhere, except maybe certain intersections that are too tricky. Or Tesla could try do FSD everywhere but limited to good weather. There are lots of ways to do the ODD for L4. It really depends on the ODD. But Elon has really talked about FSD being safer than humans in general. He has not talked about any ODD limits. So I think Elon wants to skip L4 and try for L5. But I think it is likely that Tesla will release FSD later this year that works everywhere but still requires active driver supervision. So it will not be L4 or L5.

How about a 4.20 km sq area?

 

[Sporty]

I disagree. I think the current price is based on it being a form of advanced AP on city streets, or a 'personal chauffeur' as you mentioned. Once robotaxi level of autonomy is reached the price will be significantly more than $12k. I think once true L5 is reached the price of FSD will be more like $50k with incremental prices below that for limited features.

If it gets solved, at some point others will follow and competition will kick in. It might be a premium price to start but it won’t stay there.

I voted no. They’re not going to hit L4 with or without ODD tricks for a long time. Not this year.

 

[S4WRXTTCS]

L4 implies a limited ODD and the manufacturer can make the ODD whatever they want and still be L4. So depending on the ODD, sure, Tesla could achieve L4 this year. For example, they could pick a 10 sq mi geofenced area in LA where FSD is able to drive with no human intervention and that would technically be L4. Or Tesla could try FSD that works basically everywhere, except maybe certain intersections that are too tricky. Or Tesla could try do FSD everywhere but limited to good weather. There are lots of ways to do the ODD for L4. It really depends on the ODD. But Elon has really talked about FSD being safer than humans in general. He has not talked about any ODD limits. So I think Elon wants to skip L4 and try for L5. But I think it is likely that Tesla will release FSD later this year that works everywhere but still requires active driver supervision. So it will not be L4 or L5.

About the only thing I see being potentially plausible is the Vegas Loop going from L0 to L4.

That's the one I'm curious about because if they were going to do L4 anywhere that would be the place to do it.

 

[Daniel in SD]

About the only thing I see being potentially plausible is the Vegas Loop going from L0 to L4.

That's the one I'm curious about because if they were going to do L4 anywhere that would be the place to do it.

That would not be L4. (@EVNow I think we found our answer for how a robotaxi might not be L4/L5. I could see Elon calling a vehicle operating in a Boring company tunnel a robotaxi.)

This SAE Recommended Practice describes motor vehicle driving automation systems that perform part or all of the dynamic driving task (DDT) on a sustained basis. It provides a taxonomy with detailed definitions for six levels of driving automation, ranging from no driving automation (level 0) to full driving automation (level 5), in the context of motor vehicles (hereafter also referred to as “vehicle” or “vehicles”) and their operation on roadways. These level definitions, along with additional supporting terms and definitions provided herein, can be used to describe the full range of driving automation features equipped on motor vehicles in a functionally consistent and coherent manner. “On-road” refers to publicly accessible roadways (including parking areas and private campuses that permit public access) that collectively serve users of vehicles of all classes and driving automation levels (including no driving automation), as well as motorcyclists, pedal cyclists, and pedestrians.

 

[Twiglett]

there is probably a slim chance that they will deliver a mostly working FSD by the end of the year - maybe 10% chance.
But that will still be driver assist and definitely not worth 12k
I don't believe the robotaxi (with fully autonomous driving that is supposed to make FSD worth lots of money) is going to appear in this decade - which just means it definitely isn't worth 12k today - and even more definitely will not be enabled in a car that I own today.

That is on the basis of current 10.8.1 FSD and comparing progress of regular AP over the past four years.

 

[Phil Stasik]

Having put many miles on FSD Beta over the past few months, I just can not imagine L4/5 autonomy at year’s end. It IS currently amazing, but needs some sort of camera upgrade plus processor/software revolution. My wife has given up on it, because it keeps doing stupid things to her. I kind of enjoy driving on the edge of my seat — totally focused on the machine and the driving environment — disconnecting as needed, and pushing the little camera button, in the hopes that Santa’s elves see my messages, and are working away in their workshop on the next update.
I have flown airplanes that are like this. You enjoy the good characteristics most of the time but keep away from the bite-your-ass corners of the envelope. “You snooze, you lose!” The difference here is that they are able to continually update the software to shrink those corners.
We are safe to vote “NO”.

Maybe that’s where I am — having Santa and his (two?) reindeer displayed on the UI — as I test pilot the sleigh along the snowy virtual roadway. But I’m actually in Florida, and it’s really Fantasyland!

 

[Baybay408]

Safer than humans I can believe but Level 4 means Tesla will allow no human supervision in some situations. This means that they will be responsible for accidents. I don't see any benefit for them taking on this risk. So no. No level 4 in the near future. They will still require the driver to be responsible for accidents

There is something everyone is completely missing. FSD may make fewer at fault accidents than humans but it won't make fewer total accidents. Everyone is counting how many accidents are caused by human error, but nobody is counting how many accidents have been avoided by humans. I have seen many accidents avoided by smart anticipation by humans. Anticipating that someone will run a red light or a stop sign. Or change lane when you ard in their blind spot.
I saw a video on youtube where FSD tries to cross an intersection in front of someone running a red light very fast.

 

[Bouba]

It’s true, FSD could improve....my suggestion would be to have a car chase mode...L2 could start with a OJ chase and L4 would be the French Connection

 

[Waterskidude]

Here’s an honest question that I haven’t found an answer for anywhere. My car regularly kicks me out of autopilot because 1 or more cameras are blinded. How will FSD get past this problem?

 

[JulienW]

Here’s an honest question that I haven’t found an answer for anywhere. My car regularly kicks me out of autopilot because 1 or more cameras are blinded. How will FSD get past this problem?

AKA: Level 3.

 

[byeLT4]

Here’s an honest question that I haven’t found an answer for anywhere. My car regularly kicks me out of autopilot because 1 or more cameras are blinded. How will FSD get past this problem?

Good question. HW3 MCU1 here and there are times I can't even engage cruise control. I voted no LOL

 

[aronth5]

Here’s an honest question that I haven’t found an answer for anywhere. My car regularly kicks me out of autopilot because 1 or more cameras are blinded. How will FSD get past this problem?

Same question applies to rain since I find that happens more often than blinded cameras. Once it starts raining while using FSD on the highway I expect FSD will eventually disconnect and it typically does.

 

[diplomat33]

Here’s an honest question that I haven’t found an answer for anywhere. My car regularly kicks me out of autopilot because 1 or more cameras are blinded. How will FSD get past this problem?

FSD would need to either pull over/come to an emergency stop or ask the driver to take over.

 

[nvx1977]

Here’s an honest question that I haven’t found an answer for anywhere. My car regularly kicks me out of autopilot because 1 or more cameras are blinded. How will FSD get past this problem?

The only camera that ever gets dirty on my Model 3 is the rear camera. Front, repeaters, and b-pillars stay surprisingly clean, and I've driven through some nasty sticky snow and freezing rain. Only time I've had FSD complain about the cameras is at night on a poorly lit road. the side cameras often throw warnings, but FSD doesn't disengage.

I've had NoA revert to AP during heavy rain, but I don't get a camera occlusion warning. Car uses the camera feed to detect that it's raining hard, and NoA disengages out of caution. It frequently comes back on and off during hard rains. Haven't seen this when in FSD beta local roads.

 

[dramsey]

Damn. Jalopnik used to LOVE elon. What happened?

Elon Musk Promises Full Self-Driving 'Next Year' For The Ninth Year In A Row

When he gets to 10, all Teslas will magically come to life and make sweet, sweet love to everyone who believed in them all along

jalopnik.com

Uh…they started paying attention to his years-long records of over promising and under-delivering?

As I’ve said, it could be worse. Perhaps they can interview some of the people who gave Elon $250K for the Founder’s Edition Roadsters in 2017 and see how they’re holding up.