FSD rewrite will go out on Oct 20 to limited beta

[turnem]

Just like the car reads speed limit signs so they could prevent you from speeding if they didn't want you to.

So by your...let's generously call it logic... Tesla wants everyone to speed.

Otherwise they'd disable the ability to do so...right?

We could play the extreme example game all day. As I've stated before, not everything in my world is binary but apparently in your world it is.

But honestly - none of this matters because my point is actually dead simple - so simple in fact that you can probably follow it and agree. FSD Beta works better and is safer on city streets than AP regardless of whether AP is intended to be used on city streets or not.

 

[drtimhill]

I think it's possible Tesla may not do a dribble and release a bunch of features at once. However, I absolutely disagree that 'Tesla FSD' is "all or nothing" in nature.

Self driving in general must be all or nothing absolutely, yes.... but Tesla FSD is not self driving.

And therefore a dribble of features is possible. But I agree they might release City Nav and a bunch of turns and other things all at once.

Well I did say "much closer", by which I meant many of the features are so inter-dependent that you either get them all working or none. Example: you cannot make turns safely without a good birds-eye view, which is probably a good 70-80% of the work that has been done (lane identification, road boundaries, hazard and obstruction mapping, car identification and velocity vectors etc). Add in traffic signal recognition etc etc, and you have pretty much got everything that is in the current FSD beta. That's why city driving is so hard, you have to solve many problems all at once, unlike freeways that are so much more restrictive.

 

[pilotSteve]

Well I did say "much closer", by which I meant many of the features are so inter-dependent that you either get them all working or none. Example: you cannot make turns safely without a good BEV view, which is probably a good 70-80% of the work that has been done (lane identification, road boundaries, hazard and obstruction mapping, car identification and velocity vectors etc). Add in traffic signal recognition etc etc, and you have pretty much got everything that is in the current beta. That's why city driving is so hard, you have to solve many problems all at once, unlike freeways that are so much more restrictive.

BEV view. .... Battery Electric Vehicle view? Too many TLAs here. (three letter acronyms )

 

[drtimhill]

Yea, seriously.

This is a scenario that they could absolutely not allow you to engage autopilot if they didn't want you to. They don't currently allow you you engage autopilot in certain situations so we know they could disable it if they wanted to. They don't want to.

And stoplights and stop signs?? Sure, there are a few scenarios where these show up on divided highways but come on man. You can't possibly believe this is the ONLY time they expect you to use this feature?

This whole issue has been debated several times before in this forum, and comes down to a confusion between what is technically possible versus what is necessary. You cannot extrapolate from "they could prevent this" to "therefore they want you to do this" .. it is an invalid inference. As has been noted here, they could, possibly, stop you from driving off a cliff. Just because they don't stop you doesn't mean they are encouraging you to drive off a cliff. But you are using precisely this argument to justify your position.

 

[drtimhill]

BEV view. .... Battery Electric Vehicle view? Too many TLAs here. (three letter acronyms )

BEV = Birds Eye View (and yes, sorry)

 

[turnem]

This whole issue has been debated several times before in this forum, and comes down to a confusion between what is technically possible versus what is necessary. You cannot extrapolate from "they could prevent this" to "therefore they want you to do this" .. it is an invalid inference. As has been noted here, they could, possibly, stop you from driving off a cliff. Just because they don't stop you doesn't mean they are encouraging you to drive off a cliff. But you are using precisely this argument to justify your position.

Sheesh. I'll restate it in real-world terms then.

I currently use AP on city streets despite the fact that he manual says it's not intended for that use. I believe that FSD Beta is significantly better and safer to use on city streets (even WITH it's current challenges and limitations) than the AP that I'm currently using (which is expressly what the manual tells me NOT to do... I know, I know... I'm a complete moron).

So I'd personally like to see them roll-out FSD to a wider audience so that we can begin taking advantage of the numerous benefits.

And as I've stated before I believe that are two (basic) types of people in this forum.

The ones like me that use AP daily on city streets (in pretty much any scenario that the car will allow them to) but understand they are in FULL control of the car and completely responsible for any adverse consequences. THESE folks want FSD as soon as they can get it.

The other group NEVER engages AP except for as instructed per the manual. They feel that releasing FSD to the wild is going to be an unmitigated disaster.

Nothing wrong with being in either group. I'm just making the point that things here in the real world aren't always as cut and dry as a few here make them out to be.

 

[Microterf]

I'm with you @turnem. I want FSD as soon as possible. I engage AP every chance I get

 

[drtimhill]

The ones like me that use AP daily on city streets (in pretty much any scenario that the car will allow them to) but understand they are in FULL control of the car and completely responsible for any adverse consequences. THESE folks want FSD as soon as they can get it.

The other group NEVER engages AP except for as instructed per the manual. They feel that releasing FSD to the wild is going to be an unmitigated disaster.

Agreed, and ultimately it will be for Tesla to make that call. And they will be balancing issues that are tricky for anyone outside the company to quantity, including (a) revenue realization from FSD purchases, (b) cachet and effect on stock price of such a major release, (c) potential backlash if FSD proves unreliable and/or dangerous, (d) regulatory and/or legal impact should FSD cause major issues.

fwiw, my feeling is that FSD will actually be safer than the average human driver pretty quickly (as I've noted before, that bar is pretty damn low anyway). However, public perception is sadly not made by rational statistics, but by attention-grabbing headlines ("Dog slightly hurt by self-driving car!!! 'They should be banned forever!!' says angry owner" etc). This was really my concern with the abuse of AP; will it lead to such adverse publicity that it impacts the legitimate uses of the technology?

 

[EVNow]

I think they very much intend to have it in there honestly. It's legal ease to protect them. I think they will have it in there until FSD is rolled out and even then they will have SOME form of verbiage in there to protect them.

The basic protection is in the verbiage that tells you the software is beta and you have to pay attention and take corrective action.

This bit of text will not "protect" them per se - there is plenty of contradictory statements and feature lists (not to talk of tweets and interviews) that will overwhelmingly override that one line of text.

 

[JulienW]

I want FSD also and I think it is making GREAT progress BUT it will only take one gigantic catastrophic accident to end it all. I know that 100 people are killed a day in the US by human accidents BUT NO one CARES (other that those effected).

Here is a situation from today when it starts to make an unprotected blind left turn. Imagine this happing to a kid and 3 friends of an FSD owner while they were out joy ridding and not paying attention. It would be "front page" news all over the US and the regulators would be fighting over who could regulate the most. We have just got to be patient as hard as it is.

 

[drtimhill]

I want FSD also and I think it is making GREAT progress BUT it will only take one gigantic catastrophic accident to end it all. I know that 100 people are killed a day in the US by human accidents BUT NO one CARES (other that those effected).

Here is a situation from today when it starts to make an unprotected blind left turn. Imagine this happing to a kid and 3 friends of an FSD owner while they were out joy ridding and not paying attention. It would be "front page" news all over the US and the regulator would be fighting over who could regulate the most. We have just got to be patient as hard as it is.

Exactly. I know people are keen to get it (so am i), but its better to wait than to have a catastrophe. I think Tesla will wait until they can show clear statistics that it is as safe (or safer) as a human driver in given circumstances. Then they can fight the inevitable hysteria when someone does abuse FSD and cause some bad headlines.

 

[Electroman]

[cross posted from another thread. I think this post belongs here]

After watching over two dozen videos, the results are in (in my mind):

- On city streets in slow moving conditions (less than 45 mph) it handles most situations like a champ. The perception and situational awareness is really good for slow speeds. Stop signs, red lights, right turns, protected left turns, round abouts, lane changes, parking lot maneuvers, passing parked cars, congested narrow streets - I would give it a B+. Unprotected left turns I would give a C- at low speed conditions (barely passing grade).

- Anything over 45 mph, things get a little iffy. Right turns get a passing grade only 50% of the time. Unprotected left turns simply doesn't work with fast cross traffic. Lane changes are not confident. I would give it a D- failing grade.

Now it is not clear to me whether the hardware can even see fast moving cross traffic in either directions at a fair distance, can judge the speed and has the processing power to take the right decision. Nothing that I have seen so far gives me the confidence. We will see.

 

[Electroman]

I use AP on city streets every single time I get into the car.

Yup. I use AP on city streets and highways, everywhere I get the steering icon. I am watchful and it does an amazing job.

 

[Daniel in SD]

I want FSD also and I think it is making GREAT progress BUT it will only take one gigantic catastrophic accident to end it all. I know that 100 people are killed a day in the US by human accidents BUT NO one CARES (other that those effected).

Here is a situation from today when it starts to make an unprotected blind left turn. Imagine this happing to a kid and 3 friends of an FSD owner while they were out joy ridding and not paying attention. It would be "front page" news all over the US and the regulators would be fighting over who could regulate the most. We have just got to be patient as hard as it is.

View attachment 617749

Yes the fact that we have multiple regulatory agencies at all levels of government and numerous private advocacy groups shows that no one cares about road safety.
Deaths per mile have dropped almost continuously since the automobile was invented because people do care.

I still can't wrap my head around why anyone thinks automating unprotected left turns with driver supervision is a good idea. Obviously once it can do it better than a human it's a great idea and maybe there is some way to implement a feature to prevent humans from making dangerous turns (in the same vein as the emergency lane departure assist).

 

[Snuffysasa]

I still can't wrap my head around why anyone thinks automating unprotected left turns with driver supervision is a good idea. Obviously once it can do it better than a human it's a great idea and maybe there is some way to implement a feature to prevent humans from making dangerous turns (in the same vein as the emergency lane departure assist).

I don't have data to support this. However, I am extremely confident that unprotected left turns with confirmation required would result in less accidents than just regular humans taking left turns.

 

[arghx7]

I dunno... laws?

In fact of the states that specifically already allow legit auto driving cars, AFAIK they all require them to obey all driving laws- including the speed limit.

Sure, and get passed by the cops doing 10 over. The cops need to be obey traffic laws too.

 

[EVNow]

- On city streets in slow moving conditions (less than 45 mph) it handles most situations like a champ. The perception and situational awareness is really good for slow speeds. Stop signs, red lights, right turns, protected left turns, round abouts, lane changes, parking lot maneuvers, passing parked cars, congested narrow streets - I would give it a B+. Unprotected left turns I would give a C- at low speed conditions (barely passing grade).

- Anything over 45 mph, things get a little iffy. Right turns get a passing grade only 50% of the time. Unprotected left turns simply doesn't work with fast cross traffic. Lane changes are not confident. I would give it a D- failing grade.

This tells us what they can do.

Infact, Tesla must be having *real data* - rather than anecdotes we can watch on youtube. They know exactly what scenarios FDS does well (better than avg human !) and what condition it doesn't. They should disable / ask for confirmation in those cases where it is not reliable enough.

I still can't wrap my head around why anyone thinks automating unprotected left turns with driver supervision is a good idea.

I still can't wrap my head around why anyone thinks automating very fast freeway lane changes and exits/entrances is a good idea. I don't use freeway NOA.

Unprotected left turns - with driver supervision but no confirmation - is simply not good enough now. They won't enable it until it is.

 

[Snuffysasa]

I still can't wrap my head around why anyone thinks automating very fast freeway lane changes and exits/entrances is a good idea. I don't use freeway NOA.

So you don't think it's possible to automate lane changes safer than humans performing lane changes?

Or you don't think the performance of Tesla NoA's lane changes is safe enough?

 

[Daniel in SD]

I don't have data to support this. However, I am extremely confident that unprotected left turns with confirmation required would result in less accidents than just regular humans taking left turns.

I also don't have the data. But I'm skeptical of that.

I still can't wrap my head around why anyone thinks automating very fast freeway lane changes and exits/entrances is a good idea. I don't use freeway NOA.

I also don't use NoA because it doesn't do interchanges as well as a human. It also doesn't select lanes as well as a human so I don't use that either. I do use use auto lane change when I'm confident that it won't abort. I spend such a small portion of my driving on interchanges and changing lanes anyway. I wish Tesla would do a non-beta highway FSD system so I could not pay attention to the road.

Unprotected left turns - with driver supervision but no confirmation - is simply not good enough now. They won't enable it until it is.

The problem is that unless it's really good it still might not be safe. People will trust it and become complacent once it gets much better but still below human capability. I always cringe when people say they trust features that require driver supervision, you're not supposed to trust it! A bad unprotected left turn is not something want to react to after the car has already moved ten feet.

 

[Snuffysasa]

I wish Tesla would do a non-beta highway FSD system so I could not pay attention to the road.

That would be amazing and huge value, but also Tesla knows they can't achieve that. And are not focused on that.