Elon: FSD Beta tweets

[willow_hiller]

The driver is not overly cautious. The driver lets FSDb control the vehicle and disengages whenever he senses something off about the decision FSDb makes which is what you are supposed to do as a tester. It is not his job to prove that FSDb can handle situations, it's his job to make sure an accident does not occur, contrary to some other "testers" who think it's their job to prove that FSDb is "great" and can handle "difficult' situation by letting it make dangerous decisions.

Agreed. It's his car, he can choose to disengage his ADAS whenever he wants.

But you gave the opinion "Older versions were really bad, but the newer versions are not great." based on his video. I'm saying his choice to disengage does not necessarily reflect on FSD Beta's performance, and your choice to use a couple of disengagements from a single video as evidence to write off a firmware version as not great is premature.

 

[diplomat33]

So does


suggest that driving through the middle of a field is required? After all typical human can do that.

No. The levels only apply to driving on public roads.

See the text for yourself:

“Unconditional/not ODD-specific” means that the ADS can operate the vehicle on-road anywhere within its region of the world and under all road conditions in which a conventional vehicle can be reasonably operated by a typically skilled human driver. This means, for example, that there are no design-based weather, time-of-day, or geographical restrictions on where and when the ADS can operate the vehicle. However, there may be conditions not manageable by a driver in which the ADS would also be unable to complete a given trip (e.g., white-out snow storm, flooded roads, glare ice, etc.) until or unless the adverse conditions clear. At the onset of such unmanageable conditions the ADS would perform the DDT fallback to achieve a minimal risk condition (e.g., by pulling over to the side of the road and waiting for the conditions to change)." Page 32 of J3016.

 

[JulienW]

No. The levels only apply to driving on [public] roads.....

Just a tiny addition.

 

[Bitdepth]

So does


suggest that driving through the middle of a field is required? After all typical human can do that.

It does not

As specified herein, Level 5 is distinguished from Level 4 by the fact that it is not operationally limited to a specific operational
design domain and can rather operate on-road anywhere that a typically skilled human driver can reasonably operate a
conventional vehicle.

 

[Knightshade]

So does


suggest that driving through the middle of a field is required? After all typical human can do that.

No, but your question -yet again- suggests you've never actually read J3016

 

[Dewg]

I'll just toss this in here: Where in SAE does it mention the amount of accidents allowed for each level? With each level increase, does the number of accidents allowed decrease? So L2 allows for X number of accidents, and L3 requires X-n number of accidents?

 

[diplomat33]

I'll just toss this in here: Where in SAE does it mention the amount of accidents allowed for each level? With each level increase, does the number of accidents allowed decrease? So L2 allows for X number of accidents, and L3 requires X-n number of accidents?

Of course not. J3016 does not mention accident rates for each level.

J3016 specifically says that the levels are not about performance:

"By itself, this document imposes no requirements, nor confers or implies any judgment in terms of system performance. Therefore, while it may be appropriate to state, for example, that a given ADS feature does not meet the definition of Level 4 because it occasionally relies on a remote fallback-ready user to perform the fallback (and is therefore a Level 3 feature), it is not appropriate to conclude that the feature in question is therefore “non-compliant” or “unsafe.” Section 8.1, page 36.

And the levels are not a hierarchy from bad to good:

"While numbered sequentially 0 through 5, the levels of driving automation do not specify or imply hierarchy in terms of relative merit, technology sophistication, or order of deployment. Thus, this taxonomy does not specify or imply that, for example, Level 4 is “better” than Level 3 or Level 2." Section 8.3, page 36.

You can think of the levels like types of fruit. It tells you the fruit is an apple, banana, or an orange. The levels don't tell you if the apple is good or bad or if apples are better than oranges, they just tell you if it is an apple, a banana or an orange.

@ewoodrick

 

[Bitdepth]

I'll just toss this in here: Where in SAE does it mention the amount of accidents allowed for each level? With each level increase, does the number of accidents allowed decrease? So L2 allows for X number of accidents, and L3 requires X-n number of accidents?

Lol. It does not mention performance or safety levels or how good or bad each level is allowed to be. Essentially, it makes no judgment on how good each level is which is why it states that the levels are Never Fractional you can have a L2 system that performs better than L4 system in terms of not having any accident. SAE J3016 just aims to describe the role of the actor in performing the DDT and DDT fallback.

page 30

the level of driving automation is based on the functionality of the driving automation system feature,
as determined by an allocation of roles in DDT and DDT fallback performance between that feature and the (human) user
(if any). The manufacturer of a driving automation system feature determines that feature’s requirements, operational design
domain (ODD), and operating characteristics, including the level of driving automation, as defined below. The manufacturer
also defines the proper use of that feature.

 

[spacecoin]

Please watch this:

 

[Dewg]

Of course not. J3016 does not mention accident rates for each level.

J3016 specifically says that the levels are not about performance:

"By itself, this document imposes no requirements, nor confers or implies any judgment in terms of system performance. Therefore, while it may be appropriate to state, for example, that a given ADS feature does not meet the definition of Level 4 because it occasionally relies on a remote fallback-ready user to perform the fallback (and is therefore a Level 3 feature), it is not appropriate to conclude that the feature in question is therefore “non-compliant” or “unsafe.” Section 8.1, page 36.

And the levels are not a hierarchy from bad to good:

"While numbered sequentially 0 through 5, the levels of driving automation do not specify or imply hierarchy in terms of relative merit, technology sophistication, or order of deployment. Thus, this taxonomy does not specify or imply that, for example, Level 4 is “better” than Level 3 or Level 2." Section 8.3, page 36.

You can think of the levels like types of fruit. It tells you the fruit is an apple, banana, or an orange. The levels don't tell you if the apple is good or bad or if apples are better than oranges, they just tell you if it is an apple, a banana or an orange.

@ewoodrick

Exactly - couldn't have said it better myself, Diplomat!

So, everyone, keep this in mind when discussing safety, human accident comparisons, etc. Just because something is L3, L4, or L5, doesn't mean it also has to be flawless and never get into an accident.

 

[spacecoin]

Exactly - couldn't have said it better myself, Diplomat!

So, everyone, keep this in mind when discussing safety, human accident comparisons, etc. Just because something is L3, L4, or L5, doesn't mean it also has to be flawless and never get into an accident.

This is not a false statement when looking at J3016 in a vacuum, but in practice the ADS needs to be pretty great to be deployed since the OEM will be sued into oblivion if the system is proven to be at fault in the ODD when performing the DDT.

Legislators generally want to see if the ADS is safe before granting permits. In UNECE R157 there are formal tests. Even in the new DCAS (L2) drafts. The DMV:s has a process to before granting permits if different kinds. Et.c.

 

[Dewg]

This is not a false statement when looking at J3016 in a vacuum, but in practice the ADS needs to be pretty great to be deployed since the OEM will be sued into oblivion if the system is proven to be at fault in the ODD when performing the DDT.

Here is where the free market will sort itself out. I would also imagine that laws will slowly change with tort reform, such as max payouts, etc. We may see some companies fail, and get merged/swallowed up into other mfgs while this is all sorted out.

The point is that it will never be accident free. People will die using ADAS/AV. It's inevitable, and something everyone needs to accept. The trick will be to kill less people than humans alone. We, humans, killed 42,000 people in 12 months driving ourselves (in the US). If L2+, L3, and L4 can save some of those lives, it's worth it.

 

[spacecoin]

Here is where the free market will sort itself out. I would also imagine that laws will slowly change with tort reform, such as max payouts, etc. We may see some companies fail, and get merged/swallowed up into other mfgs while this is all sorted out.

The point is that it will never be accident free. People will die using ADAS/AV. It's inevitable, and something everyone needs to accept. The trick will be to kill less people than humans alone. We, humans, killed 42,000 people in 12 months driving ourselves (in the US). If L2+, L3, and L4 can save some of those lives, it's worth it.

Yes. I apologize for the late edit, but as I was saying - most jurisdictions will regulate ADS, if not everyone. In L2 the driver’s always driving, simple as that. That’s not an ADS.

 

[Bitdepth]

Exactly - couldn't have said it better myself, Diplomat!

So, everyone, keep this in mind when discussing safety, human accident comparisons, etc. Just because something is L3, L4, or L5, doesn't mean it also has to be flawless and never get into an accident.

The document does not say that but in practice the higher the level the higher the MTBF has to be because at lower levels (L3 and below) the DDT fallback is the human while from L4 and up the DDT fallback is the ADS.

A manufacturer can say it's just a L2 system so it's not our responsibility that it failed, or it's a L3 system we gave the human 10 seconds to take over but they didn't but at L4 you can't blame anyone but the manufacture so they have to be sure that their ADS is really good otherwise they could kill someone. Would you be willing to ride in the back seat of FSDb as it currently is?

 

[Dewg]

The document does not say that but in practice the higher the level the higher the MTBF has to be because at lower levels (L3 and below) the DDT fallback is the human while from L4 and up the DDT fallback is the ADS.

A manufacturer can say it's just a L2 system so it's not our responsibility that it failed, or it's a L3 system we gave the human 10 seconds to take over but they didn't but at L4 you can't blame anyone but the manufacture so they have to be sure that their ADS is really good otherwise they could kill someone. Would you be willing to ride in the back seat of FSDb as it currently is?

Interesting question, but I don't accept your premise. FSD Beta is L2+ driver assist system, and is not designed to be autonomous (L4+). So I would never be in the back seat with FSD Beta, as there must be a driver. It's designed to handle boring driving tasks while you monitor it and take over when needed.

 

[Knightshade]

Legislators generally want to see if the ADS is safe before granting permits. In UNECE R157 there are formal tests. Even in the new DCAS (L2) drafts. The DMV:s has a process to before granting permits if different kinds. Et.c.

Meanwhile in many US states you need do no more than tell the DMV "This is level 3 and we promise it follows all required laws" and you can throw your stuff on the road the next day... (and same for 4 and 5).

Now, nobody is going to do that with a system they don't have high confidence in-- but no evidence of safety is required up front in these places.

 

[Bitdepth]

Interesting question, but I don't accept your premise. FSD Beta is L2+ driver assist system, and is not designed to be autonomous (L4+). So I would never be in the back seat with FSD Beta, as there must be a driver. It's designed to handle boring driving tasks while you monitor it and take over when needed.

Fair enough, but it was just a thought experiment, let's assume FSDb was released as a L4 or L5 system in its current state.

 

[spacecoin]

Interesting question, but I don't accept your premise. FSD Beta is L2+ driver assist system, and is not designed to be autonomous (L4+). So I would never be in the back seat with FSD Beta, as there must be a driver. It's designed to handle boring driving tasks while you monitor it and take over when needed.

There is no “Level2+” since we’re discussing J3016…

 

[spacecoin]

Meanwhile in many US states you need do no more than tell the DMV "This is level 3 and we promise it follows all required laws" and you can throw your stuff on the road the next day... (and same for 4 and 5).

Now, nobody is going to do that with a system they don't have high confidence in-- but no evidence of safety is required up front in these places.

Sure, legislators will have different approaches. but they likely won’t let the manufacturer keep it on the road if it kills people too often. Mercedes did get their L3 system approved as per unece r157 in germany, but that’s only limited 3rd party testing. with dcas (the new unece L2) it’s all self-certification.

 

[spacecoin]

Interesting question, but I don't accept your premise. FSD Beta is L2+ driver assist system, and is not designed to be autonomous (L4+).

Since this is the “Elon says stupid things“-thread, perhaps check out the Q2 earnings transcript?

“It's just that you can think of every car that we -- that we sell or produce that -- that -- that has a full Autonomy capability as actually something that, in the future, may be worth as much as five times what it is today.”

Not FSD-capable like he usually says. And then they define FSD as L2. Now he saId full autonomy. Tesla’s lawyers aren’t likely happy with that.

I guess that would be ActualFSDb then if this isn’t it.