Software Update 2018.39 4a3910f (plus other v9.0 early access builds)

[Snuffysasa]

I totally get where you're coming from, but I'll just point out that if the "feature" they were adding was the ability to look ahead approximately 10 seconds in time to know when it's about to encounter something it can't handle, then that would be a feature bringing it closer to L3.

I guess I wouldn't call that a feature. Certainly not a user facing feature.

And being able to look ahead and interpret everything well enough to always know if it can handle what's coming next, is not a feature.

 

[rnortman]

But, in my opinion, very significant improvement. The difference between when the driver reads the environment and makes a decision vs. when the car makes that decision is extremely more advanced. So, I emphatically agree with your assessment.

Yeah, it's almost like we now have approximately 1% of the extremely advanced technology demonstrated in the 2016 FSD demo video! (You know, the extremely advanced technology that just required validation and regulatory approval...) Except they don't have enough confidence in that 1% to actually let it do anything.

A significant improvement indeed, but that only speaks to how far they have left to go... and they probably could have been here 18 months ago if they'd just included rear corner radars for blind spot monitoring...

 

[tljdesign]

Another note, tesla added a new setting to car config it pushes from mothership to your car, so they can easily control which features are enabled ulc_without_confirm: false which controls one of drive-by-nav features

Remembered this post from @dennis_d from earlier this weekend and finally found it... which means it may not be that the ULC is removed from the release, but that it's present and just not enabled, AND it may not require a complete update to enable for cars? Or maybe I'm reading this incorrectly?

 

[rush6410]

Yeah, it's almost like we now have approximately 1% of the extremely advanced technology demonstrated in the 2016 FSD demo video! (You know, the extremely advanced technology that just required validation and regulatory approval...) Except they don't have enough confidence in that 1% to actually let it do anything.

A significant improvement indeed, but that only speaks to how far they have left to go... and they probably could have been here 18 months ago if they'd just included rear corner radars for blind spot monitoring...

I agree with your point. But, I have been in the technology business long enough to understand the foundational advancement even though the end result has not been achieved yet. So, IMO, they are pretty far along to reaching the stated vision.

 

[diplomat33]

There is absolutely no correlation or connection between SAE level 3 and making a lane change / deciding when to make a lane change.

I beg to differ. L3 is defined as a system that monitors the driving environment in addition to executing steering and acceleration/deceleration. it is defined by the SAE as "the driving mode-specific performance by an automated driving system of all aspects of the dynamic driving task with the expectation that the human driver will respond appropriately to a request to intervene."

Lane changing is certainly a dynamic driving task, wouldn't you say and deciding when to make a lane change certainly requires monitoring the environment (seeing other cars, predicting the path of those other cars, determining if those other cars are a collision risk, detecting the lanes and knowing where you are in relation to those lanes etc). So lane changing is definitely a piece of L3.

Based on the definition of L3, V9 certainly qualifies on the highway.

Noooooo! Stop. L3 is NOT about capability!

Let's look at the definition again: "the driving mode-specific performance by an automated driving system of all aspects of the dynamic driving task with the expectation that the human driver will respond appropriately to a request to intervene."

If "all aspects of the dynamic driving task" is not about capability, what is it about?

Adding new features in v9 has nothing to do with bringing it closer to L3.

Depends on what the features are. If the features represent a higher level of autonomy, then yeah, they will most certainly bring the system closer to L3.

There can be a system that is only capable of single lane keeping and is L3. There can be a system that is not even capable of lane keeping at all or doesn't even drive in lanes that could be an L3 system.

You seem to be fixated on lane changing. I am not saying that V9 is L3 because it can do lane changes. Heck, AP before V9 could do lane changes and it was not L3. Lane changing alone does not make a system L3. But lane changing when the car is making the decision to change lanes, is most certainly a piece of L3.

 

[conman]

Here we go again, fighting over the definitions of levels of automation... Sigh.

I guess we've run out of things to talk about while we keep waiting to get V9 ourselves.

 

[diplomat33]

Here we go again, fighting over the definitions of levels of automation... Sigh.

I guess when discussing features that are part of a driver assist system, it is inevitable that the discussion will come back to where the features fall on the self-driving scale, hence a debate over the SAE levels of autonomy. I did not want to rehash this old debate but I just could not stand by and say nothing when someone says that V9 is still nowhere near L3.

 

[Snuffysasa]

before we get started, just want to make sure you have ready every page of this document.

J3016A: Taxonomy and Definitions for Terms Related to Driving Automation Systems for On-Road Motor Vehicles - SAE International

Otherwise, I have no interest in this discussion. If you have, I am happy to hear your argument.

Lane changing is certainly a dynamic driving task

What ?? sure it could be....

but not if the dynamic driving task is staying in a single lane. or following a route that doesn't have lanes at all.

deciding when to make a lane change certainly requires monitoring the environment

lol... focus man.

deciding when to make a lane change requires monitoring yes....

however, monitoring doesn't require deciding when to make a lane change.

ased on the definition of L3, V9 certainly qualifies on the highway.

LMFAO, no it does not...

first of all, the only thing that makes it L3 is if Tesla says so... levels of autonomy are assigned by OEM only, and not measured by 3rd parties or consumers.

..... in progress

If the features represent a higher level of autonomy, then yeah, they will most certainly bring the system closer to L3.

give me an example

But lane changing when the car is making the decision to change lanes, is most certainly a piece of L3.

no.... it's not..

Where in SAE definitions does it say anything about lane changing and L3...

Also like I said.... L3 systems can exist that do not lane change, or do not even operate in areas with lanes.

lane changing is not a piece of L3


PS- if you prefer to PM me, thats fine.

 

[Snuffysasa]

Here we go again, fighting over the definitions of levels of automation... Sigh.

I guess we've run out of things to talk about while we keep waiting to get V9 ourselves.

Sorry

 

[dark cloud]

And meanwhile, @bmah is determining which level some posts are:

level 3: conditionally passes as on topic: poster responds appropriately
level 4: highly off topic: if poster remains off topic moderator will intervene
level 5: fully off topic: posts have been automatically removed from thread

 

[MarcusMaximus]

Hm. Does "technically capable" include being at least as safe as a human driver even when the person in the driver's seat is completely not paying attention, like maybe is reading a book or watching a movie or composing a sonnet? Does "technically capable" include being able to react, at least as well as a human, to a deer running across the road or debris falling off a vehicle in front of it, without relying on the driver to take over immediately? Does "technically capable" including being able to handle a tire blowout, again without requiring the human driver to take over? Does "technically capable" including being able to know at detect that it's going to encounter a situation it can't handle at least 10s before it actually can't handle the situation so it can alert the driver to take over?

There’s potentially different answers to some of these questions, but I’ll answer with a general: No. But those make up a rough description of level 4, not level 3.

One description of Level 3:
“An Automated Driving System (ADS) on the vehicle can itself perform all aspects of the driving task under some circumstances. In those circumstances, the human driver must be ready to take back control at any time when the ADS requests the human driver to do so. In all other circumstances, the human driver performs the driving task”

Another:
“Conditional
Automation
the driving mode-specific performance by an automated driving system of all aspects of the dynamic driving task with the expectation that the human driver will respond appropriately to a request to intervene”

Neither of those actually even require the removal of the nag, just removal of the confirmation pull would be enough, but I throw that one in to make it feel appropriately automated. They certainly don’t require handling every odd circumstance the car could ever find itself in without human intervention.

 

[pedriscoll]

This is the most lively discussion I have seen on this forum since the period just before the rollout of AP1 in October of 2015. I have now had my AP2 100D for 1 week and can confidently say that AP2 has equaled if not surpassed AP1 in overall functionality. v9 should finally push AP2 past that point. I am excited to try the new features but know from experience to be patient. The upgrade will be pushed out only when Tesla is ready.

 

[Snuffysasa]

They certainly don’t require handling every odd circumstance the car could ever find itself in without human intervention.

Yes a Level 3 system must be able to handle every odd cirumstance or to detect the circumstance within the agreed upon time interval with the consumer.

"expectation that the human driver will respond appropriately to a request to intervene"

the ROI is issued by the system, meaning it needs to detect every odd circumstance, with sufficient time. That is pre determined and advertised as so

If the system fails to do so, the company can be sued, and system recall. unlike with L2 systems

 

[MarcusMaximus]

Yes a Level 3 system must be able to handle every odd cirumstance or to detect the circumstance within the agreed upon time interval with the consumer.

If the system fails to do so, the company can be sued, and system recall. unlike with L2 systems

Note that I said “without human intervention”.

 

[shukenmg]

As a programmer myself, I feel that many people don't understand that sometimes a program has a lot of hidden code and features that works but is not enabled.

So saying that V9 doesn't have any big important features is not really true.

Just take ULC for example, its there in the code but not available, so probably there's more hidden frameworks that we just don't know about in V9 that will enable more and more progress in the field of EAP and FSD.

 

[Silicon Desert]

You want to avoid having multicast devices (e.g. smart devices, etc.) on the same VLAN as the car.

Not sure what you mean by a multicast device. Can you give an example please? Since I have over 50 devices on the same network as the MX and never had any issues, I am probably going to guess that one or more of those devices is what you refer to. Thus, having a hard time understanding why they may interfere.

 

[chillaban]

Not sure what you mean by a multicast device. Can you give an example please? Since I have over 50 devices on the same network as the MX and never had any issues, I am probably going to guess that one or more of those devices is what you refer to. Thus, having a hard time understanding why they may interfere.

He really means anything that broadcasts multicast packets, especially DLNA players. Seems like excessive multicast traffic causes some Teslas to drop off the network.

 

[dhanson865]

Wonder how they interpret that data. I can often be "chill" when on AP and if the guy in front of me is only going a little slower, I will just stay behind him. So I might not confirm lane change options not because they are "wrong" but just because I dont want to deal with holding up folks in the fast lane to pass a guy I am only gaining on by a little bit.

It's probably the near collisions that teach better than the other possibilities.

* Collision - allowed (maybe learns)
* near collision - allowed (learns with greater confidence / higher value to the experience)
* object close - allowed (learns with lesser confidence / safer but less value to the experience for learning)
* allowed but nothing close (not much to learn, maybe not even considered?)
* denied - not even considered?

Maybe there are multiple categories under denied and some of those cases could be reviewed but it seems like a low payout path.

My point is even if you deny lane changes they can learn more from the people that allow without your denial harming the learning curve.

 

[Mobster]

As a programmer myself, I feel that many people don't understand that sometimes a program has a lot of hidden code and features that works but is not enabled.

Ok I need to settle this.

"Any sufficiently advanced technology is indistinguishable from magic."

......my 40 years in Aerospace Engineering, Computing and Global Network Management qualifies me to steal the solution from Arthur C. Clarke

 

[dhanson865]

I agree that v9 seems like a great improvement, but this... isn't how machine learning works. Sure, you're feeding data to Tesla constantly when EAP is in use. But expecting "exponential" improvements in autonomy is far fetched. Neural network machine learning is just fancy pattern matching. Yes the more patterns you have to train a specific model more granularly the more often you'll get high confidence results FOR THAT MODEL, but it's not like learning in a human sense where you master one thing and then build experientially off of that previous knowledge to learn a whole new thing.

Consider Montana or West Virginia or other low population density routes.

Part of the exponential growth is getting cars to go down roads they've never gone down before. Or getting enough cars to go down the same road so that variance in calibration can be averaged and give a higher confidence. Or getting the same edge condition to occur more frequently so it isn't considered too low to prioritize because someone misjudged it's rate of occurrence when the fleet was smaller.

There will be new ground, new leaps and bounds, it will seem to accelerate. It can't help but be pushed at a higher rate as we go from thousands to tens of thousands to hundreds of thousands to millions of cars on the road running the same hardware in an ever growing environment.

Will the neural net do it all magically just because someone drove down a road? No. But get a bunch of people driving down a new road and the engineers behind the neural net have new data and new challenges to overcome.

The fleet growth is exponential, maybe the neural net doesn't gain capability as quickly as the fleet grows but there is more than one curve in the universe that is exponential.