Elon Musk: EAP solved, on track for FSD completion in 2019 (No one else is close!)

[diplomat33]

It's not there yet but I do think that Nav on AP will eventually bring AP up to Level 3 Autonomy on the highway. Here's why I think that. I was reminded in a previous thread that a big requirement for L3 is that the car needs to be able to notify the driver in advance when the driver will need to take back control. Nav on AP introduces some new warnings that AP did not have before that I think come close to doing that:
- It warns the driver in advance when it is about to disengage because it reaches the end of a highway.
- It warns the driver of limited functionality when it detects a construction zone ahead.
- It warns the driver of limited functionality when it detects severe weather.
- It warns the driver in advance that it will disengage when it detects an exit that it cannot handle.

So when Nav on AP is able to handle auto lane changes without driver input and thus is able to handle large parts of highway driving on its own without driver input and notifies the driver when it will need to disengage, I would think that comes pretty darn close to Level 3 Autonomy.

 

[Wooloomooloo]

So when Nav on AP is able to handle auto lane changes without driver input and thus is able to handle large parts of highway driving on its own without driver input and notifies the driver when it will need to disengage, I would think that comes pretty darn close to Level 3 Autonomy.

Well it asks for confirmation, so technically it's not "without driver input". I know it's a nuance, but it's obviously done that way because it's still beta and can't make those decisions autonomously yet.

Regardless, we need people like Elon to believe stuff can happen in 6 months, so it actually happens in 2 years, rather than 20 years...

 

[diplomat33]

Well it asks for confirmation, so technically it's not "without driver input". I know it's a nuance, but it's obviously done that way because it's still beta and can't make those decisions autonomously yet.

I was talking about in the future, when Nav on AP will not require the driver confirmation anymore.

 

[RDoc]

I think they are. There was a report awhile back that Tesla is recruiting employees into a FSD beta group:
Tesla reportedly starts incentivized Full Self-Driving beta test program for employees

Hmm. It's odd isn't it that they want the feedback by the end of next year? FSD is going to be out in the next 2 months or less, so why do they want feedback from an insider program a year after it's out?

 

[diplomat33]

Hmm. It's odd isn't it that they want the feedback by the end of next year? FSD is going to be out in the next 2 months or less, so why do they want feedback from an insider program a year after it's out?

I assume you are being sarcastic because FSD will not be out in the next 2 months. The fact that Tesla wants feedback by the end of next year is a pretty good indicator that Tesla expects to have some form of FSD for testing sometime next year. And next year is also when Tesla expects the FSD chip to be ready. So it adds up. I would say that Tesla is probably hoping to start testing their FSD in private beta testing next year when the FSD chip is ready.

 

[u00mem9]

Hmm. It's odd isn't it that they want the feedback by the end of next year? FSD is going to be out in the next 2 months or less, so why do they want feedback from an insider program a year after it's out?

You’re kidding, or the victim of a salesman?

 

[buttershrimp]

The biggest and latest thing for me that convinces me massive progress is happening is seeing how amazing the lane changing capability suddenly has been. I don't have Nav on AP yet but it's remarkable to see how much the update enables. I can't imagine how worried the folks at cadillac and nissan and every other automaker are.... just when they are starting to be able to stay in lanes on highways that are nannyscaped.... they are like, "ughhhhh, crap, lane change?"

 

[cwerdna]

Another note is those disconnects don't show how much of that is Highway???
Unless the reports which I didn't open show more Data (on Tablet right now)?
Super Cruise was built for Highway which is easier. I'd hope they're trying to expand that.

For Cruise https://www.dmv.ca.gov/portal/wcm/c...-5a2c9f10673b/GM_Cruise.pdf?MOD=AJPERES&CVID= says

Last year we drove over 125,000 miles on San Francisco’s complex city streets. All the attached data
is from this urban driving. We drive in San Francisco because it allows us to improve more quickly.
Cities like San Francisco contain significantly more people, cars, and cyclists that our self-driving
vehicles must be aware of at any given time. That makes San Francisco one of the hardest places to
test a self-driving vehicle, and creates a rich environment for testing our object detection, prediction,
and response functions. It also helps us validate our vehicles’ self-driving skills faster than testing in a
suburban location alone. So, we drive here because by doing so we get better faster.

They've posted Why testing self-driving cars in SF is challenging but necessary before.

I suspect Cruise Automation's testing efforts in cities like SF is orthogonal to expanding Super Cruise coverage. I haven't followed the latter's progress but found Ushr Powers HD Map Data for Cadillac Super Cruise™. Seems like OUR STORY is responsible. Press release says

"Ushr's HD mapping plays a critical role in assuring that Super Cruise operates as promised," said Jon Lauckner, Chief Technology Officer, General Motors. "We knew at a very early stage that high-definition mapping was critical for Super Cruise and there were few companies who could provide reliable data and that's why GM Ventures invested in Ushr."

Using high precision LiDAR technology, Ushr has mapped all road features of every mile of limited-access highway in the U.S. and Canada to under four inches in a global coordinate frame. Ushr will collect roadway changes on an ongoing basis and provide frequent updates to GM.

For Waymo, https://www.dmv.ca.gov/portal/wcm/c...-97f6f24b23cc/Waymofull.pdf?MOD=AJPERES&CVID= says

As of the end of November 2017, Waymo has operated vehicles on public roads in autonomous mode for
over 4 million miles across more than 20 U.S. cities. For the 12 month period covering this report,
Waymo completed 352,545 miles in autonomous mode in California – with the vast majority of the driving
on surface streets.

Waymo/Google specifically said long ago they've shifted their focus away from racking up as many miles as possible (easy on highway) to harder scenarios. But, I agree that SF is probably way tougher than Phoenix.

 

[E-Ryc]

The biggest and latest thing for me that convinces me massive progress is happening is seeing how amazing the lane changing capability suddenly has been.

It's not sudden, the work on v9 started probably at the moment v8 got released (year ago, v8 started to be in "maintenance mode").

Yeah, they make some progress on features, but not much progress on stability, unfortunately (i.e. memory corruption in other thread). And, believe or not, getting rid of bugs in existing program is way more difficult than properly testing and getting rid of them in the development phase.

 

[Kruegmeister]

Technically speaking. I would rather not have FSD if my car has to look like a Waymo.
If that big ugly thing on top (I assume Lidar???) Cannot come down in size, then I'll continue to use whatever Level Tesla or GM or Ford can deliver without that thing on top. But that's me...

 

[Kant.Ing]

When cars become FSD, we don't even need to own our own cars. I would care no more. Mission accomplished.

 

[buttershrimp]

It's not sudden, the work on v9 started probably at the moment v8 got released (year ago, v8 started to be in "maintenance mode").

Yeah, they make some progress on features, but not much progress on stability, unfortunately (i.e. memory corruption in other thread). And, believe or not, getting rid of bugs in existing program is way more difficult than properly testing and getting rid of them in the development phase.

Huh? Sudden as in the update has suddenly added immediate functionality and grace to lane changing where it is instantly more sophisticated, confident and able to interweave in between complicated situations .... this is not the NOA that people are talking about.... I'm talking lane change initiated in straight up bad situations.... it's patient and does a brilliant job.

Also, I'm not talking about the beautiful highways en route to Karlovy Vary, I'm talking I-5 rush hour road rage near Compton!.

 

[CyVolt]

Anyone driving a Tesla today knows it is a long way away from FSD. When a car can't merge, can't stop on its own at lights, and can't take turns then it's impossible to be FSD.

 

[BrentwoodX]

If only Elon could recognize the genius of the critics that share their comments on this forum. FSD would be solved in minutes!

 

[diplomat33]

Anyone driving a Tesla today knows it is a long way away from FSD. When a car can't merge, can't stop on its own at lights, and can't take turns then it's impossible to be FSD.

AP is a long way from FSD now but it is getting closer with every update. I am sure our cars will be able to stop at lights, merge, make turns etc in the future.

If only Elon could recognize the genius of the critics that share their comments on this forum. FSD would be solved in minutes!

If Musk listened to the critics, our Teslas would get a big ugly LIDAR pod stuck on the top. LOL.

 

[CliffSD]

I think it's ridiculous for folks to say things like "can't stop at lights"-- it's more accurate to say "doesn't". Just because a feature hasn't been released doesn't mean it's not capable of doing it.

Also, I drive 80 miles to work on Socal freeways and my Nav on AP is really, really good. At merging, driving in the slow lane when others are merging, and difficult lane changes. Sure, it occasionally appears drunk when it aborts a lane change and then tries again-- but even when this happens it's still successful in the end. Successful, SO FAR, at least, but yeah. May not drive exactly like I would, but it's leaps and bounds ahead of v8 autopilot.

Maybe I'm just fortunate to have better training and map learning/adaptation in a Tesla hot spot like Socal, so I'm curious if other states have as much success with NoA as my experience.

 

[Mjsais]

My Tesla is pre AP so I don’t have personal experience with AP. However, if Jason Hughes says that Nav on AP is not ready, I believe it. See,

Navigate on Autopilot is Useless (2018.42.3)

Also, Bloomberg - Are you a robot?

Others are building HW and SW that although not perfect seems to be at par or better than Tesla’s current version. I really would like to see Tesla become the leader in this field, but I’m not convinced that they currently are.

Finally, I would not trust any autonomous driving system and current Emergency Braking Systems until data showed thay they could avoid collisions with pedestrians, bikers and stop objects 99% of time in good or bad weather.

 

[timvracer]

Remind me again which car is out there that I can go buy that is ahead of Tesla in automated driving?

Others are building HW and SW that although not perfect seems to be at par or better than Tesla’s current version. I really would like to see Tesla become the leader in this field, but I’m not convinced that they currently are.

 

[djgarrett21]

My Tesla is pre AP so I don’t have personal experience with AP. However, if Jason Hughes says that Nav on AP is not ready, I believe it. See,

Navigate on Autopilot is Useless (2018.42.3)

Also, Bloomberg - Are you a robot?

Others are building HW and SW that although not perfect seems to be at par or better than Tesla’s current version. I really would like to see Tesla become the leader in this field, but I’m not convinced that they currently are.

Finally, I would not trust any autonomous driving system and current Emergency Braking Systems until data showed thay they could avoid collisions with pedestrians, bikers and stop objects 99% of time in good or bad weather.

Facepalm quote from Mobileye:

“It was a very unique situation,” he said. “We’d never anticipated something like this.” Shashua added that Mobileye is also making tweaks to the hardware meant to protect the car’s computers from electromagnetic interference with hopes of ensuring that similar incidents do not occur in the future.​

So when they have two sensors performing the same function and the outputs don't agree Mobileye's response is "we'd never anticipated something like this". What!? This is some very basic systems engineering stuff here. ANY time you have redundancy in a system you have to consider what you will do when outputs don't agree. And it still doesn't sound like they addressed it (at least not from the article) instead talking to how they will be hardening against EMI in the future. Kudos to them for that I guess (still a miss in the first place though).

I hope Tesla is paying attention to stuff like this. I'd love to see them become a leader in the field too and maybe they have a better chance at it than I thought seeing how other teams are making some very basic errors of their own.

 

[diplomat33]

Others are building HW and SW that although not perfect seems to be at par or better than Tesla’s current version. I really would like to see Tesla become the leader in this field, but I’m not convinced that they currently are.

Completely disagree! Yes, in terms of FSD, there are companies like Waymo that have better systems than Tesla. But in terms of driver assist that you can actually use in your car now, Tesla is THE leader. Show me another driver assist system that you can buy now that can do what Nav on AP does. It does not exist. Only Tesla! Probably the closest to Tesla is Supercruise and it can't even do auto lane changes.