Autonomous Car Progress

[EVNow]

The problem is that we don't always have a basis for accurately judging if we are being misled.

Look. There is history and thus reason for suspicion here. ”Elon is bad at predicting” just doesn’t quite cut it anymore for some of us with the history. I would say that is understandable.

But you are only looking at selective history.

You never look at all the cases when Elon promised/predicted something that all the naysayers were sure won't happen, but Elon made it happen.

If you don't have an explanation for that as well - then you are just a naysayer for the sake of being a naysayer. You no longer have a unified, consistent theory of Tesla. Infact this is the problem with the media as well - most of them focus only on things Tesla fails to deliver. Almost no word on incredible achievements. That is what makes their motive suspicious.

 

[electronblue]

But you are only looking at selective history.

You never look at all the cases when Elon promised/predicted something that all the naysayers were sure won't happen, but Elon made it happen.

If you don't have an explanation for that as well - then you are just a naysayer for the sake of being a naysayer. You no longer have a unified, consistent theory of Tesla. Infact this is the problem with the media as well - most of them focus only on things Tesla fails to deliver. Almost no word on incredible achievements. That is what makes their motive suspicious.

I don’t think I am. I think I am looking at relevant history.

I have given (and do give) plenty of credit to Tesla for basically creating the modern BEV industry. I like what they and Elon have accomplished there. I also give credit to SpaceX.

On EAP/FSD, I think the evidence is pointing towards the possibility that Elon made an uncharacteristic mistake, ovepromised — and indeed pre-sold — something he can’t deliver in a reasonable/relevant timeframe.

SolarCity may be another such mistake, though in that case I’d blame lackluster execution more so than lack of vision. But in the case of EAP/FSD it seems possible to me even Elon’s vision is uncharacteristically wrong. AP2 suite may not be up to the task and who knows if HW3 suite is either.

The evidence suggests FSD is not simply late like Roadster or Model S or Model X or SpaceX rockets were. This time it is different — and all the more so because Tesla has pre-sold the hardware and are now waiting to deliver software, again something that hasn’t usually happened in those previous instances. (I’ll not repeat the more minor P85D HP debacles etc.)

Again: My 2016 bought Model X turns three (the usual life of a lease) in a couple of months. The FSD pack has not received any features yet.

 

[EVNow]

I don’t think I am. I think I am looking at relevant history.

I have given (and do give) plenty of credit to Tesla for basically creating the modern BEV industry. I like what they and Elon have accomplished there. I also give credit to SpaceX.

The relevant history is exactly what you state in the second paragraph. For you FSD may be very different from what SpaceX has achieved - but that is you. Don't confuse yourself with Musk. You are not Musk !

When you are trying to figure out the motivation of other people, you can't think what you would do - you need to figure out how that person thinks and operates.

For Musk it is simply a tough problem - just like ramping up Model 3 volume was or getting rockets to land vertically back. History will show us whether he is right or not - but assuming that he is lying is a leap of faith in itself.

The case for an optimistic and hopeful serial entrepreneur is so much easier to make. His thinking is simple ...
- Look at a problem
- See if fundamentally ("first principles") it is solvable
- If so, all it takes in perseverance and it can be solved

Like Musk said about Model 3 ramp up - his timeline were pure guesses since he had not done it before. Same for FSD.

ps : Anyway as I've written before this line of posting is pure trolling on your part. There is a separate thread on Musk, you can post there. Or create a separate thread yourself and post there. To repeat this again and again in every thread reeks of ulterior motivation.

 

[Octo]

You seem to be under the impression that Musk is doing the actual work.

The main difference between his past successes and the FSD flop is that his personal value add (money and ability to motivate people) worked with development as in engineering something based on existing science whereas FSD is fundamentally different as it’s something humanity hasn’t even figured out yet on a basic level so no amount of optimism, cash or pressure can make it happen on a predictable timeline.

 

[electronblue]

The relevant history is exactly what you state in the second paragraph. For you FSD may be very different from what SpaceX has achieved - but that is you. Don't confuse yourself with Musk. You are not Musk !

When you are trying to figure out the motivation of other people, you can't think what you would do - you need to figure out how that person thinks and operates.

For Musk it is simply a tough problem - just like ramping up Model 3 volume was or getting rockets to land vertically back. History will show us whether he is right or not - but assuming that he is lying is a leap of faith in itself.

The case for an optimistic and hopeful serial entrepreneur is so much easier to make. His thinking is simple ...
- Look at a problem
- See if fundamentally ("first principles") it is solvable
- If so, all it takes in perseverance and it can be solved

Like Musk said about Model 3 ramp up - his timeline were pure guesses since he had not done it before. Same for FSD.

I would say in my opinion reality is more nuanced than that. I do think I get where and why Elon has succeeded — yet I am not convinced it necessarily applies to FSD given the track-record in this area in particular (and the somewhat different nature of the area as @Octo points out). This could be the exception to the rule, where Elon doesn’t deliver.

As for lies, I would not consider missing a long-term plan a lie. Of course not. But in some short-term incidents it does feel like Elon or Tesla may be a little liberal with the truth. It seems possible in some controversial instances. This is especially problematic in areas where a product has been pre-sold with specific wording and understanding presented.

ps : Anyway as I've written before this line of posting is pure trolling on your part. There is a separate thread on Musk, you can post there. Or create a separate thread yourself and post there. To repeat this again and again in every thread reeks of ulterior motivation.

I find it odd that you’d think this way: the information or direction emenating from a CEO of any company is relevant in topics related to that information/direction — let alone with a company like Tesla where the CEO is such an influential character.

You and I are not so unlike in this in any case. Our views of Elon’s track-record fuel both of our opinions on Tesla’s part in the Autonomous Car Progress — and I would say both are quite relevant in this thread.

 

[willow_hiller]

Interesting development on the policy side of autonomous vehicles: Florida Department of Transportation building facility to test autonomous cars

"Florida Department of Transportation building facility to test autonomous cars"

465 acre track dedicated to testing autonomous vehicles. Most narratives I read on the path to regulatory approval involve manufacturer-collected statistics, but it will be interesting to see how different autonomous systems respond to government testing. Importantly, for manufacturers who rely heavily on pre-mapping areas, would the Florida DOT give them access to these testing facilities in advance? Or would it be too easy to cheat the tests if manufacturers know the parameters. A generalized solution might really shine in such a scenario.

 

[diplomat33]

Interesting development on the policy side of autonomous vehicles: Florida Department of Transportation building facility to test autonomous cars

"Florida Department of Transportation building facility to test autonomous cars"

465 acre track dedicated to testing autonomous vehicles. Most narratives I read on the path to regulatory approval involve manufacturer-collected statistics, but it will be interesting to see how different autonomous systems respond to government testing. Importantly, for manufacturers who rely heavily on pre-mapping areas, would the Florida DOT give them access to these testing facilities in advance? Or would it be too easy to cheat the tests if manufacturers know the parameters. A generalized solution might really shine in such a scenario.

Personally, I don't think the DOT should give manufacturers a map of the testing area in advance because I think that would be a form of cheating. The manufacturer could "hard code" how to handle the test and it would not be a true test of the autonomous system under real world conditions. Of course, this would favor companies like Tesla with a general solution approach but I think it should be that way. A general solution is more desirable.

 

[Daniel in SD]

Interesting development on the policy side of autonomous vehicles: Florida Department of Transportation building facility to test autonomous cars

"Florida Department of Transportation building facility to test autonomous cars"

465 acre track dedicated to testing autonomous vehicles. Most narratives I read on the path to regulatory approval involve manufacturer-collected statistics, but it will be interesting to see how different autonomous systems respond to government testing. Importantly, for manufacturers who rely heavily on pre-mapping areas, would the Florida DOT give them access to these testing facilities in advance? Or would it be too easy to cheat the tests if manufacturers know the parameters. A generalized solution might really shine in such a scenario.

This seems like a huge waste of time and money. It sounds a lot like the DARPA Urban challenge which 6 teams completed successfully in 2007. The average driver gets in a serious accident every 500,000 miles. Is it really possible to verify that level of safety on a closed course? I'm skeptical.

 

[willow_hiller]

This seems like a huge waste of time and money. It sounds a lot like the DARPA Urban challenge which 6 teams completed successfully in 2007. The average driver gets in a serious accident every 500,000 miles. Is it really possible to verify that level of safety on a closed course? I'm skeptical.

Even if the testing ends up fruitless for the automakers, I'm hopeful this sort of facility familiarizes the DOT with how autonomous vehicles operate. While the software is improving, the DOT could experiment with ways to arrange construction cones to make sure they're obvious barriers to all the major brands, experiment with different lane line configurations, etc.

 

[Daniel in SD]

Even if the testing ends up fruitless for the automakers, I'm hopeful this sort of facility familiarizes the DOT with how autonomous vehicles operate. While the software is improving, the DOT could experiment with ways to arrange construction cones to make sure they're obvious barriers to all the major brands, experiment with different lane line configurations, etc.

I'm fairly confident that state of art autonomous vehicles can already deal with the standards already in place for construction zone markings and lane line markings. That's not the problem, it's the edge cases which I don't see how you can devise a test for. You can relatively easily program an AV to deal with any test and then simulate it with many variations of the same test, that's what Waymo and Cruise do. The hardest thing to do is to predict the behavior of other drivers, pedestrians, and road users. Humans are quite good at that and machines are very bad at it. Take a look at this demo video from Cruise, are you really going to be able simulate testing harder than this?

 

[motocoder]

I'm fairly confident that state of art autonomous vehicles can already deal with the standards already in place for construction zone markings and lane line markings. That's not the problem, it's the edge cases which I don't see how you can devise a test for. You can relatively easily program an AV to deal with any test and then simulate it with many variations of the same test, that's what Waymo and Cruise do. The hardest thing to do is to predict the behavior of other drivers, pedestrians, and road users. Humans are quite good at that and machines are very bad at it. Take a look at this demo video from Cruise, are you really going to be able simulate testing harder than this?

Wow, that video was incredible.

 

[Octo]

I'm fairly confident that state of art autonomous vehicles can already deal with the standards already in place for construction zone markings and lane line markings. That's not the problem, it's the edge cases which I don't see how you can devise a test for. You can relatively easily program an AV to deal with any test and then simulate it with many variations of the same test, that's what Waymo and Cruise do. The hardest thing to do is to predict the behavior of other drivers, pedestrians, and road users. Humans are quite good at that and machines are very bad at it. Take a look at this demo video from Cruise, are you really going to be able simulate testing harder than this?

Performs slightly better than AP on my Tesla.

 

[Daniel in SD]

Performs slightly better than AP on my Tesla.

Should have bought FSD
Automatic city driving is coming later this year.

 

[Octo]

Should have bought FSD
Automatic city driving is coming later this year.

Haha! Maybe for Christmas!!!

 

[thewishmaster]

I'm fairly confident that state of art autonomous vehicles can already deal with the standards already in place for construction zone markings and lane line markings. That's not the problem, it's the edge cases which I don't see how you can devise a test for. You can relatively easily program an AV to deal with any test and then simulate it with many variations of the same test, that's what Waymo and Cruise do. The hardest thing to do is to predict the behavior of other drivers, pedestrians, and road users. Humans are quite good at that and machines are very bad at it. Take a look at this demo video from Cruise, are you really going to be able simulate testing harder than this?

They don't show you all the times the Cruise cars get stuck at crosswalks or before/after intersections for various mostly unjustified reasons I see it all the time though

 

[Daniel in SD]

They don't show you all the times the Cruise cars get stuck at crosswalks or before/after intersections for various mostly unjustified reasons I see it all the time though

Yep. If Cruise cars could navigate SF for hundreds of thousands of miles without a problem we’d already have robotaxis. My point is it would be extraordinarily difficult to simulate the testing conditions in that video at a test facility.

 

[thewishmaster]

Yep. If Cruise cars could navigate SF for hundreds of thousands of miles without a problem we’d already have robotaxis. My point is it would be extraordinarily difficult to simulate the testing conditions in that video at a test facility.

Yeah, definitely.

 

[diplomat33]

We have some data on Waymo's robotaxi pilot program in California:
- Waymo transported 6,266 passengers in self-driving Chrysler Pacifica minivans in its first month participating in a robotaxi pilot program in California.
- Waymo completed 4,678 passenger trips in July — plus another 12 trips for educational purposes. An average of 156 trips every day that month.
- Waymo’s fleet completed 4,678 trips and logged 59,886 miles during the final month of the quarter.

https://techcrunch.com/2019/09/16/w...6200-riders-in-its-first-month-in-california/

 

[Phlier]

The Cruise video and the Waymo information really is incredible, but IMO, it's kind of an apples/oranges comparison to what Tesla is doing.

The Cruise video is a geofenced area, and last time I read about the tech, it was based on HD mapping and LIDAR. And again, last time I checked, Waymo was similarly based. So they are good at what they do, as long as they stay in the confines of their geofenced areas, and the HD maps are kept up to date.

Tesla's approach if very different, as we all know.

Which tech will win in the end? Stay tuned!

 

[Tam]

If you have 1 hour for Nova episode "Look who's driving" premiered tonight:

Look Who's Driving

It should be a great public relations to educate the public but both Uber and Tesla declined to participate in this document (You can guess what both got in common).

It doesn't sound like the technology is here.