Apples and oranges comparisons:
- comparing production ADAS to prototype self-driving
- comparing anything to a prototype that doesn’t even exist yet
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Waymo’s “commercial” ride-hailing service is... not yet what I hoped
- Thread starter strangecosmos
- Start date
Apples and oranges comparisons:
There were some comparisons with the Tesla demo, though and I think it would be fair to compare this demo to the other demos too.
Really? Is that why there will be millions of L2+, L3 and L4 cars in the timeframe you listed powered by Mobileye?
You know L3 and L4 cars. Which Trent says is equivalent to TACC and being able to do the below is mean-less.
:no_upscale()/cdn.vox-cdn.com/uploads/chorus_image/image/45753474/volvo-drive-me-1.0.0.gif)
Last edited:
At the time they started they didn't think about lidar yet. And in fact had a very specific goal in mind. So they went to GM to make some joint development. GM laughed at them and there was no deal made, that's when they started developing it on their own.
Soon after they realized that it will take a long time to develop the system based solely on image recognition. And since lidar has distance information and its own light source, it looked like a better alternative. No one picks more expensive parts only for a hobby.
I feel Elon is asking for a slap in the face again. Just like with the manufacturing. "Faster Than Any Car Assembly Line" ended up being the slowest.
Didn’t the Google self-driving car project use lidar from the very beginning? Self-driving cars have been using lidar since the 2004 DARPA Grand Challenge.
I may be wrong about when they started using it. The GM story was around ~2011 before that funny car.
CarlK
Active Member
That's a funny and totally fabricated story. "Project Chauffeur" started as one of those moonshot projects under the super secretive X lab. Lidar was chosen from the beginning for good reasons. They need all the help to make it to work. It was a small project with only a dozen or so engineers working on it. Accumulate millions or billions miles of data had never even entered their mind. When that design (which Elon describes as on crutches} was set they are not able to change it anymore unless if they want to start the SW over again. Also there were never intentions to make it a joint venture or even let anyone to know what's going on there. It was just a secretive small project in the beginning. Maybe they did talk to some in the later stage but those moonshot projects never had any set goals and, with the exception of Waymo, none went anywhere.
As for how Elon does things you probably did not notice he has delivered everything he promised, just not at the schedule he said he would. The same will be true for manufacturing line speed and autonomous driving capability. He never quits until it's done.
Last edited:
CarlK
Active Member
Really? Is that why there will be millions of L2+, L3 and L4 cars in the timeframe you listed powered by Mobileye?
You know L3 and L4 cars. Which Trent says is equivalent to TACC and being able to do the below is mean-less.
You still don't understand that demos or presentations mean nothing? Everyone could do those things easily.
What was later known as the Google self-driving car project started in 2009, and, as I recall, Astro Teller (Google X’s Captain of Moonshots) said the original product was a highway driving assist, similar to Navigate on Autopilot.
2012 was the year that Hinton et al. used a deep convolutional neural network to win the ImageNet challenge. So 2012 was a breakthrough year for computer vision. It was the advent of deep learning.
Google started on vehicle autonomy in 2009, and Tesla started work on HW1 Autopilot in 2013 I believe. Interesting that Google started pre-2012 and Tesla started post-2012.
Although what matters more than the timing is probably the business models. Tesla makes many thousands of production cars and sells production ADAS. Google has always been working on a small-scale R&D project. So a vehicle’s hardware can cost $200,000 for all Google cares, whereas Tesla has to tightly control every dollar of cost. There is no disadvantage to Google piling on more sensors, and Tesla just can’t afford to use high-grade lidar. So both companies are doing what makes sense for them.
A $200,000 robotaxi could probably be profitable, even assuming hardware costs don’t come down (which they of course will). So if you think you can get to full autonomy just by testing a few hundred or a few thousand robotaxis, this model totally makes sense. But if you think full autonomy requires training data gleaned from billions of miles, the only way to do that is try to put autonomy-grade hardware in production cars.
2012 was the year that Hinton et al. used a deep convolutional neural network to win the ImageNet challenge. So 2012 was a breakthrough year for computer vision. It was the advent of deep learning.
Google started on vehicle autonomy in 2009, and Tesla started work on HW1 Autopilot in 2013 I believe. Interesting that Google started pre-2012 and Tesla started post-2012.
Although what matters more than the timing is probably the business models. Tesla makes many thousands of production cars and sells production ADAS. Google has always been working on a small-scale R&D project. So a vehicle’s hardware can cost $200,000 for all Google cares, whereas Tesla has to tightly control every dollar of cost. There is no disadvantage to Google piling on more sensors, and Tesla just can’t afford to use high-grade lidar. So both companies are doing what makes sense for them.
A $200,000 robotaxi could probably be profitable, even assuming hardware costs don’t come down (which they of course will). So if you think you can get to full autonomy just by testing a few hundred or a few thousand robotaxis, this model totally makes sense. But if you think full autonomy requires training data gleaned from billions of miles, the only way to do that is try to put autonomy-grade hardware in production cars.
Last edited:
This is another poor attempt (no surprise here) to frame Tesla as having some NN advantage. First of all Tesla started working on Tesla Vision internally around 2016. (we have dates on linkedin to confirm)
Tesla's complete work on AP1 were simply developing control algorithm on Mobileye's NN output and nothing more.
It only takes a tiny bit of research to see the truth. We know exactly what happened during the NN breakthrough and it was Google immediately buying the company and the researchers and implementing NN into their cars.
Google Hires Brains that Helped Supercharge Machine Learning
Here's how the events transpired.
Inside the lab where Waymo is building the brains for its driverless cars
There was a mad dash from everyone to implement NN everywhere. This whole idea of pre NN vs post NN is complete hogwash perpetrated by Tesla fans to hint at an idea of an advantage.
In a Big Network of Computers, Evidence of Machine Learning
And google's goal was never to create a highway driver assist. Again a simple google search is all that's needed.
Tesla's complete work on AP1 were simply developing control algorithm on Mobileye's NN output and nothing more.
It only takes a tiny bit of research to see the truth. We know exactly what happened during the NN breakthrough and it was Google immediately buying the company and the researchers and implementing NN into their cars.
Google Hires Brains that Helped Supercharge Machine Learning
Here's how the events transpired.
Inside the lab where Waymo is building the brains for its driverless cars
There was a mad dash from everyone to implement NN everywhere. This whole idea of pre NN vs post NN is complete hogwash perpetrated by Tesla fans to hint at an idea of an advantage.
In a Big Network of Computers, Evidence of Machine Learning
And google's goal was never to create a highway driver assist. Again a simple google search is all that's needed.
Last edited:
Working in this industry I hear things. And they are not always the same as what's publicly available.
CarlK
Active Member
Pretty much everyone's demo runs are staged or gamed that way. Tesla's capability is way more difficult to achieve than what others are doing. Anything it says its system could do will be scrutinized by tens or hundreds thousands of users in all locations under all different conditions. The payback is in the end it will be the only company that will be able to achieve general autonomous capability to sell anyone a car to be used anywhere in the country. Not saying it's 100% sure thing that this will be achieved in how many years but it will be the only company that even has a chance to get there. Not another one is even trying.
Last edited:
CarlK
Active Member
Fair enough. Can't tell you how but I happen to know a lot of things happened there too. No I don't work for the CIA. Next guess.
Last edited:
While the Tesla video is a demo because it was a one/off, other companies ain't because its based on a test fleet and reflects the progress they have made on the way to reach their production goals/date.
You first have to define what exactly is a demo? Is a prototype car that is undergoing testing nationwide a demo? Did you call model 3 a demo when you saw it because it was tested on the streets in 2016? Or how about the Tesla Semi? Do you call that a demo too?
Did you say "don't compare Model 3 demo to Chevy Bolt, an existing car that's for sale?" I bet you no one did. So why are you calling Self driving test fleets demos? they are not demos just like the Tesla semi or the model 3 testing fleets wasn't a demo.
When you see the roadster out testing, do you say? that's not real, that's a DEMO! Please don't compare a demo to the BMW i8!
These Self Driving Systems HAS to be tested in test fleets before they are released in their scheduled release date ranging from 2019 to 2021 depending on the company.
Another issue with the statement is that AP2 sucks compared to others! You are saying don't compare A to B as though B is better than A. AP is simply lane keeping assist and adaptive cruise control. Nothing more, nothing less. Autopilot 2 isn't even better than AP1 (most people prefer AP1 over AP2) or Supercruise for that matter. Saying don't compare waymo to Tesla AP is like saying don't compare a supercomputer that is coming out next year to a T1 calculator that you have right now because the supercomputer is just a demo. Its absurd!
The question everyone should be asking is, will Tesla have a L3, L4 car when these test fleets go into production?
How is having a map of a city or the entire US highway system gaming it? you do realize the routes ain't planned and there is an infinite amount of routes that a rider can take? Elon excuse of not wanting to "game it" as to why they hadn't done cross country is a complete lie. His idea of gaming it is taking one specific route which no one does. If your system actually worked. You would be able to give it a HD map and it will drive autonomously anywhere in it.
Hello, visitor from the future. On what date was the FSD delivered and what's the TSLA stock price on that day? Anything other interesting you want to tell us about the future? When did Elon Musk die (clearly since he met all his promises he was not doing any new ones for a while too). Would be great if you can also enumerate completion dates for all his promises.
Cherry-picking some best-case clips and ignoring failures is pretty much the same as a demo.
CarlK
Active Member
It's not just HD map but also route training. To have those ready for all roads is not gaming it. To have those ready only for roads you're doing demo to give the impression the system works on all roads is gaming it.
It is ready for all roads in the geofenced area. Ofcourse what gonna be publicly displayed is the road that is most tested on until they are ready for production.
For example Waymo has a 100 sq.mile that their early riders can go anywhere in. But then they also have a bigger geofenced area that covers the entire metro phoenix area that they are testing in that they plan to enlarge to. And then they have a smaller geofenced area that they are testing with no safety driver at all.
Again there's no gaming its simply logistics. Elon gaming is simply wrong. Their system is simply incapable of going 3,000 miles (cross country) without requiring intervention. While you can define a static route to take. You can't game road users and objects. Your system still have to handle it.
Please quote where I said “L4 cars” are “equivalent to TACC”.
Lol you're reaching,
Which does equates to L3 being merely TACC.
My post to Carl was about L3 and L4, I never made a distinction between the two levels of which the statement was referring to because it wasn't needed.
First and foremost: Carl knew which one was being referred to because he's actively participating in the thread.
Secondly: Not only that, but you can have L4 Traffic Jam systems aswell and you are basically saying if its not changing lanes as part of its ODD then its worthless which is the point of the pic.
Last edited:
Similar threads
- Article
