Knightshade
Well-Known Member
Russ Mitchell.
LOL as the kids say.
He's been doing BS hit pieces on Elon for years.
Much better summary of the actual contents of that "documentary" here:
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
The catastrophe of FSD and erosion of trust in Tesla
- Thread starter SilverString
- Start date
Russ Mitchell.
LOL as the kids say.
He's been doing BS hit pieces on Elon for years.
Much better summary of the actual contents of that "documentary" here:
bradtem
Robocar consultant
Compared to the Automakers, Google was the company taking crazy risks. I was working on Google's car then. But no question, Tesla is willing to take much bigger risks. So big that most people in the industry, who are not that risk averse (and downright crazy compared to car OEMs) think Tesla is taking too many risks, putting the whole industry at risk, not just themselves. Now, to be fair, Uber did that and it didn't hurt the industry as much as people expected so the fear of that has reduced a bit, but people still are worried, and also don't like the massive confusion Tesla has caused over the term self-driving.
Agreed that Tesla's methodology is riskier. And the perception of that risk is amplified due to Musk's missed predictions as well as public FSDb videos showing sketchy situations. But.... at this point, it seems clear to me that progress is still being made with a vision-only, AI-heavy methodology. There will be a point where people will start to appreciate how capable the car is doing with just cameras (even before it's considered very good). I'm already getting there, but it's easy to call me biased.
If 10.12 turns out to be a large incremental improvement like some of the early vids are showing, and the same rate of incremental progress continues, I think by the end of this year the old narratives about FSD are going to go stale quickly.
bradtem
Robocar consultant
I have videos of Tesla FSD on my own youtube channel. Many videos of Waymo rides are available on YT. Videos of Cruise are just starting to show, though not fully open yet. Chinese videos are specially selected so hard to test.
But while the intervention numbers published use different methods for different companies, the numbers of the best companies are in the tens of thousands of miles per intervention, and for Waymo in Chandler, we know that they have done probably 8 million miles without causing an accident. I personally can't get FSD to go more than a handful of miles without needing a serious intervention, and while there are videos, like the above of it going for longer periods, the existence of any trip with interventions needed at that frequency means you aren't even in the game yet, let alone comparable. A lot of people make this basic mistake, they say, "Hey, wow it drove for an hour, I'm impressed" when the bar is to drive an entire human lifetime. Any intervention seen by a single driver within a year of driving is an immediate flunk-out. People don't seem able to get that through their head. If you see even a single intervention necessary to prevent an accident after you drive for a year, you are still at the starting gate. To quote Elon Musk:
Now, as it turns out most people don't think it requires full real-world AI, at least not to drive in commercially viable taxi service areas, and that is what they are doing, and getting success at. None of the top teams are even trying what Tesla wants to try, and it's not because they don't think they are as smart as Tesla. It's because they think it's a fool's errand, to use Musk's phrase about lidar, for now, or rather a very longshot bet. There are a few smart teams, like Wayve and Waabi and a couple others, trying to do what Tesla is doing, but they won't pretend they are past the starting gate yet, let along proclaim they will have it working in 2022.
I appreciate your experience in your field. That said I am not looking to get stats. I want to see videos to get a warm fuzzy feel of how the competition is doing.
fyi - My history is in industrial automation where I have achieved reducing a factory floor to less than 10 employees that where traditionally required 350+ workers. I am providing this information only to let you know I am no joker unaware of how automation works. Been there. Know the difficulty level.
bradtem
Robocar consultant
Understand that all videos mean very little. They can give you some hints and clues but no video, on its own, will tell you that a project is close to ready. No year of driving can tell you that. Really.
On the other hand, a single drive can tell you that a project is not ready, and that it is very, very far from ready. Every drive I have had with Tesla FSD has been such a drive. Even when it doesn't need an intervention, it drives erratically. It pauses too long at turns and stop signs. It gets honked at. It lashes the wheel back and forth, it does jerky braking and acceleration. Even when it doesn't try to hit something. Which sometimes it doesn't do.
It baffles me that Elon thinks it's close. That he dares people to try it thinking they will agree with him. I guess he does that because some people who see it drive for 7 minutes like the drive above come away "impressed." Those people simply don't know what is impressive. It's not their fault, they don't have the experience. They should however know that they don't have the experience. Elon has no excuse.
It is not close at all. Even in controlled environments of an industrial shopfloor, it takes a lot to get automation working close to actual targets. An open road with unpredictable humans is going to require an amazing amount of neural processing that is predicting and processing reactions at the very least 250ms ahead of it itself.
Knightshade
Well-Known Member
To be fair, things like CNC machines need vastly more precision than cars do.
You're expecting to stop even at slow speeds with a foot or more between you and the car in front of you,... so if you're off by a couple inches either way no big deal....Likewise lane centering you've got inches on both sides to spare.
Run tolerances like that on an industrial shop floor and you'd be out of business pretty quickly.
a CNC machine is not what I referred to. In our industrial automation, the machines were designed to inspect, measure and determine the disposition on the next step of the component being manufactured at a rapid rate of few hundreds per minute with a tolerance of +/- 0.0002 inches.
A car going at 40mph will travel 58 feet per second, and the braking distance is 80feet. A life will be lost at this rate. The reaction time of the FSD system needs to be detect the human at least 110feet ahead of time in order to brake safely.
lolachampcar
Well-Known Member
Reading all this it occurs to me that I've seen this Musk movie before.
Keep crashing rockets till you succeed or run out of money.
Keep pushing out FSD attempts till you succeed or (take your pick) kill someone and get shut down or run out of credibility.
The all in Monopoly/Risk approach is entertaining.
Keep crashing rockets till you succeed or run out of money.
Keep pushing out FSD attempts till you succeed or (take your pick) kill someone and get shut down or run out of credibility.
The all in Monopoly/Risk approach is entertaining.
As soon as our world adopts glass scale encoders to our streets and zoned predefined motion of everything in its environment you will have an industrial comparison. CNC is Not a close comparison in my opinion.
Exactly the point. Either you turn the roads into a very well defined and predictable environment or bring up the neural net to make decisions even in the absence of frameworks & predictors. Those cars will then even run in India, Africa, etc
Terminator857
Active Member
Your not going to get good answers from videos. Waymo works well, but is only in tiny areas. Will they continue to expand at a snails pace over the next 5 years? I think so, but hopefully I'm wrong. On the flip side, one day FSD will get it right, likely 5 years down the road in my opinion. Then we will see Tesla way ahead. I also think Mobileye based products will be doing well in 5 years.
Intel's Mobileye Expands Self-Driving Car Tests, So You Can Buy One in 2025
Exclusive: Expanding to Miami and Stuttgart, Germany, means more rain, snow, traffic and freeways to challenge self-driving car engineers.
www.cnet.com
Last edited:
Knowing what it took to get SpaceX to where it is today. Knowing what it took to take Tesla to where it is today. Knowing that he seems to be hands on/engineer, etc.
Do we REALLY believe that he truly, 100% thought it was close? Do we TRULY think he thought we would be able to summon our cars from NYC to LA by 2017? Do we TRULY think he thought Tesla's would be driving around on the streets EMPTY, going to pickup/deliver humans by 3 years ago? Do we TRULY think he genuinely had honestly miscalculated FSD capabilities that badly?
Or is it more realistic to believe, that he knew by making those statements/combined with peoples knowledge of his prior successes("i have credibility" aspect), that those statements would equal hundreds of millions of dollars coming in from consumers for FSD purchases?
Similar threads
