WHOOOOA! Sandy Munro calls Tesla FSD CRAP!

[tidmutt]

Munro knows car manufacturing.

He has opinions on a lot of things, but what he knows is manufacturing.

You need to understand that when you watch his videos.

FSD is designed to operate on roads designed for human drivers. While FLIR might give some details normal vision doesn’t, it will also miss things which normal vision does see. Since the roads are designed for human sight that range of vision needs to be covered. Maybe cameras that cover both infra and normal ranges would work, but you gotta have the human visible spectrum covered 100%.

FSD always has the potential to do what humans do in low visibility conditions: Slow down. That’s what any other traffic on the road is going to do.

This.

I listen to Munro in regards to engineering, manufacturing cars but AI, neural networks etc. are not even slightly in his wheelhouse. He complained to Elon about how road markings need to be improved for FSD showing his complete lack of understanding of the approach Tesla is taking.

 

[tidmutt]

Ye

Not entirely inaccurate.

When I reserved my Cybertruck I didn’t bother with FSD because it seemed like a waste*. Likewise on my Model Y.

The big question is whether it will ever be more than a really interesting experiment. It *seems* like it’s turning a corner. It almost feels like convergence in calculus. Every year FSD gets better, but is always just a little shy of good enough.

Munro’s assumption is part of the problem with FSD is the cameras they use, but I don’t think that is the limiting factor here.


* In retrospect maybe I should have clicked that button so I had the option, but water under the bridge.

Yes, I do wonder if they will always be chasing the next 9. Clearly we can drive so it should be possible, but maybe the current approach will need yet another ground up rewrite as they figure out one fundamental flaw after the other.

Either way I admire their gumption and have to give them props.

 

[Ogre]

Yes, I do wonder if they will always be chasing the next 9. Clearly we can drive so it should be possible, but maybe the current approach will need yet another ground up rewrite as they figure out one fundamental flaw after the other.

Either way I admire their gumption and have to give them props.

Tesla has spent more than 5 years chasing this with billions of miles of sensory data and millions of man-hours tagging images.

Nobody else tilts at windmills that big.

 

[NKSplitter]

As an Aptera reservation holder I'm wondering if Sandy signaled some cause for concern at the end. At the 12:42 mark he says that he's learned his lesson and doesn't buy stock in companies anymore. I know Aptera isn't publicly traded yet; but he has previously said in relation to Aptera that he has broken his own rule and invested in a company he is working with. He may well just be stating that he doesn't invest in the stock market anymore; but I have a hard time shaking the feeling he was hinting that he regrets investing in Aptera looking at the current state of the project.

 

[Ogre]

As an Aptera reservation holder I'm wondering if Sandy signaled some cause for concern at the end. At the 12:42 mark he says that he's learned his lesson and doesn't buy stock in companies anymore. I know Aptera isn't publicly traded yet; but he has previously said in relation to Aptera that he has broken his own rule and invested in a company he is working with. He may well just be stating that he doesn't invest in the stock market anymore; but I have a hard time shaking the feeling he was hinting that he regrets investing in Aptera looking at the current state of the project.

I’m pretty sure Munro has said he doesn’t invest in companies in previous videos as well.

I sure hope Aptera is doing well. I don’t have a reservation but I’m rooting for them.

 

[JBee]

I am a little shocked that Tesla are using such low resolution and particularly low FPS cameras and are still achieving those results and high data rates. For robot CV it is fairly standard to use those types of resoultions as that works well enough for things that are close, but typically the frame rate is much higher.

The other thing is that theses sensors already have IR sensitivity without a IR filter in front of the lens, and you normally don't drive "normally and at speed" in bad weather conditions because of the other drivers not being able to see as well.

I can agree that more is better (I still haven't been persuaded that removing radar was necessary) and adding FLIR will definitely help with wildlife and country driving, but I can't see it improving vision based FSD in most common driving conditions.

 

[firsttruck]

This.

I listen to Munro in regards to engineering, manufacturing cars but AI, neural networks etc. are not even slightly in his wheelhouse.
....

Agree.

For FSD I listen to James. He has actual done neural network programming.

---------------------------------------------

James Douma on AI - 30 videos in playlist
Dave Lee on Investing

https://www.youtube.com/c/DaveLeeonInvesting/playlists



---------------------------------------------

 

[John K]

Bias tends to cloud judgement on solution approach.

FLIR to put simply is a sensor system. Programming to manage a left turn at an intersection is not solved by the sensor. What would FLIR add over the current camera systems? Would the costs be justified? Argument seeing through fog? Even with greater visibility, the car needs to drive at the same safe speed as as visual driving.

On the scale between hating on FSD and FSD is magical perfection, I take a pragmatic approach and view it as an assist tool.

@Ogre you should cancel your first reservation since your second has FSD reserved.
??

 

[Ogre]

Bias tends to cloud judgement on solution approach.

FLIR to put simply is a sensor system. Programming to manage a left turn at an intersection is not solved by the sensor. What would FLIR add over the current camera systems? Would the costs be justified? Argument seeing through fog? Even with greater visibility, the car needs to drive at the same safe speed as as visual driving.

On the scale between hating on FSD and FSD is magical perfection, I take a pragmatic approach and view it as an assist tool.

@Ogre you should cancel your first reservation since your second has FSD reserved.
??

The rest of your post is so spot on, that last idea must be great. I’ll get right on it.

 

[CostcoSamples]

I'm not saying Sandy is wrong, but I'm sure Tesla has considered FLIR in great detail.

 

[Dids]

Hey Sandy. Can Flir tell me if the light is red, green or yellow? No? Guess we still need cameras. I'm with you that Flir is great for occluded situations such as fog or snow but let's deal with those once fsd works for clear vision. Before you say sensor fusion let me remind you that we removed the cameras from the guidence system because we could not prioritize which sensor to follow when they disagreed. Tesla is right to go with pure vision, will it work in fog or snow? Not without decreasing speed but it will still be better than humans.
Besides the problem isn't the sensors. FSD problems are interpreting the data and predicting actions.

 

[Crissa]

I think Munro is right about the markings, but not for the reason he thinks.

We tolerate alot of mistakes by humans that are totally preventable just if we built things to encourage the safe behavior. Narrower lanes people naturally drive slower. Bike and pedestrian paths separate from the cars would be safer. And that autopilot wreck on 101 that took a driver's life? No one cared a whit to write an article about the other four who crashed there in the previous six months, or those afterwards.

The AI will be safer, but it will slow down and say, 'there's a hazard there'. And it'll be automated and undeniable.

-Crissa

 

[Zabhawkin]

Hey Sandy. Can Flir tell me if the light is red, green or yellow? No? Guess we still need cameras. I'm with you that Flir is great for occluded situations such as fog or snow but let's deal with those once fsd works for clear vision. Before you say sensor fusion let me remind you that we removed the cameras from the guidence system because we could not prioritize which sensor to follow when they disagreed. Tesla is right to go with pure vision, will it work in fog or snow? Not without decreasing speed but it will still be better than humans.
Besides the problem isn't the sensors. FSD problems are interpreting the data and predicting actions.

Actually yes, but only because the positioning of the red yellow and green lights are standardized. (so that colorblind people can tell)

With the bombs down Sadam Husseins chimney, those were laser guided bombs, yes the sight was infra-red, but was also at night where the infra-red cameras do much better. They are significantly less capable during the day when everything is hot. They do still work though.

 

[John K]

(Predator joke) FLIR is terrible seeing off road vehicles after they went mudding. (/PJ)

 

[Ogre]

Yes
He also stated China will be taking over the car industry. I lived in China and China is 10-20 years behind Tesla Chian and will never take over the world car market.

China is not 10-20 years behind the electric vehicle industry. Tesla buys batteries from Chinese CATL so a big chunk of the Model 3 SR is Chinese. So we’d better hope they are 10 years behind.

BYD and CATL are huge parts of the EV industry. It is entirely likely Tesla will use cells from both companies. If BYD can make an LFP “Blade” based vehicle for the US market, it would likely do quite well.

Geely group (Chinese) owns Volvo and Polestar and much of their manufacturing is done in China.

The Chinese don’t need to beat Tesla, most of the US and EU based legacy auto-companies are doing an absolutely terrible job moving to EVs. GM already outsourced most of the Bolt design to a Korean company. Would not surprise me to see similar partnerships with Ford or GM and either BYD or Geely.