WHOOOOA! Sandy Munro calls Tesla FSD CRAP!

[firsttruck]

Point counterpoint: Is Sandy Munro right that Tesla FSD IS CRAP?! Do we need FLIR??
In a recent E For Electric interview (linked below), Sandy Munro, automotive and engineering expert par excellence claimed that Tesla's FSD is crap because it needs FLIR (Forward Looking InfraRed) in order to work properly and solve full self driving. I take the opportunity here to provide some counterpoints, and explain why in my humble opinion FLIR is not necessary for Tesla to solve FSD. Let's take a look!
Mar 21, 2022
Dr. Know-it-all Knows it all

-----

 

[Bill906]

What bothered me about the interview was I believe Alex wanted to ask Sandy the same thing I did, but did not ask it correctly. He kept asking if FLIR would see a stop sign. I think the more precise question would have been if FLIR could READ signs. I don't know the limitations of FLIR (which is why I wanted to ask), I'm guessing FLIR could see a stop sign and determine it's a stop sign by it's shape. But can it differentiate between Speed Limit 35, Speed Limit 75, Pass With Care and Do Not Pass signs? What about construction signs. Bridge Out, vs. Flag Person Ahead, vs Right Lane Closed.

Addendum: I wrote this before seeing the youtube video @firsttruck posted above. The video covered my question very well.

 

[Zabhawkin]

What bothered me about the interview was I believe Alex wanted to ask Sandy the same thing I did, but did not ask it correctly. He kept asking if FLIR would see a stop sign. I think the more precise question would have been if FLIR could READ signs. I don't know the limitations of FLIR (which is why I wanted to ask), I'm guessing FLIR could see a stop sign and determine it's a stop sign by it's shape. But can it differentiate between Speed Limit 35, Speed Limit 75, Pass With Care and Do Not Pass signs? What about construction signs. Bridge Out, vs. Flag Person Ahead, vs Right Lane Closed.

Yes, it would still be able to make out the letters on the stop sign. We use one at work and it can see writing on painted metal.

Using infra-red would definitely give FSD the ability to be significantly better than a human as far as night/rainy/foggy driving. But yeah cost would be an issue.

 

[SSonnentag]

FSD is kinda hot garbage rn tbh

Summon is klunky, autopark is slower than my grandma, phantom braking is real. The beta is much better, and is advancing quickly, but...it's just a beta out to a few thousand people.

EAP is much better than FSD Beta in my opinion. The phantom braking every 30 seconds to 3 minutes during night driving while using FSD Beta makes cruise control and FSD/AP unusable.

 

[Rozonoe]

FSD cameras in Cybertruck HW4.0

Tesla's FSD hardware 4.0 to use cameras with LED flicker mitigation
January 13, 2022
By Nuno Cristovao

At Tesla's AI Day event last summer, Elon answered some questions during the Q&A portion of the event that revealed Tesla's plans for FSD hardware.

Elon informally announced that Tesla is already researching the next major hardware revision for full self-driving.

Tesla's camera housing

Elon said that he expects the current hardware in today's vehicles to be capable of "super-human" driving, but Tesla will continue to update their cars to have the most powerful hardware available.

Elon guessed that Tesla may release FSD hardware 4.0 with the introduction of the Cybertruck. Since the Cybertruck won't be released until the end of this year at the earliest, we may even see new hardware before its arrival, although this is unlikely.

In another question related to new hardware, Elon also said that Tesla is looking into new cameras. He was clear that the current cameras aren't a hindrance, but if Tesla can do better, then they will.

Elon didn't go into details about whether the new cameras would be tied to the Cybertruck or hardware 4.0. He also didn't mention specifically whether the new cameras would be higher resolution, contain larger sensors, or possibly even be placed in different locations.

Tesla currently utilizes three separate front-facing cameras in the housing at the top of the windshield. The camera unit contains wide-angle, standard, and narrow view lenses with three separate CMOS sensors.

Each sensor has a resolution of 1280x960 or about 1.2 megapixels.

New Camera
According to Chris Zheng on Twitter, who has connections with some Tesla suppliers, Tesla is planning to use Sony's new IMX490 automotive sensor in their upcoming hardware 4.0 revision.

Wide-angle camera comparison

This new sensor offers many benefits over the current sensors that Tesla uses in their vehicles.
The new sensors support a much higher resolution of 2896x1876. That's 5.4 megapixels compared to the 1.2 megapixels that Tesla's current front-facing cameras support.

At more than four times the resolution of current sensors, these new sensors will allow Tesla to more accurately detect objects that are required for FSD.

Detecting road edges and lane markings that are further away are especially difficult. Due to the angle of the road and the current image resolution, Tesla often has to make assumptions based on just a few pixels.

The increased resolution will offer a more detailed image and should improve the accuracy in these situations.

The new wide-angle sensor may also allow Tesla to drop the number of cameras used in the front-facing module. Due to the higher resolution in these sensors, Tesla may be able to use this wide-angle sensor to interpret objects that previously required the use of the standard or telephoto lenses before. This may allow Tesla to reduce the number of cameras that are required in the front of the vehicle from three down to two or even a single lens.

The new camera sensor also supports HDR shooting at 10 bit and 40 FPS, offering better contrast and richer colors. It's not clear whether Tesla would take advantage of HDR as it would require more power to process, but it's something Tesla may consider depending on the compute power available.

LED flicker mitigation

More importantly, the new sensor will also support LED flicker mitigation. Camera sensors often have a hard time capturing the light emitted from LED lights in a constant fashion. Since LEDs are widely used in electronic signs and traffic lights this will be an important feature and should make it easier for Tesla to interpret such signs.

Tesla is working closely with Samsung to produce their next-generation chips that will be used in FSD hardware 4.0. Tesla will likely launch updated cameras at the same time they introduce a new FSD computer.

It makes sense for Tesla to introduce this new hardware in a new model first before retrofitting new camera modules into existing models.

https://www.notateslaapp.com/news/679/tesla-s-fsd-hardware-4-0-to-use-new-cameras#:~:text=According to Chris Zheng on,their upcoming hardware 4.0 revision.&text=This new sensor offers many,Tesla uses in their vehicles.

Note: Not mentioned in this article is Samsung will also provide cameras for the Cybertruck.

I will not buy FSD in it's current configuration. I think Munro knows a bit or 2 more than most of us can comprehend & more than I do!

 

[Zapharus]

FSD is kinda hot garbage rn tbh

Summon is klunky, autopark is slower than my grandma, phantom braking is real. The beta is much better, and is advancing quickly, but...it's just a beta out to a few thousand people.

Exactly! People love to inflate FSD's current capabilities knowing damn well is not living up to expectations based on the many delivery timelines spewed by Elon. It's pretty dishonest because anyone who's not blind by fanboyism and has used FSD knows it still needs a lot of work to be labeled as being safer than a human driver.

 

[Ogre]

Exactly! People love to inflate FSD's current capabilities knowing damn well is not living up to expectations based on the many delivery timelines spewed by Elon. It's pretty dishonest because anyone who's not blind by fanboyism and has used FSD knows it still needs a lot of work to be labeled as being safer than a human driver.

FSD is a giant contradiction. It is both an absolutely amazing technology which has the potential to transform society and transportation; and fallen way short of the expectations set early on.

I think it’s awesome, but that’s easy for me because I kind of had the lowdown on it’s state when I bought my car and resisted buying it. If I’d bought it for $7,000 when I bought my Model Y, I might feel otherwise.

 

[Crissa]

That we're even talking about FSD like it's just something you can buy off the shelf is the fanboyism.

Entire companies have lived and died and spent billions of dollars on a handful of vehicles and we're arguing about buying it for $7 grand.

-Crissa

 

[tidmutt]

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

No doubt, what a treasure trove of information we’ll be able to mine.

It can’t be a prerequisite for FSD though.

 

[anionic1]

This just reminds me of when computers were picking up steam in the 90s. Nothing was compatible or streamline and things crashed and really it was a rough process working with software. But now you can buy some random Bluetooth device on Amazon and it very likely will just sync with your computer upload the software itself and it just works. Or better yet, when Siri or Alexa first started. 75% of the time they got it wrong and it was more frustrating than its worth, but now it seems like 90% of the time it gets it right and thats doable. In ten years FSD will very likely be to the point where it’s almost flawless. And it will need to be and society needs to just be able to work through the struggle to get there. I really don’t see any other way.

 

[Ogre]

This just reminds me of when computers were picking up steam in the 90s. Nothing was compatible or streamline and things crashed and really it was a rough process working with software. But now you can buy some random Bluetooth device on Amazon and it very likely will just sync with your computer upload the software itself and it just works. Or better yet, when Siri or Alexa first started. 75% of the time they got it wrong and it was more frustrating than its worth, but now it seems like 90% of the time it gets it right and thats doable. In ten years FSD will very likely be to the point where it’s almost flawless. And it will need to be and society needs to just be able to work through the struggle to get there. I really don’t see any other way.

More like: Buy some random thing off of Amazon, it requires you sign up for some special online account, phones home, gets infected with malware you don’t know about for 2 years until it finally breaks. Meanwhile your house is part of a swarm of crappy mostly unused IOT junk which takes part in regular DDOS attacks and malware infections.

 

[anionic1]

I was also thinking about the brilliance of tesla to look at human brain processes to design and mimic with their software and hardware. Our vision, mental and reaction processes have evolved over millions of years in very challenging circumstances. If any FSD system has a chance at truly succeeding I think the FSD system tesla is working on has the best chance because of exactly this.

 

[anionic1]

More like: Buy some random thing off of Amazon, it requires you sign up for some special online account, phones home, gets infected with malware you don’t know about for 2 years until it finally breaks. Meanwhile your house is part of a swarm of crappy mostly unused IOT junk which takes part in regular DDOS attacks and malware infections.

But it worked when you tried to connect it and it was fun for at least a week.

 

[firsttruck]

Investing Against the Grain rebuts E for Electric hatchet job on Tesla FSD

Alex of E for Electric edited Sandy Munro statements to make it sound like Sandy was saying FSD software was crap when Sandy was talking about lidar/radar vs FLIR .

E for Electric Creates Tesla FUD / Sandy Munro is Wrong ! Ep 133
E for Electric creates FUD about Tesla and in my opinion abuses his relationship with Sandy Munro. Stick until the end to see how he abuses and lies about what Sandy Munro says. Sandy Munro believes that Tesla needs to use FLIR to solve FSD. I disagree!
Mar 23, 2022
Investing Against the Grain

-----