FSD discussion

[bkp_duke]

Also can someone explain to me why all of these cars that uses Mobile Eye like the Ford Mach E, or all the Lidars they put on EQS(L3 they claim), Lucid Air, or whoever can't seem to beat a Tesla at active safety braking when NCAP actually put them through a standardized test? So what Mobile Eye is hiding their best stuff behind closed doors while allowing the Mach E to hit bicyclists right now?

So far all the stuff in actual customer's hands are trash, while the parent companies are bragging about FSD everywhere next month. Only Tesla to my knowledge actually released real FSD to customers and charging for it. Everyone else seems to be hiding something.

I have heard a theory about this, and it has some good weight behind it.

All of these systems rely upon ECU modules to do the processing for them. This adds latency for those modules to talk with other systems.

There was a fantastic video about the wet traction on a BMW i4 vs. a Tesla M3 P. The Tesla had far far better wet traction than the BMW (faster 0-30, shorter stop time, you could literally see the i4 wheels spin and try to grip - nothing to immediate grip from the Tesla), simply because it could talk to the wheels much faster.

This is what happens when you rely upon 3rd party companies to build out most of your functionality.

 

[LPEREZDANCER]

FSD Beta is not a paid feature, and nobody is "entitled" to it from owning FSD.

You should have access to all the features included in the actual paid FSD package (which is- everything included in EAP, plus stop light and stop sign recognition). And eventually it'll include city streets once that's moved to the mainline FSD branch instead of the limited beta.

THAT said- your car can't do the beta because it needs updated cameras. Which you are entitled to get, for free... (you might also need a new HW3 computer if you still have the old one- wasn't clear from your post- which also would be free).

Once you get those you'd be able to get the beta, HW wise anyway.

Tesla offering Autopilot Camera Upgrades to owners with FSD capability

Tesla is offering Autopilot camera upgrades to FSD customers with high Safety Scores. The upgrades come just before FSD's wide release.

www.teslarati.com

Link about the upgrades for cameras and how to request them.

Thank you sir! I did have it in 2 weeks ago for the eMMC upgrade and upgraded cameras. Wondering if anyone with MCU1 is able to run FSD? Perhaps I need to start a new thread

 

[EVNow]

And TSLA doesn't?

Also I like how Tesla with the best AI engineers on self driving are not trying the method all these so called competitors who are claiming to have found the panaceia to true FSD.

This is the problem with not completely following the conversation.

My point was - we can't say with certainty that others will find Tesla's position insurmountable. MobilEye and companies like XPeng are not far behind (though it is difficult to judge) - atleast we can't make that assumption. Its not like Tesla will solve FSD but others will not be able to ...

BTW, I'm actually very interested in finding out where exactly other competitors are. I started a thread to track it (its even in my sig !). I've not updated in a while ... but the table isn't all that outdated.

Wiki - Consumer AV - Status Tracking Thread

This thread is to track the progress of Consumer AVs - from ADAS to full self-driving. I'll keep this info updated. Let me know if you have any updates & corrections.

teslamotorsclub.com

 

[bkp_duke]

This is the problem with not following the conversation.

My point was - we can't say with certainty that others will find Tesla's position insurmountable. MobilEye and companies like XPeng are not far behind (though it is difficult to judge) - atleast we can't make that assumption. Its not like Tesla will solve FSD but others will not be able to ...

Completely disagree, and I'm going to be detailed on why.

People glance at MobileEye's PR sheet and see glitzy words and think all the added sensors they are bringing to bear (LIDAR, I'm looking at you) means that they must be superior. But it's not really the sensors that are important, it's the DATA PROCESSING and logical heuristics that are key. Often called "sensor fusion". The problems with trying to prioritize different types of data streams (LIDAR vs. vision vs. Radar) and aggregating those signals is what I believe is the main reason Tesla dropped low-res radar, and refused to add lidar in to the mix. By focusing on just one data type, vision (i.e. images), the processing heuristics can be essentially universal, and an order of magnitude higher in quality (less garbage in, garbage out).

Everyone else is just tossing sensors at things, but their ability to manage those datastreams to date is very VERY poor (as shown by how bad everyone's ADAS system tests out compared to Tesla). If they can't get ADAS working better, why do we think they can get an FSD competitor working in a reasonable time-frame?

 

[EVNow]

People glance at MobileEye's PR sheet and see glitzy words and think all the added sensors they are bringing to bear (LIDAR, I'm looking at you) means that they must be superior.

I'm not sure you have seen much of MobilEye. I've spent hours listening to their detailed talks and we have had a lot of discussion about them in the AP forum. They are quite close to what Tesla is doing, infact (camera based).

ps : I didn't know they are partnering with Porsche now ...

https://twitter.com/x/status/1655891255887294464

 

[Singuy]

This is the problem with not completely following the conversation.

My point was - we can't say with certainty that others will find Tesla's position insurmountable. MobilEye and companies like XPeng are not far behind (though it is difficult to judge) - atleast we can't make that assumption. Its not like Tesla will solve FSD but others will not be able to ...

BTW, I'm actually very interested in finding out where exactly other competitors are. I started a thread to track it (its even in my sig !). I've not updated in a while ... but the table isn't all that outdated.

Wiki - Consumer AV - Status Tracking Thread

This thread is to track the progress of Consumer AVs - from ADAS to full self-driving. I'll keep this info updated. Let me know if you have any updates & corrections.

teslamotorsclub.com

What I find incredible is the amount of....misleading data? coming out of these companies. Cruise claimed 60k miles per intervention. So you are telling me all these cruise cars getting stuck almost every few days or running into a bus have driven 60k miles/car between each problem? So far even geofenced system is not all that impressive compared to FSDb as we see them taking easy routes, slams on the brakes, getting stuck, and giving passengers heart attacks as it swerve into pedestrians as it tries to avoid other pedestrians.

 

[EVNow]

Cruise claimed 60k miles per intervention.

60k miles between "accidents". Getting stuck, apparently isn't (I guess).

So far even geofenced system is not all that impressive compared to FSDb as we see them taking easy routes, slams on the brakes, getting stuck, and giving passengers heart attacks as it swerve into pedestrians as it tries to avoid other pedestrians.

Where do you see those ? Definitely something I'm sure @diplomat33 would be interested in.

There is a guy who has recorded and posted dozens of Waymo drives in Chandler ... don't remember seeing anything like that.

 

[bkp_duke]

I'm not sure you have seen much of MobilEye. I've spent hours listening to their detailed talks and we have had a lot of discussion about them in the AP forum. They are quite close to what Tesla is doing, infact (camera based).

ps : I didn't know they are partnering with Porsche now ...

https://twitter.com/x/status/1655891255887294464

I'm very familiar with them, have been following them closely since they were the provider for the hardware for Tesla's AP1.

I don't think their approach is going to work well, as it partially relies upon HD maps. The Achilles heel of HD maps is that they go out of date almost instantly (what happens when a car double-parks? That wasn't on the map). What happens if there is construction ongoing with lane shifts since the mapping?

No, Tesla's design is the only one that has the ability to massively scale.

 

[Singuy]

I'm not sure you have seen much of MobilEye. I've spent hours listening to their detailed talks and we have had a lot of discussion about them in the AP forum. They are quite close to what Tesla is doing, infact (camera based).

ps : I didn't know they are partnering with Porsche now ...

https://twitter.com/x/status/1655891255887294464

How about give mobileye to 400k users across 4 million miles of road and lets see how close they are.

 

[EVNow]

I don't think their approach is going to work well, as it partially relies upon HD maps. The Achilles heel of HD maps is that they go out of date almost instantly (what happens when a car double-parks? That wasn't on the map). What happens if there is construction ongoing with lane shifts since the mapping?

Actually their "hd maps" are crowdsourced and something Tesla should do as well. As a well liked Tesla analyst put it - its low hanging fruit.

https://twitter.com/x/status/1643032819336183811

 

[EVNow]

How about give mobileye to 400k users across 4 million miles of road and lets see how close they are.

We obviously can't know that ... as you well know, in the market perfect data is hard to find

We also don't have independent videos of MobilEye driving - though there are a lot of "raw" videos they have put out.

The point is not to say X is more advanced than Y - but it is to acknowledge that Tesla's may not have an unsurmountable lead, after all.

 

[Singuy]

60k miles between "accidents". Getting stuck, apparently isn't (I guess).


Where do you see those ? Definitely something I'm sure @diplomat33 would be interested in.

There is a guy who has recorded and posted dozens of Waymo drives in Chandler ... don't remember seeing anything like that.

It's per disengagement, not accident.

 

[bkp_duke]

Actually their "hd maps" are crowdsourced and something Tesla should do as well. As a well liked Tesla analyst put it - its low hanging fruit.

https://twitter.com/x/status/1643032819336183811

No no no. HD mapping is a logical "dead end" for this kind of product.

Humans don't memorize detailed maps. We think along the way, THAT is the hardest part. Not the mapping. The mapping is chump change and buzz words for PR people.

 

[Singuy]

We obviously can't know that ... as you well know, in the market perfect data is hard to find

We also don't have independent videos of MobilEye driving - though there are a lot of "raw" videos they have put out.

The point is not to say X is more advanced than Y - but it is to acknowledge that Tesla's may not have an unsurmountable lead, after all.

Then why are people making claims using only marketing video from mobile eye? "They are very close".....at what exactly without anyone having access to it let alone almost half a million people?

 

[EVNow]

No no no. HD mapping is a logical "dead end" for this kind of product.

Humans don't memorize detailed maps. We think along the way, THAT is the hardest part. Not the mapping. The mapping is chump change and buzz words for PR people.

Separate discussion - but both I & Tesla disagree with you. Lets not use the word "HD map" - since it can mean a lot of different things. But adding a lot of useful metadata to the map (and maintain it by crowdsourcing) is absolutely doable and scalable.

Tesla is now using more metadta with maps that it downloads when you select a route per @verygreen.

 

[EVNow]

Then why are people making claims using only marketing video from mobile eye? "They are very close".....at what exactly without anyone having access to it let alone almost half a million people?

If you don't know what their capability is - why are you claiming they are inferior. That is what I'm pointing out ...

ps : MobilEye is obviously different from some other fly-by-night startup. They need to be taken seriously.

 

[Singuy]

If you don't know what their capability is - why are you claiming they are inferior. That is what I'm pointing out ...

ps : MobilEye is obviously different from some other fly-by-night startup. They need to be taken seriously.

No, the point is now that the tech is in the hands of almost half a million people, we find almost an infinite amount of problems Tesla needs to solve which will take awhile and it's frustrating. However they already have the data set and are building models toward that goal. Mobile Eye haven't even started yet. The case that was being made is no one will be able to skip this long tail end grind, and no one have even started attacking the problem besides Tesla.

 

[bkp_duke]

No, the point is now that the tech is in the hands of almost half a million people, we find almost an infinite amount of problems Tesla needs to solve which will take awhile and it's frustrating. However they already have the data set and are building models toward that goal. Mobile Eye haven't even started yet. The case that was being made is no one will be able to skip this long tail end grind, and no one have even started attacking the problem besides Tesla.

BINGO - it's a data accumulation and computational problem. The data accumulation has taken 5+ years for Tesla, and the dataset has been filtered repeatedly to make it high-quality (human labelers, auto labelers, etc.). All of these tools still have to be built by the competition, that's what everyone (even here) is completely ignoring.

 

[diplomat33]

60k miles between "accidents". Getting stuck, apparently isn't (I guess).

Where do you see those ? Definitely something I'm sure @diplomat33 would be interested in.

There is a guy who has recorded and posted dozens of Waymo drives in Chandler ... don't remember seeing anything like that.

Certainly, there have been some stalls with Cruise and Waymo that got a lot of media attention but they are rare compared to the total driverless miles. To argue that the stalls are happening all the time and that the rides are terrible with constant phantom braking is complete hyperbole IMO. Waymo is doing 10k driverless rides per week. I think we would know if the rides were constantly bad all the time. AFAIK, the overwhelming majority of those rides are very smooth. Furthermore, according to the CPUC, Cruise had only 5 collisions while driverless since June 2022 and none of the collisions involved any injuries and Waymo has had zero collisions while driverless that resulted in injuries. As a result, the CPUC has said that Cruise and Waymo are reasonably safe and can deploy driverless 24/7 in all of SF. Does anyone seriously think that FSD Beta could do 10k driverless rides per week in SF or downtown Phoenix with only a dozen or so stalls per year and with no collisions resulting in injury? I doubt it.

But if you don't believe me, ask Maya. She is a regular Waymo rider who posts a lot of videos of her driverless rides in SF. She can testify to the quality of her rides. Here is a link to one of her videos where she responds to the naysayers. But you can check out her other videos. She has lots of videos on her youtube channel.

 

[bkp_duke]

Certainly, there have been some stalls with Cruise and Waymo that got a lot of media attention but they are rare compared to the total driverless miles. To argue that the stalls are happening all the time and that the rides are terrible with constant phantom braking is complete hyperbole IMO. Waymo is doing 10k driverless rides per week. I think we would know if the rides were constantly bad all the time. AFAIK, the overwhelming majority of those rides are very smooth. Furthermore, according to the CPUC, Cruise had only 5 collisions while driverless since June 2022 and none of the collisions involved any injuries and Waymo has had zero collisions while driverless that resulted in injuries. As a result, the CPUC has said that Cruise and Waymo are reasonably safe and can deploy driverless 24/7 in all of SF. Does anyone seriously think that FSD Beta could do 10k driverless rides per week in SF or downtown Phoenix with only a dozen or so stalls per year and with no collisions resulting in injury? I doubt it.

But if you don't believe me, ask Maya. She is a regular Waymo rider who posts a lot of videos of her driverless rides in SF. She can testify to the quality of her rides. Here is a link to one of her videos where she responds to the naysayers. But you can check out her other videos. She has lots of videos on her youtube channel.

The problem with Cruise and Waymo are the inability for these solutions to SCALE UP in any reasonable time-frame.

Specifically:
1) they are geofenced to TINY TINY areas
2) they require cars with thousands, sometimes tens of thousands of dollars of extra equipment
3) until recently Cruise could only operate in low-traffic environments (and both still completely avoid highway driving) - tell-tale sign of "cherry picking" the variables under which your system has the best chance for functionality (can't blame them, it's reasonable - but it is telling)

I did an intervention-free, door-to-door drive in FSDb 11.3.6 this weekend from Orange County, CA to north San Diego County (~70 miles). It involved city streets with heavy traffic, entering and exiting the freeway, and traversing multiple construction zones. That's a first for me with FSDb, but I'm seeing marked improvements in a system that can generally handle all driving conditions (not specific ones that are easier). I also find that most disengagements currently are not safety-related, but preference related (i.e. I take over because I'm an aggressive driver and would prefer to be in the left lane going faster passing people, etc.).