Wiki FSD Beta 10.4

[TravelFree]

Making left and right turns from a traffic light was still erratic and in some cases caused the abort as it wanted to turn into oncoming traffic-- That was real scary.

Happened to me too once. I understand the first reported FSD-Beta crash happened due to this very failure of FSD Beta.

Now that we have the first accident using FSD Beta, I see Tesla has pushed 10.5 back for another 10 days. It was bound to happen. I hope the Tesla haters in the regulatory agencies don't ban us from further use of FSD beta testing. Even worse, that one new Tesla hater wants to ban Auto Pilot because Tesla refused to use her favored LIDAR from her former employer.

 

[Gotchaa]

Driving around in LA and I can tell you I am not surprised. I had 3 interventions on left and right turns today that would have resulted in accidents if cars were around. Right turns where there is bike lane is a problem, AP gets confused. Turning onto a street just after a train track crossing almost resulted in turning onto the track. And going down a street with cars parked on both sides with oncoming traffic almost resulted in a side swipe. This is alpha. I hope the snapshots are helping. My current setting is now chill mode, with speed set to -5mph of the speed limit. I rather manually increase the speed as I don’t trust it. AP forces you to be more attentive and is stressful. Bright spots are there too. Generally good at 4 way stops, going to try round-about tomorrow, left hand turns are better than expected. It sucks at turning on a grade, usually need to take it at an angle or it will scrape the bottom. Lastly it cannot see a folding barricade with yellow reflective light, tried to turn left directly into it.

I think it’s going to take a long time for this to be safe for general release. I don’t see it happening anytime soon, there is just too many variances in street markings and signage. Tesla should start working on lobbying for policies to standardize street markings/signage for the benefit of all AP systems. We are many years away from this being street safe IMHO.

 

[AndreP]

Driving around in LA and I can tell you I am not surprised. I had 3 interventions on left and right turns today that would have resulted in accidents if cars were around. Right turns where there is bike lane is a problem, AP gets confused. Turning onto a street just after a train track crossing almost resulted in turning onto the track. And going down a street with cars parked on both sides with oncoming traffic almost resulted in a side swipe. This is alpha. I hope the snapshots are helping. My current setting is now chill mode, with speed set to -5mph of the speed limit. I rather manually increase the speed as I don’t trust it. AP forces you to be more attentive and is stressful. Bright spots are there too. Generally good at 4 way stops, going to try round-about tomorrow, left hand turns are better than expected. It sucks at turning on a grade, usually need to take it at an angle or it will scrape the bottom. Lastly it cannot see a folding barricade with yellow reflective light, tried to turn left directly into it.

I think it’s going to take a long time for this to be safe for general release. I don’t see it happening anytime soon, there is just too many variances in street markings and signage. Tesla should start working on lobbying for policies to standardize street markings/signage for the benefit of all AP systems. We are many years away from this being street safe IMHO.

Who pays for standardizing street markings/signage, especially when the intent is to move away from gasoline/diesel and the tax revenues they generate for municipalities etc?

There have been all sorts of ideas bounced around within this space in terms of making roads friendlier for autonomous vehicles, they tend to fall flat when juxtaposed against financial realities. Even just ensuring roads are properly painted/marked all year round would go a long way, and that one step comes with huge cost implications.

Dig into this further and you'll find that places are asking private companies to help with this, they think Tesla etc should help foot the bills.

 

[TravelFree]

I watched the FSD beta videos for months but it didn't dawn on me until I was behind the wheel witnessing for myself, how difficult autonomy will be.

I boiled it down to two basic problems.

1. Some roads are designed that do not follow a standard. Almost like the lowest level construction worker decides how to make them as they are being built.

2. People drivers do not follow rules of the road.

There is no such thing as artificial intelligence. What the geniuses are building is a super calculator that follows a very large collection of rules.
Until artificial intelligence can think and react on the spot the way humans can, deciding how to deal with something new and strange or just deciding when to break the rules for personal convenience, other drivers be damned, we will never get to full Level 5 autonomy and the Robo Taxi idea is just another idea like perpetual motion machine. I believe getting to Level 3 may be possible if we have special roads that are only accessed by cars using Level 3 autonomy.

The closest to Level 5 autonomy and RoboTaxi will just be a human driver chauffeur on our payroll. I hope someone can prove me wrong with real evidence. And, don't just say Dojo because what they describe is just another bigger calculator.

 

[BMW740iL]

The closest to Level 5 autonomy and RoboTaxi will just be a human driver chauffeur on our payroll. I hope someone can prove me wrong with real evidence. And, don't just say Dojo because what they describe is just another bigger calculator.

 

[Gotchaa]

Who pays for standardizing street markings/signage, especially when the intent is to move away from gasoline/diesel and the tax revenues they generate for municipalities etc?

There have been all sorts of ideas bounced around within this space in terms of making roads friendlier for autonomous vehicles, they tend to fall flat when juxtaposed against financial realities. Even just ensuring roads are properly painted/marked all year round would go a long way, and that one step comes with huge cost implications.

Dig into this further and you'll find that places are asking private companies to help with this, they think Tesla etc should help foot the bills.

It should be a shared cost, we already pay for highways, local municipalities are where the challenge is. I would say, after 2025, any ICE car sold should have an added Tax imposed to pay for it as well a tax for any ICE car registered. I can see these funds going towards infrastructure improvements while at the same time helping accelerate EV adoption. I believe it should happen at the state level. I can see CA doing this first as usual.

Looks like we are already going to be paying for some progress in this realm at the federal level with the new infrastructure bill. Here are some relevant sections in the bill the President is signing next week:

Sec. 11504. Study of impacts on roads from self-driving vehicles



Sec. 24208. Crash avoidance technology.

Sec. 24209. Reduction of driver distraction.



Sec. 24219. Research on connected vehicle technology.

Sec. 24220. Advanced impaired driving technology.



Sec. 25006. Electric vehicle working group

Long ways away from any significant changes happening, hopefully tech improves enough to allow for minor road enhancements to have an impact on AP performance …

 

[AndreP]

It should be a shared cost, we already pay for highways, local municipalities are where the challenge is. I would say, after 2025, any ICE car sold should have an added Tax imposed to pay for it as well a tax for any ICE car registered. I can see these funds going towards infrastructure improvements while at the same time helping accelerate EV adoption. I believe it should happen at the state level. I can see CA doing this first as usual.

I don't think expecting tax revenues for infrastructure improvement/maintenance from a declining source makes a lot of sense but we're not in the positions of people making these decisions, I just know what I've seen and heard discussed at autonomous vehicle conferences. These are not new ideas, the concept of road improvements to improve autonomous vehicle performance have existed for as long as the concept of autonomous vehicles.

I'll sooner expect EV- and Autonomy-specific taxes if private companies won't step up, those will be the long-term revenue generators if we're actually heading in that direction

 

[MrPinrel]

My safety score just got to 98. Does anybody know if I will get the FSD now (even if I didn’t have 98 last Friday or whenever) or do I have to wait for the next wave of new users?

 

[Gotchaa]

I don't think expecting tax revenues for infrastructure improvement/maintenance from a declining source makes a lot of sense but we're not in the positions of people making these decisions, I just know what I've seen and heard discussed at autonomous vehicle conferences. These are not new ideas, the concept of road improvements to improve autonomous vehicle performance have existed for as long as the concept of autonomous vehicles.

I'll sooner expect EV- and Autonomy-specific taxes if private companies won't step up, those will be the long-term revenue generators if we're actually heading in that direction

Based on latest bill, it seems like they are taking it more seriously now given higher EV adoption and goal towards zero Carbon emission. The only way this happen is by taxing, through EV purchases and registrations as part of the shared cost for improvements. I can also see a portion of higher business taxes for larger companies with carbon emitting businesses over a certain level in a given local municipality going towards these improvements…shared costs

 

[AndreP]

Based on latest bill, it seems like they are taking it more seriously now given higher EV adoption and goal towards zero Carbon emission. The only way this happen is by taxing, through EV purchases and registrations as part of the shared cost for improvements. I can also see a portion of higher business taxes for larger companies with carbon emitting businesses over a certain level in a given local municipality going towards these improvements…shared costs

Revenue from carbon taxes is used to curb emissions though, they won't take carbon tax revenue and use it to improve roads to help autonomous vehicles. Road maintenance has historically been funded by taxes on fuel and it only makes sense, because road use scaled with fuel consumption. More fuel usage = more miles driven = more road maintenance required and more funds generated to support that maintenance.

Similarly, autonomous vehicle usage is what will dictate the need for road improvements to help autonomous vehicles function.

 

[aronth5]

Who pays for standardizing street markings/signage, especially when the intent is to move away from gasoline/diesel and the tax revenues they generate for municipalities etc?

Good question but 18 states have already added extra annual fees to EV owners and I suspect that number will certainly increase. How close that new revenue source will off set the loss of gasoline/diesel tax revenues I don't know but appears states are well aware of the revenue shortfall.

 

[Reddy Kilowatt]

My safety score just got to 98. Does anybody know if I will get the FSD now (even if I didn’t have 98 last Friday or whenever) or do I have to wait for the next wave of new users?

There is an article out this morning with a tweet from Elon stating 98 is delayed ten days and they will stat pushing 98's with the 10.5 release. I was a 99 and waited a while and just got Beta this week.

 

[tmoz]

Once collisions are reduced by 90% +, a lot of jobs are going to be lost, such as physical therapists, injury lawyers, doctors of all sorts, body shops, etc, etc.
The cost to society of car accidents is a very big number, 100s of billions I believe. People wont be dying from car injuries, lives wont be ruined by automobile accidents. This might be 50 or 100 yrs from now, or 5 or 10. I don't know, but it can only get better. These are first baby steps towards that goal. FSD Beta testers are part of the progress towards that goal, even if the current approach doesn't work, somebody has to show that it doesn't work and another approach is needed - keep up the good work guys!

 

[boiler81]

I think folks who want FSD beta and don't have the safety scores needed to get it are in for a rude awaking.
Being a beta tester is hard work and the testers have to be very vigilant when using beta because it can and does make mistakes at the worst times.

However, it is improving presumably with the help of the beta tester input.
If you are willing to drive for the "good of the technology development", you in time will be rewarded, but if you looking for something useful today, don't be a beta tester.

 

[EVNow]

But yeah... lane changes in 10.4 are darn near humanly good.

FSD and AP before that has been very good at changing lanes. Infact its been good enough that I always use it to change lanes.

 

[EVNow]

People wont be dying from car injuries, lives wont be ruined by automobile accidents.

Among teenagers, accidents are #1 cause of death.

https://www.cdc.gov/nchs/data/databriefs/db37.pdf

Accidents account for nearly one-half of all teenage deaths. As a category of accidents, motor vehicle fatality is the leading cause of death to teenagers, representing over one-third of all deaths.

 

[mark95476]

Yup. I always check my side mirrors before signaling and FSD is there to verify the coast the clear.

FSD and AP before that has been very good at changing lanes. Infact its been good enough that I always use it to change lanes.

 

[lolder]

FSD 10.4 is my first beta as I just got my SS to 100 a week ago. Long time FSD beta tester " Dirty Tesla" in Mich. wondered if 10.4 was broken it's so bad. It can't make simple decisions. Beta testers were supposed to be ferreting out the dragons tail. Had a relaxing drive today on EAP/NOA.

 

[EVNow]

Long time FSD beta tester " Dirty Tesla" in Mich. wondered if 10.4 was broken it's so bad.

But other long time beta testers (Frenchie) say they had "ALMOST Perfect" drives with 10.4.

I don't think 10.4 is any worse than other builds - just that individual drives & weather etc change.

 

[nvx1977]

But other long time beta testers (Frenchie) say they had "ALMOST Perfect" drives with 10.4.

I don't think 10.4 is any worse than other builds - just that individual drives & weather etc change.

Even during AP/NoA days, we had a wide range of reports on how well it works. I think a lot of it is just our subjective tolerances and how much we appreciate the tech's progress. Some people might think steering wheel jitter is NBD while others find it completely intolerable. Some might judge FSD based on how similar it is to how they personally drive. Same behavior between cars, wide range of opinions.