Tesla’s FSD Beta 10.3 Coming This Friday 10-22-2021

[S4WRXTTCS]

Well, so far I've been able to usually get it to drive me from home to work and back (about 35-40 miles each way), and also from home to a friends place about 45 minutes away and back home again- without interventions. There's some caveats there of course.


1) In all those cases the vast majority of the drive is highway...probably at least 80-90% of it.... so really it's more "can drive me from driveway to highway" or "highway to work" or "highway to friends house" and then reverse of them.... with no on-actual-city-streets portion being more than maybe 10 minutes of driving tops.

2) It does brake quite often for little apparent reason, but it's always quite minor.... most commonly 2-3 mph slowdowns on corners or hills where I guess it can't be absolutely sure as far ahead as it'd like.

3) It seems to often decide a road I'm turning onto has a WAY lower speed limit than it really does so I find myself having to dial that up- this is not technically a disengagement but it would be if that were the only way to make it not be doing 25 mph on a 55 mph road.

4) A couple of times it's been REALLY slow to go through a turn... it would eventually do it... and it'd do the same turn another time fine... but that probably WOULD have been a disengagement if there'd been cars behind me as I wouldn't have wanted to make them wait through that.

5) The one and only real "this was legit wrong and dangerous" disengagement I did have was coming around a tightish corner up to a line of maybe 6 or 7 cars lined up stopped at a red light (single lane each direction road).... FSDBeta decided these guys were stopped for no reason and began to go into the oncoming lane to pass them.

6) It waits too long to get into the correct lane for upcoming turns. Since most of my driving is in light/little/no traffic this hasn't big a big issue, but it would be in denser driving situations.

7) And lastly as I think I've mentioned- having seen the creeping behavior with low side visibility, I don't believe they can achieve better than L2 on city streets without at least 1 additional per side forward-mounted side looking cams.... that wouldn't prevent them from getting as high as L4 on highways though.

I haven't been able to have it drive me from home to work as there are real fundamental issues with it. The entirety of my drive to work is 9 miles through suburbia.

1.) There is a blind corner before a stop light where it tries to maintain the set speed of 35mph through the corner so I either have to dial the set speed way down or take over before reaching the corner. This is where I've seen other people have to panic stop because they simply came in too fast, and didn't realize cars were backed up waiting for the light. If it wasn't a blind corner the 35mph speed would be perfectly fine. This issue I reported.

2.) The over braking mostly matches your experience in terms of the slow down. But, it's been doing it even when there was no obvious reason.

3.) It tends to be a bit twitchy. Like the steering will suddenly jerk slight which is really unnerving.

4.) It hasn't handled a single left turn yet. Now partly this is simply from me not being comfortable with it. It doesn't seem to want to do a nice smooth turn, but breaks it up into pieces which is really unnerving.

5.) I haven't been able to go from work to home because the autobrights keep getting in the way so I turn it off during times where I know it will annoy other drivers. This part is sucky as I really want to test it at night with less traffic, and the navigation has the correct path to take to home. The path to work it chooses a road where I'd have to take an unprotected left over multiple lanes of fast traffic.

6.) It's gotten way too close for comfort to curbs during some corners so I've taken over each time as I don't want to damage my rims.

7.) I wish it would slow down earlier for traffic stopped in front of red light. For some reason today it wanted to wait to last minute to go from 45-0 at a spot where I'd already be at 20 using re-gen to fully slow me down. I did report this one because it was noticeably braking later than typical.

8.) I find it annoying that it doesn't always turn the turn signal on BEFORE getting into a lane to turn left.

9.) It's too slow to really use without taking over in spots. Like turning a right out my residential zone.

One note is that it's basically been raining none stop since I got the beta so its definitely getting tested in typical Seattle area weather.

With exception to the phantom braking I do enjoy it as long as there are no turns, and I like how its handled the 4+ way stops so far.

 

[Terminator857]

It's not ready so even if the coast was clear to release right now they wouldn't be able to. ...

What do you mean? Sure they can in Arizona, Florida, and elsewhere.

 

[Knightshade]

I did some testing on rural roads with stop signs that seem to not be mapped. These were at rail road crossings and the signs were in good condition it was dark with high beams active. The car has always braked really late for these and I have slowed the car down from 55 to ~40 when approaching and it has been able to stop although very hard. I decided to see if I left it at 55-60 mph what would happen. The car wasn't going to stop in time, so I setup some safety checks (used a stop that had good visibility, where I live so I know the time trains run, had a lookout just to be extra safe that no trains or other cars were around) and decided to let it play out and let the car fully do its thing. As I expected the car with fsd beta driving ran the stop sign. It tried to slam on the brakes and sounded an alarm at the last second as we blew through the stop sign. I took the car back to the stop sign but this time paid more attention to the display (I didn't let it run it this time) it seems (based on what the display showed) that the car just can't see far enough forward (unless it can see more than displayed) to be able to react in time to unmapped stuff which is super worrying to me. By the time the forward arc of vision got to where the stop sign was and displayed it, it would be too late for it to make the stop at that speed. So I am worried about vision being able to handle this type of thing just based on what I can see from the display. This also makes me wonder about if a person was standing in the road instead of a stop sign (they may have another braking algorithm for that but IDK). This is also a word of caution to everyone that if you are driving in a rural area and there are stop signs that may not be mapped just know this can happen (although it is clear that its either not going to stop or it will be a very uncomfortable stop so a driver would already have started applying the brakes typically). I did send a clip of this to the fsd team with an email.

When you say the sign had good visiblity- how good are you talking?

Like it was visible from 500 feet away... or like it was visible 50 feet away cresting a hill?

In the second case there's not really any solution to this...If it's an unmapped stop sign right over a hill for example, a human wouldn't know it was there till it was visible either.

Otherwise the cameras should easily be able to do this....the cameras can see 450-750 feet ahead (depending which of the 3 cams you're talking about.

At 55 mph the car can stop to 0 in roughly 90-120 feet (depending on tires) if it has to.

 

[flyhighboi20]

When you say the sign had good visiblity- how good are you talking?

Like it was visible from 500 feet away... or like it was visible 50 feet away cresting a hill?

In the second case there's not really any solution to this...If it's an unmapped stop sign right over a hill for example, a human wouldn't know it was there till it was visible either.

Otherwise the cameras should easily be able to do this....the cameras can see 450-750 feet ahead (depending which of the 3 cams you're talking about.

At 55 mph the car can stop to 0 in roughly 90-120 feet (depending on tires) if it has to.

I am going to delete my original post here as I posted it in the experiences thread which is probably more the correct thread I should of put it in but I wanted to answer you first. It was flat ground leading up to the stop sign and the sign is reflective and in good condition, not occluded by any branches or other signs in front of it. There is a rail crossing X above the stop sign but it is separated with a height gap between them and it does display the stop sign. It doesn't even start to hit the brakes until very late at 40mph (this is tested on at least 3 different crossings) but it does stop but is a hard stop that tosses stuff thats loose in the car. At actual travel speed for that road it runs the stop (I had tested this once before months ago before having beta and it did the same thing)

 

[S4WRXTTCS]

What do you mean? Sure they can in Arizona, Florida, and elsewhere.

There are two parts to being able to release a "In most any situation" L4 system.

One part is internal acceptance of the liability involved versus the potential revenue. It doesn't do much good to release a system where the liability far exceeds any revenue from it. This is why there are no L4 systems that fit this description released anywhere in the world that I know of. There are extremely geofenced systems from Waymo, Cruise, etc that are essentially autonomous driving on a leash meaning that they have to call home.

Then there is external acceptance like regulatory, suppliers, customer insurance, etc. I simply can't see taking any chances on releasing an L4 system like this in any state without a nod of approval from the NHTSA. If you don't do that there is a chance they can put a stop to it, and force you to disable the system in the installed user base. This essentially happened with AP1 after a single fatality accident happened in Florida (of course) back in 2016. Basically the NHTSA blamed Tesla for releasing a system that allowed the driver to become complacent. To satisfy the NHTSA demands Tesla added a nag in a SW update thereby making it less user friendly. Not only did they have to add the nag, but MobileEye (which supplied the ADAS system at the time) wasn't happy with Tesla because they claimed the system wasn't meant to be use the way Tesla did. This is a critical aspect because Tesla lost a lot time having to reinvent things MobileEye already had when the two companies split ways. On the plus side they became more vertically integrated, and they're one of the only manufactures that I know of that have their own ADAS system top to bottom (maybe Subaru does too?). It's why their able to still ship their system while there is no supercruise due to the IC shortage for example.

I put customer insurance in there as well because with L4 the situation is a bit murky. In most situation means the user can and will still drive on occasion. So they need to have insurance, but how do you handle all the "who's in control" question? In my experience so far with FSD Beta a fair number of take over events is out of personal preference. Things like "lets not run over that protrusion in the road where the manhole cover is". I think Tesla insurance will simplify this as it won't matter how the switch happened. Or if the L4 system left the driver in a bad spot on the side of the road.

 

[Terminator857]

Waymo didn't get approval from NHTSA for their system.

 

[Daniel in SD]

Waymo didn't get approval from NHTSA for their system.

Which makes sense if you think about L3-L5 systems as being automated drivers. Right now the states regulate driver licensing and insurance requirements.
It seems likely that liability for collisions will ensure that no company will release a L3-L5 system less safe than the average human anyway. I'm not sure why the NHTSA will ever have to make regulations. L2 driver assist system regulation seems almost certain though.

 

[TonySellitto]

I haven't been able to have it drive me from home to work as there are real fundamental issues with it. The entirety of my drive to work is 9 miles through suburbia.

1.) There is a blind corner before a stop light where it tries to maintain the set speed of 35mph through the corner so I either have to dial the set speed way down or take over before reaching the corner. This is where I've seen other people have to panic stop because they simply came in too fast, and didn't realize cars were backed up waiting for the light. If it wasn't a blind corner the 35mph speed would be perfectly fine. This issue I reported.

2.) The over braking mostly matches your experience in terms of the slow down. But, it's been doing it even when there was no obvious reason.

3.) It tends to be a bit twitchy. Like the steering will suddenly jerk slight which is really unnerving.

4.) It hasn't handled a single left turn yet. Now partly this is simply from me not being comfortable with it. It doesn't seem to want to do a nice smooth turn, but breaks it up into pieces which is really unnerving.

5.) I haven't been able to go from work to home because the autobrights keep getting in the way so I turn it off during times where I know it will annoy other drivers. This part is sucky as I really want to test it at night with less traffic, and the navigation has the correct path to take to home. The path to work it chooses a road where I'd have to take an unprotected left over multiple lanes of fast traffic.

6.) It's gotten way too close for comfort to curbs during some corners so I've taken over each time as I don't want to damage my rims.

7.) I wish it would slow down earlier for traffic stopped in front of red light. For some reason today it wanted to wait to last minute to go from 45-0 at a spot where I'd already be at 20 using re-gen to fully slow me down. I did report this one because it was noticeably braking later than typical.

8.) I find it annoying that it doesn't always turn the turn signal on BEFORE getting into a lane to turn left.

9.) It's too slow to really use without taking over in spots. Like turning a right out my residential zone.

One note is that it's basically been raining none stop since I got the beta so its definitely getting tested in typical Seattle area weather.

With exception to the phantom braking I do enjoy it as long as there are no turns, and I like how its handled the 4+ way stops so far.

#8. I would always put my blinker on to signal that I’m entering the left turning lane. Not Beta.

#10 It tailgates!

 

[Terminator857]

... L2 driver assist system regulation seems almost certain though.

Interested in your vision on this one. What do you think may happen or should happen?
From my perspective I don't like the fleecing of customers done by Tesla with no full self driving. I'm not interested so much in regulation, but fixing the false advertising and false naming.

 

[EVNow]

I'm not interested so much in regulation, but fixing the false advertising and false naming.

How about changing "zero emission" for hybrids

 

[Terminator857]

How about changing "zero emission" for hybrids

Haven't seen that claim.

 

[Daniel in SD]

Interested in your vision on this one. What do you think may happen or should happen?
From my perspective I don't like the fleecing of customers done by Tesla with no full self driving. I'm not interested so much in regulation, but fixing the false advertising and false naming.

I think the NHTSA should require very detailed collision reporting from manufacturers so they can figure out if these systems improve safety. I have no idea what kind of regulations they should implement, I was just making a prediction that there will be regulation.
False advertising is a whole different issue in my mind unless it encourages people to overestimate the capabilities of L2 systems. FSD isn't really false naming since the design intent is full self-driving, it's just still in beta. They should make it more clear that they're not actually promising that it will ever be out of beta.

 

[Terminator857]

I think the NHTSA should require very detailed collision reporting from manufacturers so they can figure out if these systems improve safety.

Agree and I believe the NHTSA has required Tesla and other car manufacturers to do this. Theory: the reason why Tesla all of a sudden wants the documents to be secret, is that they don't want people to see the large number of accidents that occur while under autopilot, although perhaps that can be calculated from their safety report. Although it might make headlines: Tesla driving with autopilot has an accident every day or every x hours.

 

[Terminator857]

A partial zero emission vehicle, in the United States, is an automobile that has zero evaporative emissions from its fuel system, has a 15-year (or at least 150,000-mile) warranty on its emission-control components, and meets SULEV tailpipe-emission standards. Source: Partial zero-emissions vehicle - Wikipedia

Read about AT-PZEV on that page, too. Of course, anything that burns petroleum is not "zero emission," but compared to non PZEV vehicles, they are hellacleaner.

Thanks, that is misleading, although I would put FSD in a different category, especially since they took money made promises and didn't deliver. Here we have just bad naming: "partial zero emission", that is an oxymoron. Partial does meet the definition of zero.

 

[CWhite]

I am confused, are people having accidents while using autopilot and keeping it a secret? Sure sounds like that is what is being suggested. I have seen plenty of stats showing # of miles driven per accident of all brands of cars. Are Tesla drivers so “loyal” that they won’t report anything detrimental to the brand? Is Autopilot perfect, no, but I can drive down I-5 for hundreds of miles with just applying the weight of my hand in the steering wheel. Is FSD-beta ready for prime time? Not even close. Of course I haven’t heard anyone suggest it is. Am I sorry I paid $7k for it? Not in the least. As someone who has worked as a computer programmer since the early 1980’s I am fascinated by the technology. I paid the money knowing it could be a very long time before it’s perfected. Practically speaking I was actually thinking it might not be possible but I wanted a front row seat for the ride. Am I worried I could be in an accident because of it? No, because I’m paying attention when using it. Has it done some ‘dumb’ things, yes, I press the buttons and send an email. It is what a beta tester is suppose to do.

 

[Daniel in SD]

I am confused, are people having accidents while using autopilot and keeping it a secret? Sure sounds like that is what is being suggested.

I don't think anyone would ever try to evaluate safety using self-reported collisions.
Though Teslas are by far the safest cars on the road if you ask Tesla owners!
I wonder when the IIHS will publish fatality rates for Teslas?

 

[rxlawdude]

Thanks, that is misleading, although I would put FSD in a different category, especially since they took money made promises and didn't deliver. Here we have just bad naming: "partial zero emission", that is an oxymoron. Partial does meet the definition of zero.

No more of an oxymoron than "military intelligence."

 

[Terminator857]

... are people having accidents while using autopilot and keeping it a secret? ...

People are not expected to publicize autopilot (AP) accidents.
Lets do some math and see if we can come up with estimate of rate of AP accidents. Whatever the rate is, it does not mean it doesn't have lowest probability of accident.

Tesla Vehicle Safety Report | Tesla

At Tesla, we believe that technology can help improve safety. That’s why Tesla vehicles are engineered to be the safest cars in the world. We believe the unique combination of passive safety, active safety, and automated driver assistance is crucial for keeping not just Tesla drivers and...

www.tesla.com

recorded one crash for every 4.41 million miles driven in which drivers were using Autopilot technology (Autosteer and active safety features).

Tesla Vehicle Deliveries and Autopilot Mileage Statistics - Lex Fridman

This page provides estimates of Tesla vehicles delivered and Autopilot miles driven segmented by Autopilot hardware version, based on the following milestones: Sep-2014: Autopilot Hardware 1.0 installed (Autopilot not enabled). Oct-2015: Autopilot enabled. Oct-2016: Autopilot Hardware 2.0...

lexfridman.com

3.3 billion miles driven on AP 2020-04-22
5.1 billion miles estimated drivin 2021-01-01
1.8 billion miles between 2020-04-22 and 2021-01-01.
247M miles per month = 741 million miles per quarter.
741 million miles per quarter / 4.41 miles per accident = 168 accidents per quarter.
90 days per quarter
168 / 90 = 1.87 accidents per day on autopilot. AP accident about once every 13 hours.

 

[CWhite]

People are not expected to publicize autopilot (AP) accidents.
Lets do some math and see if we can come up with estimate of rate of AP accidents. Whatever the rate is, it does not mean it doesn't have lowest probability of accident.

Tesla Vehicle Safety Report | Tesla

At Tesla, we believe that technology can help improve safety. That’s why Tesla vehicles are engineered to be the safest cars in the world. We believe the unique combination of passive safety, active safety, and automated driver assistance is crucial for keeping not just Tesla drivers and...

www.tesla.com

recorded one crash for every 4.41 million miles driven in which drivers were using Autopilot technology (Autosteer and active safety features).

Tesla Vehicle Deliveries and Autopilot Mileage Statistics - Lex Fridman

This page provides estimates of Tesla vehicles delivered and Autopilot miles driven segmented by Autopilot hardware version, based on the following milestones: Sep-2014: Autopilot Hardware 1.0 installed (Autopilot not enabled). Oct-2015: Autopilot enabled. Oct-2016: Autopilot Hardware 2.0...

lexfridman.com

3.3 billion miles driven on ap 2020-04-22
5.1 billion miles estimated drivin 2021-01-01
1.8 billion miles between 2020-04-22 and 2021-01-01.
247M miles per month = 741 million miles per quarter.
741 million miles per quarter / 4.41 miles per accident = 168 accidents per quarter.
90 days per quarter
168 / 90 = 1.87 accidents per day on autopilot. AP accident about once every 13 hours.

Considering NHTSA says a person dies every 15 minutes in a car accident it certainly indicates that Teslas are pretty safe. Stats show something like an accident every 60 seconds in the US.

 

[Daniel in SD]

Considering NHTSA says a person dies every 15 minutes in a car accident it certainly indicates that Teslas are pretty safe. Stats show something like an accident every 60 seconds in the US.

Tesla's are a tint fraction of registered vehicles in the US (276 million).
I predict that when the IIHS publishes fatality data it will show that Teslas are not the safest modern cars (though mostly because I think Tesla drivers are bad drivers (not necessarily excluding myself. haha)).