Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register

FSD Beta Videos (and questions for FSD Beta drivers)

This site may earn commission on affiliate links.
Yea, I'm not too sure. It's possible that the car simply plans out a path and proceeds or not based on the traffic controls and projected paths of dynamic objects, so it doesn't really "know" it's an unprotected left or not. It's also possible that it does know for many of them but not all of them.
They need to know whether the car that is coming from the opposite direction is going to stop (protected left) or not going to stop (unprotected).
 
  • Like
Reactions: Matias
They need to know whether the car that is coming from the opposite direction is going to stop (protected left) or not going to stop (unprotected).

Yea, after some more thought, you're probably more right than me.

Tesla can have the software chime whenever the left signal is on + there's no left arrow light + pathing dots are blue and angled left.

Some part of my programming intuition tells me that it might be more difficult than we're assuming though. Keep in mind that this is the same software that makes obvious lane choice mistakes, so something that may seem obvious to humans might not be for the car software (at least the software might not be right often enough to make it into a deployable feature).
 
I've actually worked with huge fleet companies that have tried to use this kind of data. It doesn't work.
You mean all those insurance ads about discounts for safe driving are misleading? And those devices that permanently attach to your OBD 2 port are just eye candy? Tesla's database has got to be better at calculating insurance risk than either an OBD 2 monitor or a driver safety course requirement.
 
You mean all those insurance ads about discounts for safe driving are misleading? And those devices that permanently attach to your OBD 2 port are just eye candy? Tesla's database has got to be better at calculating insurance risk than either an OBD 2 monitor or a driver safety course requirement.
Just as a data point, my insurer ("State Farm" which is fairly large with 16% market share) still uses a very similar beacon (doesn't require plugging into OBD 2, it syncs your mileage using your phone, the beacon keeps track of events, probably with a built in accelerometer and IMU). I didn't opt in to that however (was in a previous program where I just reported mileage manually).
Drive Safe & Save™ Mobile - State Farm®
10 Largest Auto Insurance Companies (September 2021)
They also have a similar partnership with Onstar to just use that system instead of needing to add a beacon:
OnStar® Drive Safe & Save™ - State Farm®

There's an article on these devices (and similar apps). Pretty much every major insurance company on that top 10 list (which makes up 72% of the market) has either a device or an app that does the same thing. The article below only had the top 9 listed, and didn't list Liberty Mutual, but they have one too. I don't see evidence they have discontinued using them to offer discounts based on usage.
https://www.usnews.com/insurance/auto/how-do-those-car-insurance-tracking-devices-work
RightTrack® Plug-in | Liberty Mutual
 
Last edited:

A couple good maneuvers, and a couple bad. Another example of the beta struggling with intersections that aren’t two perpendicular crossings as it tried to make a right turn on a diagonal on red. A couple of really good unprotected left turns including one dealing with a jaywalking pedestrian.
 

A couple good maneuvers, and a couple bad. Another example of the beta struggling with intersections that aren’t two perpendicular crossings as it tried to make a right turn on a diagonal on red. A couple of really good unprotected left turns including one dealing with a jaywalking pedestrian.
This maneuver starting at about 07:45, cuts a wide berth around a stopped car on the left hand side but doesn't seem to recognize or show the open door and exiting driver of the white car on the right hand side

1632058536561.png


1632058582859.png


I'm almost surprised the door didn't clip Frenchie's side mirror, that is way too close and too fast
 
  • Like
Reactions: daktari
This maneuver starting at about 07:45, cuts a wide berth around a stopped car on the left hand side but doesn't seem to recognize or show the open door and exiting driver of the white car on the right hand side

View attachment 711296

View attachment 711297

I'm almost surprised the door didn't clip Frenchie's side mirror, that is way too close and too fast

I agree. FSD Beta's behavior in that clip was very unsafe. It was way too close to that driver getting out of their car. They could have been injured or worse if they had step out just a bit more.
 
  • Like
Reactions: AndreP
...wut?

Teslas insurance algorithms explicitly include using autopilot.... (for example forced disengagements from ignored AP alerts count against you)


Also since you seem continually unclear, FSDBeta (really city streets) is an L2 system, not L4.
How do you know all the metrics they programmed in to look for and data mined to come up with this? Could you provide us with a document or link of what Tesla, or any insurance company for that matter is doing, to determine their rates and what constitutes safe driving?
 
I agree. FSD Beta's behavior in that clip was very unsafe. It was way too close to that driver getting out of their car. They could have been injured or worse if they had step out just a bit more.
Legitimately, that looks inches + a split second away from what could have been a potentially serious accident and the visualization just shows a regular ol' parked car
 
  • Like
Reactions: diplomat33
I'm almost surprised the door didn't clip Frenchie's side mirror, that is way too close and too fast

Nah, it only looks close because of the wide angle camera, but I agree it was a little too fast for the level of uncertainty. There are many instances watching these videos with wide angle cameras where I thought the car would clip the curb but didn't because of the way wide angle cameras distort the corners / edges of the image.
 
Last edited:
Nah, it only looks close because of the wide angle camera, but I agree it was a little too fast for the level of uncertainty. There are many instances watching these videos with wide angle cameras where I thought the car would clip the curb but didn't because of the way wide angle cameras distort the corners of the image.
I tried to take that into consideration and is why I posted the first image showing the approach with the white car closer to the centre of the lens — Frenchie’s car didn’t budge to the left from that point until after it passed the white car and the trajectory already looks way too close for comfort

Giving more space to the stopped car on the left and cutting so much closer to an open door and someone exiting on the right is not good driving behavior, but I don't think FSD had any idea an open door or driver were there
 
Last edited by a moderator:
  • Like
Reactions: daktari
I tried to take that into consideration and is why I posted the first image showing the approach with the white car closer to the centre of the lens — Frenchie’s car didn’t budge to the left from that point until after it passed the white car and the trajectory already looks way too close for comfort

I'm pretty sure Frenchie saw the door opening, as that car had just parked there and opened the door before Frenchie's car passed it. Since Frenchie didn't comment on the door at all, I don't think it was that close. I think it was at least 1-2 feet between the door and Frenchie's car. I do agree it was a bit too fast.

FSD beta did see the driver though, as a pedestrian displayed on the visualization. Right now, FSD beta doesn't display anything for open doors, but it does take them into consideration, as we've seen in many other instances.

You can see how deceptive wide angle cameras are in this frame immediately after the door incident, where it looks like the car is inches away from the green container:

Screen Shot 2021-09-19 at 8.05.06 AM.png
 
Last edited:
I'm pretty sure Frenchie saw the door opening, as that car had just parked there and opened the door before Frenchie's car passed it. Since Frenchie didn't comment on the door at all, I don't think it was that close. I think it was at least 1-2 feet between the door and Frenchie's car. I do agree it was a bit too fast.
Personally I’d stop and let the lady close the door and move out before proceeding. That is not only safer but also common courtesy.
 
  • Like
Reactions: pilotSteve