Does Tesla Believe that its own Autopilot system engages in "unsafe following"

[GEMINI 18]

I did the best experiment I could do. My wife and I got onto the freeway at 3 am and drove for 54 miles. I set the autopilot following distance to 5. I purposely stayed away at least 100 yards from her vehicle getting to the freeway from my house and returning to my house. No other cars were involved in this experiment as it was 3 am and we stayed in the #2 lane to avoid any possible cars getting on and off the freeway.

The car was not driven for the rest of the day. My safety score went from 95 to 94 the next day when I checked it. The only change was an indication on my daily graph that I had engaged in "unsafe following."

Why would Tesla allow its own system to engage in "unsafe following?" Or is there another reasonable explanation?

 

[MY1stEV]

My understanding is, anything that happens while FSD is engaged (other than forced disengagements) does not count against your safety score.

 

[JulienW]

Any score changes happened when you had AP off. When AP is on the is NOTHING that can happen that will change the score unless you disengage AP.

 

[ATPMSD]

This has not been my experience! You get hit for unsafe following when AP is engaged, as your test shows. This is why I dropped out of the Beta, the entire Safety Score system is deeply flawed.

 

[SilverString]

It happens quite regularly to me when beta 10.6 is engaged. The car will put itself into a situation where either it disengages on its own, or the driver is forced to disengage. At that point a maneuver is then required that will cause a hard braking, unsafe following, or forward collision error in order to exit the unsafe situation CAUSED BY AUTOPILOT. I have a forward collision warning ding that occurred for no reason, with no other cars near, and on an open stretch of road. At least half of my safety score dings are a direct result of Autopilot mistakes requiring emergency maneuvers. If you are beta testing FSD, this situation is common. Since I have Tesla insurance these errors cost me money.

 

[ATPMSD]

these errors cost me money.

Just to emphasize the point, you are allowing Tesla to adjust your insurance premium based on BETA software that has a lot of known flaws. Stop and think about that.

 

[whoosierdaddy]

Never had any issues with AP on vs. AP off. AP was always perfect, only ever got dinged when out of AP. Was always able to pin where and why it happened too. 600 or so miles on safety score before I got into Beta, 100 but not a perfect 100. AP was on every chance I got. The most frustrating issue was hitting a deep manhole cover in the middle of a turn and getting dinged for aggressive turning. Had one instance of unsafe following while trying to quickly drive around a dump truck dropping rocks and getting as close to the car in front of me as I could while we were passing.

 

[EVRider-FL]

This has not been my experience! You get hit for unsafe following when AP is engaged, as your test shows. This is why I dropped out of the Beta, the entire Safety Score system is deeply flawed.

Unless you can enable AP as soon as you leave home, the test is not conclusive. Nevertheless, I agree that the Safety Score system is flawed and that’s why I dropped out of the Beta queue a while ago.

 

[ATPMSD]

Unless you can enable AP as soon as you leave home, the test is not conclusive. Nevertheless, I agree that the Safety Score system is flawed and that’s why I dropped out of the Beta queue a while ago.

Supposedly, “unsafe following” is not active until you get to 50 mph (see post #3 above.) In any case, I suggest it is fairly easy to drive to a highway on a Sunday morning, do a test and then get home without tripping an unsafe following event while not on AP. But like you, as I dropped the Beta, this is just an academic point to me.

 

[SilverString]

Just to emphasize the point, you are allowing Tesla to adjust your insurance premium based on BETA software that has a lot of known flaws. Stop and think about that.

Yep. Even so, Tesla Insurance remains my cheapest option even if I drop to a safety score of 90 or lower. I pretty much know when I have had an incident that will drop the score and lately I just reboot at the end of the drive before I leave the vehicle. I already have the beta so as crazy as it is, I still use it. I could turn the beta off and just use the regular AP stack but despite all this my score is still at 98.

 

[jakimo]

Auto-pilot beta (or non-beta) certainly does not match what the Safety-score is seeking. However, it sure does 'train' folks how to drive a bit safer and pay better attention for Auto-pilot beta testing and training..

If for NOTHING else, I myself love the visuals.. and barely use Auto-pilot because of how unenjoyable it's decision-making is. But I do love the visuals, even though I drive without engaging Auto-pilot maybe 80% of the time. Now-a-days, I'll use AP only on easy areas and disengage before it starts acting young and drunk (for sake of a smooth and sensible ride, and for battery efficiency)

In my observation, we have many years to go before Auto-pilot is both smooth and battery-efficient. My 2021 MY LR doesn't even seem to have the right cameras and sensors to really pull it off correctly..

 

[RowdyMY]

I had ~3k miles on safety score before I was given FSD access. My early scores were abysmal before I figured out how I was getting crushed for follow distance.

I made an entire 250 mile trip on autopilot and had 40% unsafe following (or so I thought).

Like it doesn't hurt you, AP also doesn't help. I was getting dinged for on and off ramps, then when I engaged autopilot I was racking up huge miles with those bad scores for the day.

If you were in fact "at least 100 yards" away the entire time that wasn't even a factor for your follow distance. It doesn't measure and account until you're at 50 and you would have been greater than 3" away up until ~69 mph. Follow distance isn't graded if you're over 3" back.

 

[GEMINI 18]

I was only at least 100 yards away from my wife vehicle when getting to and from the freeway. I did not want getting to and from the freeway to be a potential cause of possible unsafe following. Once on the freeway, AP was engaged the entire time at a following distance of 5. My score went down. Based upon my experiment, I cannot find any other reason why my score was decreased from a 95 to a 94. Nothing else changed in my daily graphs except for the unsafe following distance that clearly indicated the day before there was significant time spent at an unsafe following distance which could have only been the several miles spent on the freeway with AP engaged at following distance of 5. I may do the test again at a following distance of 1 and see if that hurts my score even more.

 

[RowdyMY]

I was only at least 100 yards away from my wife vehicle when getting to and from the freeway. I did not want getting to and from the freeway to be a potential cause of possible unsafe following. Once on the freeway, AP was engaged the entire time at a following distance of 5. My score went down. Based upon my experiment, I cannot find any other reason why my score was decreased from a 95 to a 94. Nothing else changed in my daily graphs except for the unsafe following distance that clearly indicated the day before there was significant time spent at an unsafe following distance which could have only been the several miles spent on the freeway with AP engaged at following distance of 5. I may do the test again at a following distance of 1 and see if that hurts my score even more

Don't take this as being snarky. I want to help you and sincerely want to see you succeed.

I already tried things similar to you before I figured it out. My worst day of follow distance score was when I set it to "7". Thought surely my scores were coming from autopilot.

The tight window for follow distance (> 50mph, < 3") makes that one tough category and even with very little time in the negative condition without any manual time in the positive condition causes a huge impact to your daily score. If you exceed 49 mph at less than 1" briefly but then never again meet the graded criteria save for autopilot driving your follow distance score is going to be awful. The autopilot time won't buoy that transient moment at < 1".

 

[TresLA]

I did the best experiment I could do. My wife and I got onto the freeway at 3 am and drove for 54 miles. I set the autopilot following distance to 5. I purposely stayed away at least 100 yards from her vehicle getting to the freeway from my house and returning to my house. No other cars were involved in this experiment as it was 3 am and we stayed in the #2 lane to avoid any possible cars getting on and off the freeway.

The car was not driven for the rest of the day. My safety score went from 95 to 94 the next day when I checked it. The only change was an indication on my daily graph that I had engaged in "unsafe following."

Why would Tesla allow its own system to engage in "unsafe following?" Or is there another reasonable explanation?

Safety Score is currently being used to test for FSD beta. They’ve acknowledged it will probably change before ever used for insurance. Think of it less about safe driving and more about assessing the risk of the car’s environment, and also how well you as a driver can follow instructions, pay attention, etc. required for a better FSD beta tester. The merits can be debated, but in your specific post I can say from my experience that it’s pretty clear while on AP, you’re not dinged but the miles do count. This means, to game the Unsafe Following factor of the Safety Score system you need to rack up miles while driving manually over 50 MPH and NOT while on AP.

In other words, if you’re driving anywhere over 50 MPH (not just on freeways), make sure to follow behind a car that’s within 1 and 3 seconds ahead of you in the same lane otherwise it doesn’t help your Unsafe Following. If no one in front? Just sit on AP.

But if you get on a highway, happen to be behind Someone within 1 second following distance, get over to a safe lane and then turn on AP for 200 miles, then while exiting the freeway you turn off AP, someone cuts in front of you, your Unsafe Following score will plummet because all those miles on AP didn’t count towards Unsafe Following score, but percentage of time while manually driving you were within 1 second of lead car. On top of that you racked up so many miles that day your score will be dinged the next day because your Unsafe Following percentage was so low.

 

[GEMINI 18]

Thanks TresLA... that is a reasonable explanation.

 

[Mardak]

My early scores were abysmal before I figured out how I was getting crushed for follow distance

Unsafe following has a maximum value of 60%, and even at that value showing up as red, if that's the only negative factor, the Safety Score will still be 99 (98.6 rounded up). I.e., the weighting of unsafe following factor is so low that it doesn't really matter if trying to get in to FSD Beta with 99 score (and even less so when it expands to 98 or 97 scores).

 

[Acisplat]

Thankfully for me the safety score in California is illegal to be used for Tesla insurance.

 

[RowdyMY]

Unsafe following has a maximum value of 60%, and even at that value showing up as red, if that's the only negative factor, the Safety Score will still be 99 (98.6 rounded up). I.e., the weighting of unsafe following factor is so low that it doesn't really matter if trying to get in to FSD Beta with 99 score (and even less so when it expands to 98 or 97 scores).

I guess "abysmal" was the wrong word. I was using it relative to the beta enrollment metric.

I started with a 94 for combination of unsafe following and hard braking. Had to work back up to a 99 which took some time.

Much happier now without the SS always in the back of my mind.

 

[Zextraterrestrial]

Unless you can enable AP as soon as you leave home, the test is not conclusive. Nevertheless, I agree that the Safety Score system is flawed and that’s why I dropped out of the Beta queue a while ago.

I roll w/ a score of 69....it's getting better because there are days I drive my classic S & no autocross in winter time
Safety score Beta is dumb AF. I understand the use of it if they are trying to collect slow (crappy in many senses ) driver data. at <0.4 G's it surely is easier to look at video feeds and data without extra noise from pavement surface etc.