Autopilot user from day one, 10 years ago, and it's actually worse now

[eye.surgeon]

I was an early adapter, buying a Model S before autopilot was even released to the public, it was activated in my car a month after I bought it and it was great right out of the gate. I drove it for 6 years and sold it and have been away from Tesla since, until I recent test drove a new MY. I was very disappointed to find that Autopilot has actually gotten worse. After 10 years. So many nags, false braking. When AP first came out I could literally sit with my hands in my lap for 30 minutes on the freeway, observant but not driving the car. Now, it's so naggy, what's the point I might as well drive it myself. 10 years of development, what a disappointment. Honestly they could go back to 1.0 and it would be better.

 

[Tam]

I was an early adapter, buying a Model S before autopilot was even released to the public, it was activated in my car a month after I bought it and it was great right out of the gate. I drove it for 6 years and sold it and have been away from Tesla since, until I recent test drove a new MY. I was very disappointed to find that Autopilot has actually gotten worse. After 10 years. So many nags, false braking. When AP first came out I could literally sit with my hands in my lap for 30 minutes on the freeway, observant but not driving the car. Now, it's so naggy, what's the point I might as well drive it myself. 10 years of development, what a disappointment. Honestly they could go back to 1.0 and it would be better.

That's because your AP1 is MobilEye.

Different brands, different practices.

MobilEye sells consumer quality products while Tesla sells unproven experimental ones.

The difference is: MobilEye is tested inhouse first. Once it passed the tests, it sells to consumers. Tesla sells products for consumers to test first. Tesla consumers pay first whether the products are useable or not (pay $12,000 now and test your way to see the Summon and Auto Park will work or not).

If you want hands-free, GM Super Cruise (MobilEye components) has worked since 2017. No need to test. What you pay is what you get.

 

[snorp]

There is something special about the Tesla MobilEye implementation. None of the other manufacturers that have implemented MobilEye have delivered as good (in terms of lane centering, accelerating and braking, and using it on almost all roads) of a system as what Tesla AP1 vehicles offer. So it's not just MobilEye. Maybe Tesla pushed them to allow it's use beyond their comfort zone.

Whatever it is, I agree with OP that AP1 is very good for what it does. I find it's a more relaxing driving experience on the interstates that AP2 and beyond, including FSDb.

 

[Ostrichsak]

There is something special about the Tesla MobilEye implementation. None of the other manufacturers that have implemented MobilEye have delivered as good (in terms of lane centering, accelerating and braking, and using it on almost all roads) of a system as what Tesla AP1 vehicles offer. So it's not just MobilEye. Maybe Tesla pushed them to allow it's use beyond their comfort zone.

Whatever it is, I agree with OP that AP1 is very good for what it does. I find it's a more relaxing driving experience on the interstates that AP2 and beyond, including FSDb.

AP1 hardware w/MobileEye software was easily the best driver assistance tool and probably still is today based on my experience. That's a sad statement when you consider that package is 10 years old. Sadder yet when you juxtapose it to all of the hyperbole and empty promises that we've been fed over that same time period with little to no actual gain in user experience. I'd even argue it's a step back as my AP1 cars were and are the smoothest Autopilot version you can drive for all of the important functionality.

 

[jmaddr]

From someone who never experienced AP1, what was obstacle avoidance like? Anyone have it react to something sudden and avoid an obstacle? I’m wondering if the current AP state is all the added situational logic and learned behavior that cause the instability.

 

[Tam]

From someone who never experienced AP1, what was obstacle avoidance like? Anyone have it react to something sudden and avoid an obstacle? I’m wondering if the current AP state is all the added situational logic and learned behavior that cause the instability.

Its obstacle avoidance is not superior to HW4.

That means, without driver intervention, it will still run under the side of a semi-truck. That said, there's no evidence that newer HW/FSDb would avoid doing that. In 2019, a HW 2.5 2018 Tesla Model 3 did just that in Delray Beach, FL.

I guess the point of this thread is although both the old and the new AP systems can kill but the old one is more stable and more pleasant to use.

 

[eye.surgeon]

I never viewed obstacle avoidance as a key component of my AP1 system because always viewed it as a highly convenient stress-reducing drivers assistant rather than a self-driving system. IMO Tesla's quest for self-driving is misguided and has derailed them developing from what I actually want which is good comprehensive drivers assistance. From what I saw on my test drive, they've made what I want worse and failed at what they want.

 

[GBMaryland]

I'm not going to lie to you: I'm not impressed with the manner in which Tesla is limiting driver input to the FSDb / AP stack.

For example, currently you are prevented from turning off the option for FSDb to change lanes automatically. Basically, they've made FSDb into a potentially deadly system. I know they want to learn, but there are times when FSDb should not, under ANY circumstances, change lanes. ...and you can't turn off the ability for the car to decide to change lanes on it's own.

(FSDb is notorious for driving into turn lanes on the right side of the road, as if it's going to pass someone in the local [eg non-highway] roads.)

There are other things, but at this point there is no way Tesla doesn't know this is happening. All they need to do it add the feature that originally existed to turn of automatic lane change.

 

[Ostrichsak]

I'm not going to lie to you: I'm not impressed with the manner in which Tesla is limiting driver input to the FSDb / AP stack.

For example, currently you are prevented from turning off the option for FSDb to change lanes automatically. Basically, they've made FSDb into a potentially deadly system. I know they want to learn, but there are times when FSDb should not, under ANY circumstances, change lanes. ...and you can't turn off the ability for the car to decide to change lanes on it's own.

(FSDb is notorious for driving into turn lanes on the right side of the road, as if it's going to pass someone in the local [eg non-highway] roads.)

There are other things, but at this point there is no way Tesla doesn't know this is happening. All they need to do it add the feature that originally existed to turn of automatic lane change.

To put it simply: If Tesla's goal is safety (as they claim) and their AP/FSD product is the answer to dramatically increased safety (as they claim) how does limiting the access to said safety features make the situation safer if a driver is looking at the screen or texting?

How does this "punishment' create for safer situations to remove safety features? If anything, it should auto-enable of an inattentive driver is detected to make it safer for them and everyone around them. But no, they slap your hand by telling you something they claim makes you safer is now being taken away for a week because you dun been a bad boi!

It would be akin to telling someone with an unsafe driving record with speeding tickets and accidents in their history that their airbags & ABS are being disabled until they can drive safer. It's completely incongruent with Tesla's statement on their vision for the future.

 

[gsmith123]

That said, there's no evidence that newer HW/FSDb would avoid doing that.

I disagree with this. I've tested FSD Beta a lot, and it's static obstacle recognition is good. It's hard to really test it, but among other things it has handled a stationary, dark, abandoned vehicle in the middle of a major road successfully.

(I don't have a link handy but it was very similar to a AP crash video I think from an AP 2 or 2.5 car running straight into a parked vehicle. My car didn't crash.)

And the car accurately shows parked and disabled vehicles along the edge of the road, even at night at highway speeds. Perception is pretty good for these things.

Admittedly, I haven't tested the exact scenario that lead to someone running full speed into the side of a semi truck trailer (hard to find that situation for testing). But I've never seen a situation where the car just failed to perceive a vehicle in it's path.

 

[buckets0fun]

Admittedly, I haven't tested the exact scenario that lead to someone running full speed into the side of a semi truck trailer (hard to find that situation for testing). But I've never seen a situation where the car just failed to perceive a vehicle in it's path.

I’ve got a story along these lines and would agree with your statement.

1am, middle of utah leaving SLC on 80E, light snow. Car is on FSDb, vmax. Car falls over like it’s going to phantom brake, which I’m prepared for with a hover foot. Start to push it, FSDb takes over and commands the left lane before I have a chance to understand, blaring warnings and red alerts on the screen. Well, a pipeline/mine site worker’s service bed wasn’t wired correctly and had no rear lighting while doing 55mph-ish. Since then I’ve trusted AP a bit more than I used to.

 

[Singuy]

With all the nagging you can thank the government and the mountains of MSM attacks on autopilot. AP1 didn't have a lot of nags sure, but they were dropped after the first death in 2016.

Tesla's safety monitoring and cautious object avoidance increase dramatically as they approach the laws of big numbers. Now that AP is in 100x more cars vs 2016, the number of deaths would sky rocket if people were to continue to die at the rate prior to this massive growth. Injury and death due to AP has dramatically decreased or else we would hear 20 deaths a month based on the total of AP miles driven vs a decade ago. You can probably attribute this to heavier monitoring by Tesla which I know irritates some people.

 

[Tam]

With all the nagging you can thank the government and the mountains of MSM attacks on autopilot. AP1 didn't have a lot of nags sure, but they were dropped after the first death in 2016.

Tesla's safety monitoring and cautious object avoidance increase dramatically as they approach the laws of big numbers. Now that AP is in 100x more cars vs 2016, the number of deaths would sky rocket if people were to continue to die at the rate prior to this massive growth. Injury and death due to AP has dramatically decreased or else we would hear 20 deaths a month based on the total of AP miles driven vs a decade ago. You can probably attribute this to heavier monitoring by Tesla which I know irritates some people.

I think many people were thinking automous driving was a solved problem before the first AP fatality.

As people then learn it's not the case, together with the increased nags and penalties from Tesla, drivers now are more careful with AP/FSD.

It's not the advancement of technology but the punishments and awareness.

 

[buckets0fun]

I think many people were thinking automous driving was a solved problem before the first AP fatality.

As people then learn it's not the case, together with the increased nags and penalties from Tesla, drivers now are more careful with AP/FSD.

It's not the advancement of technology but the punishments and awareness.

A personal dvd player and a semi truck trailer. I’m not sure more nags would have done anything in that situation.

Driver responsibility has to come into play.

 

[Singuy]

I think many people were thinking automous driving was a solved problem before the first AP fatality.

As people then learn it's not the case, together with the increased nags and penalties from Tesla, drivers now are more careful with AP/FSD.

It's not the advancement of technology but the punishments and awareness.

There's definitely both an advancement in technology and more punishments/awareness. If it's just punishment without advancement then people would use it less. However advancement brings in more complacency so you have to increase awareness. Eventually it becomes so intertwined that you have no idea if it's the software or the driver monitoring that decreased the total number of crashes so it's best to analyze disengagement data which Tesla doesn't provide. Either way, the trajectory is less death but more annoyance...I rather take annoyance than death.

 

[GBMaryland]

To put it simply: If Tesla's goal is safety (as they claim) and their AP/FSD product is the answer to dramatically increased safety (as they claim) how does limiting the access to said safety features make the situation safer if a driver is looking at the screen or texting?

I don’t really understand how that applies to what I said?

I’m talking about presets that you can turn on for your vehicle globally that don’t require you to spend time eyeballing the screen.

A bunch of different things such as the inability to turn off regenerative breaking (which was available), and more recently the ability to turn off automatic lane change.

Basically, they are attempting to take any sort of driving preferences out of the hands of the owner of the vehicle. if I want to use FSD on the back roads then I sure as *sugar* don’t automatically passing anybody, for example.