What I want my autonomous car to really do well...

[TMeister]

My driving mix is probably not like a lot of you since most of my miles (25K in two years) are done in cross country trips. I average 200 miles a month when at home. I try to use AP when driving around town and pretty much give up right away. TACC is still jumpy, AP can't deal with minor disturbances around here.

Back in the day, AP kinda worked for these cross country trips on the interstate system. Initially the car wouldn't pass trucks in sweeping right turns apparently thinking they were going to infringe in my lane. Over time things have improved but it's still not good enough. I disengage a lot when the road is somewhat curvy, when approaching construction, when traffic increases, mostly because the car doesn't drive the WAY I drive. Now with NOA, lane changing is done way too late and many times the system misses the correct exit entirely. So yet again, I drive instead of letting the car do its thing. It's really disappointing.

What I want instead of features like city driving, red light recognition, interstate ramp merge/exit, enhanced summon is a solid TACC and an AP that can manage the mundane interstate hours. I don't get that with my car. This would be rock solid L2 autonomy on interstate roads. Forget trying to get L3-4-5. How about getting the existing system to work 6 nines on interstates? That kind of development is not flashy but it's what I want. Can I have that, please?

 

[diplomat33]

My driving mix is probably not like a lot of you since most of my miles (25K in two years) are done in cross country trips. I average 200 miles a month when at home. I try to use AP when driving around town and pretty much give up right away. TACC is still jumpy, AP can't deal with minor disturbances around here.

Back in the day, AP kinda worked for these cross country trips on the interstate system. Initially the car wouldn't pass trucks in sweeping right turns apparently thinking they were going to infringe in my lane. Over time things have improved but it's still not good enough. I disengage a lot when the road is somewhat curvy, when approaching construction, when traffic increases, mostly because the car doesn't drive the WAY I drive. Now with NOA, lane changing is done way too late and many times the system misses the correct exit entirely. So yet again, I drive instead of letting the car do its thing. It's really disappointing.

What I want instead of features like city driving, red light recognition, interstate ramp merge/exit, enhanced summon is a solid TACC and an AP that can manage the mundane interstate hours. I don't get that with my car. This would be rock solid L2 autonomy on interstate roads. Forget trying to get L3-4-5. How about getting the existing system to work 6 nines on interstates? That kind of development is not flashy but it's what I want. Can I have that, please?

Sounds like you just want Autopilot do drive your way on the roads that you drive. Fair enough. I can't blame you for that.

Personally, I find Auto Steer and TACC to already be a solid L2 on highways. But I get that it might not drive the way you like to drive.

 

[J X 3]

The best L2 is L5. you can't have a 5y old that is an ace on the highway and terrible in the city. The fleet is growing, the data volumes with it, progress will accelerate.
NoA is cautious to avoid costly mistakes. As they increase accuracy ,they can be a bit more aggressive. It's a new feature, give it a minute. They need interventions to use that data and make it better, disengagements are gold for them.

A system that uses just the markings of a single lane and the car in front to navigate,, it's never gonna be all that competent. Tesla is trying to use the entire geometry of the scene and extract as much usable information as possible, city or highway, doesn't make much of a difference. City is a bit more difficult , more things going on but good in the city makes it better on the highway too.

 

[TMeister]

Sounds like you just want Autopilot do drive your way on the roads that you drive. Fair enough. I can't blame you for that.

Personally, I find Auto Steer and TACC to already be a solid L2 on highways. But I get that it might not drive the way you like to drive.

That's what I was trying to say... I want a system that'll drive like me. Somehow, AP just doesn't cut it at the moment.

The best L2 is L5. you can't have a 5y old that is an ace on the highway and terrible in the city.

Ok, that kinda makes sense. Except that I want a savant for a driver on the interstate. Give me the 5 year old for city driving and I'll realize that I want to drive instead. Right now, I get a 5 year old everywhere.

Good points everyone.

 

[diplomat33]

That's what I was trying to say... I want a system that'll drive like me. Somehow, AP just doesn't cut it at the moment.

Ok, that kinda makes sense. Except that I want a savant for a driver on the interstate. Give me the 5 year old for city driving and I'll realize that I want to drive instead. Right now, I get a 5 year old everywhere.

Good points everyone.

I figured out what you were trying to say.

Maybe you are an amazing driver in which case, yes, it should drive like you do. But in general, I think a would be self-driving car should drive well, not necessarily drive like you or me. I would echo want J X 3 wrote. Ultimately, if you can get a car that is L3/4/5, it won't matter because it will do what you want and more. You are asking for a subset of full self-driving. Plus, it will drive better than you.

 

[TMeister]

Well, yes 1/2 of us are above average drivers!

I do have quite a bit of driving experience as well as track time in various cars and situations. City driving (like in Boston in the 70's) will educate you very quickly.

I really hope my Tesla will drive better than me at some point. Right now, not so much. Which is why I don't let it drive unless the road is boring. So yeah, make those boring stretches the focal point and I'll be a happy camper. It just seems like partitioning the problem space would yield optimized solutions for those spaces. Instead Tesla are trying to solve the whole space at the moment.

 

[diplomat33]

Well, yes 1/2 of us are above average drivers!

I do have quite a bit of driving experience as well as track time in various cars and situations. City driving (like in Boston in the 70's) will educate you very quickly.

I really hope my Tesla will drive better than me at some point. Right now, not so much. Which is why I don't let it drive unless the road is boring. So yeah, make those boring stretches the focal point and I'll be a happy camper. It just seems like partitioning the problem space would yield optimized solutions for those spaces. Instead Tesla are trying to solve the whole space at the moment.

I get it. Certainly, if you are an excellent driver, then the FSD bar would be higher. it would take more for FSD to impress you. I consider myself an ok driver but my friends say I am a bad driver. So I guess for me, the FSD bar is a lot lower.

 

[ewoodrick]

Don’t try thinking that “urban interstates” are really Interstates. I just finished 4 500 mile trips and FSD worked pretty awesome!

 

[Red Fox]

I have to agree with TMeister. I just finished a 4,000 road trip and found that I was disengaging the AP very frequently. (I have a 2016 'S' with the original hardware - for reference.) My main issues were the amount of drift within a lane and no noticeable response when an adjacent vehicle drifted toward the line. When a large truck drifts to the line and the AP is also drifting to the line, I have to take control back and move back to center or nearer the opposite line. I have not had any problem with the AP seeing the lines, it just doesn't stay centered very well and it does not seem to matter whether it is on a curve or straight section.

Does anyone know if the AP is expected to respond to vehicles crossing lanes next to you? It's possible that my comfort zone is just not wide enough to wait and see.

 

[diplomat33]

I have to agree with TMeister. I just finished a 4,000 road trip and found that I was disengaging the AP very frequently. (I have a 2016 'S' with the original hardware - for reference.) My main issues were the amount of drift within a lane and no noticeable response when an adjacent vehicle drifted toward the line. When a large truck drifts to the line and the AP is also drifting to the line, I have to take control back and move back to center or nearer the opposite line. I have not had any problem with the AP seeing the lines, it just doesn't stay centered very well and it does not seem to matter whether it is on a curve or straight section.

Does anyone know if the AP is expected to respond to vehicles crossing lanes next to you? It's possible that my comfort zone is just not wide enough to wait and see.

So are you on AP1 or AP2? "2016 S with original hardware" is not very clear to me which you are referring to.

 

[Red Fox]

I believe it is AP1. It's a 2016 but it pre-dates the hardware update.

 

[Uncle Paul]

Many of us are the same way with any other driver. When my G/f drives I also mentally find fault with her driving. She does not drive smooth enough for me, does not anticipate other drivers actions, does not look far enough ahead, does not flow smoothly into exit and entrance lanes, remains in the fast lane too long etc.

Think it is just part of our psyche to think our way is the best way.

For me I find Autopilot to be extremely helpful on long drives and also in rush hour traffic. I consider it a drivers aid for me, but am always ready at a moments notice to take over manually, especially in difficult situations.

It seems to be getting better with time, and for that I am thankful.

 

[diplomat33]

I believe it is AP1. It's a 2016 but it pre-dates the hardware update.

Thanks. I asked because AP2.5 is way better than what you described.

 

[diplomat33]

I still think that the best approach is simply to make a really good self-driving system. No two drivers will agree on how to drive. One driver may like to make assertive lane changes while another driver may like to drive more passively and cautiously. If you make NOA lane changes very assertive, the first driver will be happy but the second driver will freak out and think it is dangerous. But if you make NOA lane change too conservative like it is now, the second driver will be happy but the first driver will be unhappy. You are never going to make everyone happy. So the best bet is just to make an excellent FSD that drives the right way. If the car can self-drive correctly then you've automatically solved these issues of wanting solid lane keeping or wanting smooth TACC. Those are subsets of FSD.

 

[CarlK]

Many of us are the same way with any other driver. When my G/f drives I also mentally find fault with her driving. She does not drive smooth enough for me, does not anticipate other drivers actions, does not look far enough ahead, does not flow smoothly into exit and entrance lanes, remains in the fast lane too long etc.

Think it is just part of our psyche to think our way is the best way.

For me I find Autopilot to be extremely helpful on long drives and also in rush hour traffic. I consider it a drivers aid for me, but am always ready at a moments notice to take over manually, especially in difficult situations.

It seems to be getting better with time, and for that I am thankful.

Good point. I found fault with how many people drive too but I still feel safe in their cars. I just know they will not get into an accident. We also need to change the psyche when FSD becomes available. It could take a little time to get used to though.

 

[EVNow]

How about getting the existing system to work 6 nines on interstates? That kind of development is not flashy but it's what I want. Can I have that, please?

It is extremely important for Tesla's future to work on NOA on the city streets. Also, broadly speaking one doesn't preclude the other.

BTW, are you on AP1 ? I find TACC, steering and lane keeping quite good. I wonder how good NOA on freeways is with HW3.