Really feels like AP2 > AP1 to me

[chillaban]

So…. I hesitated for almost a week deciding whether or not to post this. I don't want to ruffle any feathers, but these are my honest and objective opinions and experiences.

As background, I have had an AP2 car ever since 17.17.4 (30k+ miles). Prior to that, I put 18,000 miles on an AP1 car. This last week I had some work done on my car and had an AP1 loaner, running 2018.21.9. The reason I mention this is: I've previously been told that AP1 has increased a lot recently and my earlier observations on AP1 are not necessarily valid, which was a good point.

So long story short, during my time with AP1, I really missed my AP2 car and found that AP2 was more capable. Now that I got my AP2 car back, I've validated most of these observations back-to-back. Some quick thoughts:

• AP1 is suuuuuuuper smooth. Whether or not it's right. It has a plan and sticks with it. For example, if there's a sudden lane shift 2 feet over, AP1 will happily drive straddling the lane market for 200 feet, slowly nudging itself back into where the new lane is. Or if a curve suddenly sharpens or changes, AP1 will smoothly depart the lane, and then lazily work its way back into its own lane. AP2 on the other hand seems to be willing to make rapid adjustments to correctness, even if it's at the expense of smoothness. For those situations, AP2 often can follow sharp lane shifts without departing the lane, but it jolts the cabin. So if you really like smooth driving, it's understandable why many folks still feel AP1 is more confident compared to AP2. Personally, I'd rather my robotic car drive correctly rather than smoothly.
• AP1 is far more reluctant at allowing activation in borderline situations. I drive through subdivisions with wide lanes and mainly just a center divider. AP2 allows activation on these roads even with just one divider and an implicit curb / parked car boundary on the other side. AP1 simply won't allow activation in these conditions. Overall, many roads that I take for granted being able to use AP2 on a daily basis are simply not available for Autosteer on AP1.
• AP1's lane identification performance is relatively poor for nonstandard lane lines. It had trouble recognizing lane lines in construction areas where lane lines were just those little plastic flaps. It had trouble understanding a green-painted bike lane in the far right lanes, and guessed an incorrect and overly left-hugging implied lane line. It eventually threw a TAKE OVER IMMEDIATELY when my lead car turned away.
• Speaking of that: AP1 performed much poorer overall when there was no lead car, or the lead car was doing something stupid. Even on highways going 70mph around a gently curved hill, AP1 had trouble maintaining center with no cars ahead. In the city, when lead cars run over lane lines (such as when they're making super slow lane changes or trying to change lanes at a stoplight), AP1 would lose that side of the lane boundary and sometimes that results in a troubling drift into another lane.
• Stopped car approach performance was worse than AP2. Right now, AP2 for me reliably stops for stopped cars if I'm approaching below 55mph. On AP1, it failed both times at 50mph. At 35mph it still failed when approaching a landscaping truck carrying some trees. Even at 25mph it failed around a curve — it seems to require more than 2 seconds of unobstructed view of a stopped car before it's willing to trust it. I find AP2 is far quicker at identifying stopped cars and identifies more different types of stopped vehicles and at different angles.
• AP1 seems completely ignorant to cross traffic, oncoming turning traffic, etc. AP2 actually starting from 2018.21.9 is fairly reliably detecting and braking for that kind of traffic.
• Man I love love LOVE AP1 for allowing me to make lane changes on local roads!

So overall, my experience so far is that without a doubt in my mind, AP2 is more capable than AP1 now in 2018.21.9 and beyond. Sure, I believe there might be counterexamples to this, but at least for my driving pattern, I am much happier with AP2 performance than AP1 performance, and this makes me optimistic about the future of AP2.

With that said, I still think AP1 is an excellent ADAS system, and I would not feel bad about having an AP1 car right now.

 

[Electroman]

Agreed on all points. Exactly my experience too between AP1 and AP2. Except the occasional scare of slowing down for imaginary obstacles. But those have been relatively rare.

 

[Robwoodruff]

AP2 definitely better than AP1 now in most cases.
The things AP1 still does better:
1) slows and excellerates more smoothly
2) allows lane changes on local roads.
3) recognizes and draws trucks, motorcycles and pedestrians, not just cars

 

[chillaban]

I have to say, the area I was most disappointed with AP1 was how it ignored most anomalous obstacles. I actually have been using AP2 on my city commute home for months. Other than for stopping at stoplights and making turns, it is days between needing to make interventions for safety.

AP1 this week, basically every time there was cross traffic (person turning left and cutting me off, oncoming driver accidentally veering into my lane, driver running a red light and still in the intersection when my direction is green), it would require an intervention.... It showed no signs of slowing or stopping for anything other than for cars traveling in the same direction as me.

AP2 actually has been getting all of these situations right, I'd say 90%+ of the time. I think the downside to this is that AP2 sometimes does brake and slow down more than necessary if it mistakes something irrelevant for a possible stationary car in my path of travel. However, at the end of the day, I really do feel safer driving AP2 than AP1. Sure, on rare occasion it does something sudden and non-smooth and that's unsettling. But seeing what kinds of scenarios it's able to recognize and slow for, while AP1 would happily plow right through it? It gives me more of a sense that the sensor suite has my back.

Long story short, AP1 seems like it's currently (and likely forever will be) a really high performing adaptive cruise control + lane following system for driver assist. AP2 on the other hand feels more like it's on the path to becoming a robotic driver. Both of course still require an attentive driver at the moment watching over the system like a hawk -- smooth or not, both systems' limitations can end really badly for an overconfident human.

 

[S4WRXTTCS]

It looks to me it's like this

AP2 is way more capable in handling stationary objects and cross traffic, but there is an increase risk of phantom braking events.

AP2 is less likely to follow the leading car in doing something stupid, but this still happened with the fatality accident involving the Model X. Where it followed the other car into the gorge area.

AP2 is really hampered by public perception which is really negative right now. Part of this wasn't really AP2's fault as most of the recent accidents into stationary vehicles (like fire trucks, police vehicles) were AP1 cars.

In August we're supposed to see a significant upgrade to the AP2 software. Something that might make it compelling enough for me to upgrade from AP1.

 

[number12]

I was going to post the opposite.

Model 3 ap2.5
Vs
Model X ap1.0 (sept 2016 still not sure if all ap1.0 is made equally - or if last few were better then 2015 for example).

Ap2.5 can not handle an intersection. Just know recognizes other cars. Etc.

Parity might be considered, but it’s give and take. I don’t think anyone can say ap2.0 equals or betters every aspect of ap1.0.

Also consider 2015 ap1.0 on V7.0 will less nags and not much different than ap1.0 today.

Pretty sad that 2015 tech is still being “compared” or argued which one is better. It’s been a solid 3 years.

 

[araxara]

AP2 still does not read speed limit signs

 

[TaoJones]

Lately, I've been testing AP2 (*21.9) with stopped traffic at stop lights *and* elevation, either ascending or descending toward the intersection. In terms of reacting to the stopped traffic, there appears to be improvement across the last handful of updates and in particular with 21.9. Improvement on flat terrain and descending terrain, straight on or around curves even; vehicles are recognized *and reacted to* much further away than in, well, months/years-ish past with AP2. This is one of the best features of *21.9 imnsho.

Ascending, not so much, and it gets worse. AP2 has this chronic propensity to drive toward and attempt to straddle raised medians (curbed) especially when driving uphill. Here, AP1 gets bonus points as well for the second-biggest thing I miss from AP1 locally, and that's (as noted above) the ability to change lanes on any road not considered a divided highway. Well, there are a lot of divided roadways not considered highways, and many with as many or more lanes than the highways. So there's that.

100% agreed that AP2 "feels like" something good is coming, that *gasp* might make use of the hardware that's been built into the car for the past 2 years (by the time v9 and certainly v9.1 are released by year endish). Someday, AP2.x will be better than AP1 across the board, or at least I'd like to think so.

Also agreed that AP1 remains a smoother ride. It'll top out eventually in terms of functionality but it'll never be less smooth than it is today, presumably.

Unfortunately, and here's where the proponents of AP2>AP1 and I diverge markedly, until speed limit reaction is fixed, every day I'm in the position to get tailgated or ticketed for speeding due to AS = +5mph > database value. I couldn't count the number of times that happened off-highway on one hand *ever* with AP1, and probably not even if I included highway obstructed views of speed limit signs.

I don't know that anyone would disagree with the assessment that once the cars make use of 10cm GPS and we get traffic signal reaction (versus recognition), all will be much better with AP2.x. We've already seen what Mobileye can do with great patience and 17 executions (see December 2016 video) with stop signs, so clearly we will get there one way or the other. Stack on top of that the yuuuuuuuge competitive advantage that Starlink will bring, and one has to believe that the future is indeed bright with AP3 and beyond.

It's the present that's a tad vexing in terms of navigating reality versus past promises.

Briefly stated, too late I know, anyone with an AP1 car (which will hopefully be me again in the next few months) should be in great shape for the next couple of years while this all shakes itself out (hello actual FSD and a non-silent E in EAP). As well, anyone who buys an AP2.5 car today is just going to get happier as time passes.

Personally, I'm done buying new Teslas until AP3, circa 2020-2021 or so. I don't expect Starlink-assisted cars by then, but a couple years thereafter for sure. If you think of Teslas as iPads or iPhones with some compelling reason to buy a new one every 3 years or so, the budget rationalizations tend to go much more easily .

 

[Saghost]

I was going to post the opposite.

Model 3 ap2.5
Vs
Model X ap1.0 (sept 2016 still not sure if all ap1.0 is made equally - or if last few were better then 2015 for example).

Ap2.5 can not handle an intersection. Just know recognizes other cars. Etc.

Parity might be considered, but it’s give and take. I don’t think anyone can say ap2.0 equals or betters every aspect of ap1.0.

Also consider 2015 ap1.0 on V7.0 will less nags and not much different than ap1.0 today.

Pretty sad that 2015 tech is still being “compared” or argued which one is better. It’s been a solid 3 years.

I can't speak to AP2/2.5 vs AP1, but I disagree strongly with the second half of your argument.

The Autopilot I got under 7.1 was nowhere near as capable and trustworthy as it is now. I'll grant they haven't added significant user facing features to it, but the refinement in handling of awkward situations and bad conditions has been very substantial.

 

[chillaban]

I was going to post the opposite.

Model 3 ap2.5
Vs
Model X ap1.0 (sept 2016 still not sure if all ap1.0 is made equally - or if last few were better then 2015 for example).

Ap2.5 can not handle an intersection. Just know recognizes other cars. Etc.

Parity might be considered, but it’s give and take. I don’t think anyone can say ap2.0 equals or betters every aspect of ap1.0.

Also consider 2015 ap1.0 on V7.0 will less nags and not much different than ap1.0 today.

Pretty sad that 2015 tech is still being “compared” or argued which one is better. It’s been a solid 3 years.

I can't speak to AP2/2.5 vs AP1, but I disagree strongly with the second half of your argument.

The Autopilot I got under 7.1 was nowhere near as capable and trustworthy as it is now. I'll grant they haven't added significant user facing features to it, but the refinement in handling of awkward situations and bad conditions has been very substantial.

I can't speak to AP2.5 on the model 3, but it would be inaccurate to say that AP2 cannot go through intersections. I've found in recent builds the consistency with which it navigates intersections is on par with AP1. It's just AP1 with a lead car can do much better mimicry of exactly what the lead car did through the intersection. Unfortunately this means that if the lead car changes lanes during the intersection (legal in California), you'll have to intervene too.

FWIW, AP1.0 on 7.0 and 7.1 is nothing compared to what AP1 has become. When it was first activated, I test drove it and was extremely underwhelmed. It had a ton of issues with exit diving, ghost braking when passing cars, true truck lust, etc etc etc. It was slowly refined all the way into early 8.0 days before progress plateaued. So no, it's not been a solid 3 years. AP1 had gradually improved for a solid 3 years. AP2 was introduced during that 3rd year and while progress was slow at first, it's now almost on year 2 and it's more exponentially improved rather than gradually.

Completely agree with what @Saghost said. I joined AP1 around the 2.20 days. It was pretty nice the way it was, but it was way better than the first AP1 I test drove when the feature was turned on in v7.0.

 

[dhanson865]

Stack on top of that the yuuuuuuuge competitive advantage that Starlink will bring, and one has to believe that the future is indeed bright with AP3 and beyond.

Starlink won't have anything to do with Autopilot in any way shape or form. It is for stationary ground systems, not mobile data like cars or cell phones.

It may be that 3rd parties will relay it just like the cell network does for internet now but you won't have a starlink satellite to car direct transmission.

The system will not compete with Iridium satellite constellation, which is designed to link directly to handsets. Instead, it will be linked to flat user terminals the size of a pizza box, which will have phased array antennas and track the satellites. The terminals can be mounted anywhere, as long as they can see the sky

You can see Elon pretty much rule out the idea of using it in a car at about 36m15s in on

 

[Electroman]

but it would be inaccurate to say that AP2 cannot go through intersections.

Agreed. In fact I think it crosses curved Xns much better than AP1. Here is an Xn I have noticed that AP1 almost always ends up correcting at the last minute after crossing the Xn to be on the correct lane because it is curved. AP2 on the other hand seems to follow the (imaginary) curve although there are no lanes in the Xn itself (the orange highlighted region).

 

[chillaban]

So I'm back in an AP1 loaner (long story), but yeah, a couple more thoughts:

(1) When an intersection curves and does not have lane lines, AP1 cannot extrapolate well to the other end. I've frequently had to take over because it was getting too close to a neighboring car, or is 1ft into the neighboring lane straddling a lane marker by the other end of the intersection. If an intersection is straight through, or there's a lead car very close in front of me, then it seems okay. Overall I prefer AP2's ability to go through intersections, and I drive on a 50mph speed limit road with 20 intersections every day.

(2) Speed limit reading is a blessing and curse depending on where you live. Around here, the database is generally correct and I've had multiple problems with AP1 reading speed limits. Here I am on 880 with a 65mph speed limit but it saw a 45mph ramp speed limit and stuck with it:

On some of the commercial streets I drive, it's 45mph undivided during normal hours but 25mph in a school zone during restricted hours. AP1 will read the 25mph limit unconditionally and then lock me down at 30mph max Autosteer speed indefinitely.


Overall, I don't find myself trying AP1 on residential/commercial streets anymore. County highways or real highways are the environments where it does well. OTOH, I've had a lot of success using AP2 on a broader variety of streets.


EDIT: Overall for AP1, on a sample size of 6 commute trips (mixed residential + county highway), I'm doing about 5-10 AP interventions per trip, not counting being first to a red light or making a turn. This is much higher than AP2, where I'm down to 2 or fewer.

On interstate driving it's more neck-in-neck but the longer I have this AP1 car, the more I'm noticing that for non-interstate driving, AP2 seems to get things right more often.