Tesla Autopilot update

I'm coming up on having owned my Tesla Model Y for 5 months, and thought it might be useful to give a little update on my experiences using Autopilot. I use it pretty consistently where I can, with a few exceptions that I'll get into below. Since I don't have the FSD beta yet, that means that it's usually offered whenever there are lines on the road. I've found that it'll be available even on two lane roads without a center double yellow line, as long as there is a line on the edge of the road. So for example, there's a residential street nearby that has a bike lane marked, and thus has nice stripes along the edges but no double-yellow line. That seems to be enough for autopilot.

Right now, if you are on a divided interstate highway and some selected limited access US and State route highways, it'll offer Navigate on Autopilot, which is a slightly more autonomous mode, where if you've set a destination in the nav system, it will recommend, but not execute until you confirm, lane changes to get you into a faster lane when you're overtaking slow traffic, to keep you from clogging the left lane, and to get away from construction cones, as well as to make sure you're in the correct lane when the road divides or you need to take an exit. It will also actually signal and make the exit for you so that you follow the navigation. It's pretty seamless, and the amount of interventions I need to make in this mode are very minimal, though occasionally it is fussing at me to get out of the left lane when I am literally in the process of overtaking a slower vehicle. My only real complaint here is that it is ultra-conservative about speed on cloverleafs, to the point that I always feel like I'm annoying the driver behind me if I leave it to its own devices. But given what I'll say below about how it manages curves elsewhere, that's probably appropriate.

Regular Autopilot is what happens everywhere else. It follows lanes, will change lanes when safe to do so if you signal your intent with a turn signal, it stops at stop signs, traffic lights, and maintains a safe following distance. On 2 lane and some 4 lane roads, the max speed you can set autopilot to use is speed limit + 5MPH. On other 4 lane roads, the logic allows you to set whatever speed you want. There are settings in the system regarding how you want it to manage set speed vs speed limit, so that when you invoke Autopilot, or the speed limit changes, it responds in the same way you'd be when driving. It still will not make turns that are more than just following a curve in the existing lane of travel, so even if you have navigation active, it won't make that right turn at the intersection it just told you to do. As I said initially, it works well for what it is, as long as you remember that it's not autonomous and pay proper attention. Inclement weather is a mixed bag. I was honestly super-impressed with how well it did on a secondary two lane road on a foggy, rainy night where the visibility wasn't great, but the driver is still definitely responsible for managing its safe speed - it doesn't slow down because the wipers are on, or because of temperatures, or whatever. I have also had situations where Autopilot was unavailable due to crud on the front or side sensors when driving in a post-snow melt where that salt and slushy crud all mix to throw a bunch of trash on your car and you use a half-gallon of winshield washer fluid in one drive.

But I now have a better sense for what it does poorly, what confuses it, etc. and that has led to both some situations where I don't use it, and areas where I have to intervene more frequently to keep things literally between the ditches. Here's a list in rough order of increasing severity: 

• Traffic lights - AP is pretty conservative about lights. Unless it's following a vehicle at close range so it can see that they're proceeding through the intersection, it requires you to tap the stalk or press the accelerator briefly to confirm it's safe to proceed through the green light even though it's clearly recognizing it as a green light. I had hoped this was a matter of teaching the AI, but it hasn't improved at all over time. It will stop for a yellow if it hasn't entered the intersection yet, including if you have already given it permission to proceed because the light was green and now isn't. This all makes sense in terms of safe, conservative driving, so I can't really complain all that much other than to wonder what is different where FSD is concerned.
• Flashing lights that aren't normal traffic lights - those warning signs at the side of the road for pedestrian crossings, school zones, etc. usually are yellow and make it think it needs to slow down for what is soon to be red, or per driving rules, slow for a flashing yellow at an intersection to be sure it's safe to proceed. It especially doesn't know what to do with various flashing yellow lights that look and are placed like traffic lights, such as the emergency signal in front of my fire department that has a normal red on top that is only ever lit when the trucks need to exit, and a smaller yellow underneath that is always flashing, or the horizontal flashing lights signifying an intersection that is flashing red in the cross direction. Given the yellow light logic I mentioned above, you're having to forcefully override it with constant pedal application to get it to go past those, and if you're not paying attention, you are going to annoy the driver following you.
• shadows on the road - the lovely sun-dappled road from your car commercial occasionally makes Autopilot think there are obstacles in the road. Same for lane changes where a large truck is casting a huge shadow of itself into the open lane between the two of you where you want the car to go. It'll occasionally decide the truck is actually in that lane and refuse to make the lane change until the angle changes and the shadow isn't there. This is usually only a problem when it's a huge contrast difference, like bright, direct sun filtering through tree leaves.
• Indistinct lanes - areas where it goes from 2 to 1 or vice versa, lanes without a center line, etc. confuse it occasionally. It can guess where the lane is, but it doesn't always guess right, and in situations where it's 1 lane becoming 2, I have had it kinda dither for a minute before it commits to a lane, and when the lane is not clearly delineated by lines on both sides or is exceptionally wide, such as during a 2 to 1 merge, the car does tend to wander a little trying to decide how to center itself in the lane. 
• Blind hills - those fun little whoops that make your stomach jump a little and make the car go a little light on the wheels when you take them quickly? Autopilot basically panics when it can't see the road ahead, and the threshold for that is just long enough that it'll be upset before you get to the crest of the hill where it can see again. Usually it just disables itself, but it has panic braked on me at least once. This is generally worse at night. 

The above things aren't really unsafe, and are more situations where the system is being understandably conservative, but significantly more conservative than a human would be, so it's more something I mention here as a quirk or limitation in the system. The following are more likely to actually cause an accident if not caught quickly enough, and this is the area where a lot more work is needed before full self driving even as Tesla defines it is going to be a workable thing. 

• Misaligned intersections - Autopilot sorta guesses where the path through an intersection is, because the alternative would be to shut down due to the lack of any lines on the road to guide it. Reasonable assumption is "continue roughly straight" but often intersections shift slightly, and there are a couple where it has tried to drive into the median because it didn't realize fast enough that the lane shifted almost a full lane-width to the right, and times where it unceremoniously changes lanes in the middle of an intersection because it picks up the wrong lane marker when re-acquiring the lane. Also if you're the first car in an intersection, stopped at a light with cross traffic going past, it will frequently just give up and tell you AP isn't available because it can no longer make a reasonable guess as to the correct path of travel. Intersections with stripes for various turning lane paths sometimes confuse it too, rightly so. 
• curvy secondary roads - you know the type, those roads that those of us who enjoy spirited driving in a suitable vehicle relish the chance to use without someone in front of us. They're probably posted 35 or 40, but they have plenty of curves, some with those yellow signs suggesting the maximum safe speed is 30, or 25, or even 15. The Model Y actually handles pretty well in spirited driving. It's not my 911, but it can be hustled through those roads safely and enjoyably. The problem is that Autopilot is not capable of managing that sort of road even at speed limit + 5 mph without leading to some potentially scary last-minute interventions. After playing with it for a while, here's the conclusion I've come to: Autopilot is not responding quickly enough to the clues being provided, including the line direction, possible warning signs and even those arrow markers along the side of the road to consistently decelerate to a safe speed that it can manage, nor is it willing to apply aggressive enough steering angle to compensate for maybe carrying too much speed into a turn and getting a little bit of understeer. It has briefly crossed center lines, been close to crossing the outside line and being on the shoulder, and generally not been great at keeping equal distance between the lane markers through turns if more than about 40° of steering angle, or any braking, is required. As a result, I have to do one of two things on roads like this - either I have to manually dial back the max speed at the entrance to a turn it typically botches, or I can only use Autopilot when it is following another vehicle, since they're likely to slow down at the point you'd expect, thus the robot will too in order to maintain following distance. 

There are multiple instances of pretty much all of the above concerns where I have wished for the ability to take a more active role in "teaching" the autopilot to do better, whether it's the ability to give it a voice keyword to tell it that the flashing light it thought it needed to stop for isn't a traffic light so it remembers for the future, or to invoke a "watch my path" mode so that it can store some data about appropriate entry and exit speeds and steering angles on secondary roads and gradually build a crowd-sourced knowledge about how to manage them. Even being able to more successfully navigate roads that I drive all of the time by storing some of that locally, and realizing it needs info from a server for a new road would be a huge improvement in the performance in some of these fuzzy areas. 

I joked with a friend after seeing a Tesla with a Student Driver sticker that given the fairly stupid robot piloting them a lot of the time, probably that should be a factory-installed option. Don't get me wrong, it's still amazing, and I am definitely glad I have it, but objective discussion of where it can be improved is always going to be useful and necessary.  

Man rants about phones

I've been a Sprint customer for 20+ years, partially because I used to work for them, and now mostly because of inertia (generally not having enough issue with them to switch), and still having a few friends that work there and some stock left over from my employee days that make me moderately biased toward their continued success. Now that they've formally been absorbed by T-Mobile, I have been getting emails asking me to come to a store to switch out my SIM to "take advantage of the new combined T-Mobile network", which I did... or at least tried to do, today. 

My daughter's LG G7 Fit, which I bought unlocked from woot for cheap, swapped SIMs no problem. My nicer, purchased directly from Sprint, G7 Thinq was declared ineligible for said SIM swap "due to device". After talking to someone with a technical clue in one of the repair stores, it's apparently being blamed on antennas and bands, i.e. Sprint's specified cocktail of bands and CDMA support for their SKU of the G7 apparently makes it not completely compatible with the long-term network/band layout TMO is using. So while my phone isn't exactly going to stop working tomorrow, I am not gaining what is arguably the primary benefit for the merger - better coverage, and quite likely my coverage via the legacy Sprint network will get worse the longer we go post-merger, eventually terminating in a message that I need to upgrade or risk losing service when they want to turn the lights out on the old stuff at (I'm guessing) the 12-18 months post-merger mark. I'm sure TMO doesn't want to keep CDMA around any longer than absolutely necessary, and Verizon has already announced (but delayed, or else it'd already be gone) the sunset of their 3G network too, so the writing is on the wall. 

So suddenly I am realizing I'm in the market for a new phone. I looked, and I've owned this for 2.5 years. My typical dwell time per phone is about 2.5-3 years, mostly because I tend to not trash them and I keep using them until there's a real reason to upgrade, so this is maybe a few months early in the upgrade cycle, but not too terrible I guess, since it's been paid off for 6 months. 

At this point, I'd be stupid to buy a new phone that isn't 5G, and I still want Android, but I'm finding that this is tweaking a few recurring annoyances in the phone market that might be worth writing about briefly. 

There are a few features that I care about on a phone beyond the obvious table stakes that define it as a smartphone. These may not be interesting to anyone but me, and are admittedly somewhat specific to my use case, but either way:

• Fingerprint reader

My phone and my laptop both have a reader built in, and now that I have set it up so that 1password and some other apps can use it in addition to basic login/unlock, I would never buy another phone without it. I've actually considered getting an external one for my desktop because I miss it when I'm using that. The dedicated hardware reader on the back side of the phone makes so much sense on account of the way you hold it anyway, so I have a strong preference for that, since that's my muscle memory now. Even before the current phone with the reader, the LG G4 had the power button back there, so I'm going on 5 years of turning on my phone that way.

• Software update cadence

This has been a problem for Android since the beginning, because Google didn't have enough market power to exert direct control over updates, and was left beholden to both the device manufacturers and the carriers, both of whom have their own development cycle where they add/update/break their own software and UI skin, and test cycle before they actually release updates to the end user. Google addressed this to some extent by making more and more of the OS carrier and device independent so that more bits could be updated directly via the Play Store, and moving to a monthly security release so they could patch holes in the rest of it in a way that could be deployed more rapidly without the extensive dev and test cycle that comes with a major feature release. My experience is that is still hit and miss, and I'm not sure whether the carrier or the device manufacturer or both is to blame for that. I was getting monthly security updates about quarterly (i.e. 3-ish months behind current) on this phone, but that dried up when the merger closed, so I'm currently stuck on May 2020.

• Expandable storage (microSD slot)

No matter how much storage you spring for when you buy the phone, it's probably not going to be enough, especially given the arms race that seems to be happening with camera sensors and their megapixel counts and support for 4K and even 8K video. Granted, with the accelerating transition to cloud services, streaming audio, etc, maybe that's not as important as it was a few years ago, but I'd rather just take advantage of how cheap ridiculous amounts of storage have gotten, slap a microSD chip in my phone (current one is 256GB) and not have to think about storage use, ever, even if I put a large portion of my entire music collection on it, and keep every photo and video I've ever taken with a smartphone, and back up all of my SMS weekly, and, and, and... 

• Wireless charging support

I have several wireless chargers now. They're not a complete substitute for USB-C charging, but they're useful, and I'd be annoyed if I couldn't charge my phone by setting it in the built-in phone holder in the Tesla.

• Headphone jack

I have Bluetooth headphones, and a bluetooth speaker, and a bluetooth adapter to let my phone talk to my home stereo, and even in the old pickup with the aux jack, I got a little bluetooth to aux adapter so a real headphone jack is more of a nice to have anymore, but having the ability to put the big wired over-ear cans on for noise reduction and feed them from my phone is still helpful.

In the US Android market, with basically 3 carriers, there are a few main brands that pretty much everyone considers because they're consistently offered and supported by the big 3 as the flagship devices. There are others that are considered budget offerings, but that's not really my focus, nor is how much "better off" I'd be with an iPhone. This is specifically about the tradeoffs for my options in Android. Pros and cons to switching to Apple is, as Alton Brown says, another show. 

• Samsung

Samsung makes pretty good hardware, but it is increasingly hamstrung by their insistence on forcing you to use their (usually inferior) versions of apps and features that already come with Android, most notably their voice assistant Bixby, along with their heavy-handed UI reskin of Android. And they're simply not good enough at software to justify this. I can't find the story to link to it, but I remember reading some really impressive horror stories around their development practices that made me pretty leery of ever owning another Samsung device. Between that and the lack of an MicroSD card slot [edited to add 4/5] (apparently this resurfaced briefly on the S20 but is gone again on the 21) and often a headphone jack, I left Samsung behind after the Galaxy S4 and mostly haven't regretted it. Part of why I use Android is that I like the app ecosystem. Google has a good keyboard, voice command features, tap and pay app, photo app, and OS UI. I don't want to have to fight with my device to use those things. 

• OnePlus

I haven't been paying a lot of attention to OnePlus. Brief glance through shows generally solid hardware, though it depends on the specific phone model whether it has wireless charging, none of them have microSD, and they're mostly using the under-display fingerprint reader. I had hoped they were pure Android as well, but it looks like if they were, they've become infected by the need to "improve" it with their own touches. To be fair, it seems fairly well reviewed, and the 8Pro is a reasonable competitor to the LG V60 from a spec and pricing perspective though it doesn't seem to be offered through TMO as some of the others are. 

[Edited to add 4/5] The Oneplus 9 has been released since I wrote this, and initial reviews all say it is a very solid phone, a good flagship offering capable of competing with Samsung, and generally better hardware than the current Pixels, though the above concerns still exist. 

• Google

Google has their Pixels, and the main selling point for those is that you're getting a pure Android experience. This is about as close to the Apple direct software upgrade cycle as you get in Android, for better or for worse. One assumes that the hardware and software integration means that they work well together and bug escape is minimal, but Google is a pretty siloed company, and seems to be getting increasingly dysfunctional, so I'm not sure how valid that assumption is anymore. Google also gives you access to the newest Android features here first, and their camera/photos app has been doing well competing with Apple for impressive quality in poor conditions. The current gen has a hardware fingerprint reader in the back where I like it, but no provision for external storage unless you count plugging something into the USB-C port. Google assumes that the only thing you needed all that storage for was pictures, and gives you unlimited storage on Google Photos, but at "high quality" which is not original quality. Also, Amazon's included photo storage for Prime members is better as a photo backup solution, as it has no such restrictions. Google really wants you to subscribe to one or more of their services to address this. I'm also sad that they have gone away from the Nexus name, because I enjoyed the hat tip to Blade Runner, but since ultimately I'm going to have to compromise somewhere, I'm considering the Pixel 5 more seriously than I have considered previous Google phones. 

• LG

I've had 2 LG phones, the G4 and the G7. Been very happy with both, as LG has gotten pretty good about mostly leaving Android stock (my phone has a Google button to invoke Google's assistant), and the hardware was quite good, with one or two exceptions. LG had fairly good speakers in the phone as far as those go, and they also had a real headphone jack with a fairly decent quality DAC such that it could drive a good set of headphones surprisingly well. 

The exceptions? As it ages, my fingerprint reader is getting flaky, which is kind of annoying but not bad enough to be an issue yet. Really, the big issue is that LG sucks at software updates. My phone, released in mid-2018, has already been abandoned on Android 9, despite Android 10 coming out roughly a year after the phone's initial release. Some flavors of G7 (looks like Korean and EU versions, and possibly the TMO version) got Android 10, but neither mine nor my daughter's appear to be among the chosen ones. And there are now rumors flying [edited 4/5 to add] confirmations that they might be interested in exiting the phone business entirely, which makes me at least a little leery of buying another LG and potentially ending up with a truly orphaned device [edited 4/5 to add] a complete no op, since they're still coming from the factory with Android 10, despite us being on to Android 11 already. LG has also brought out their own payment app to compete with Google Pay and Samsung Pay, so it seems they're losing the plot on not substituting their own apps again. 

Setting all that aside for a moment, LG has a weird collection of what might be considered their flagship devices right now. The V60 is probably the closest analog to the G series, supports 5G and has a microSD slot, but unfortunately has moved to using the under-display (and thus front-mounted) fingerprint reader that apparently doesn't work as consistently as the hardware ones, so it's on the short list, but I'd have to relearn my muscle memory of reaching for the phone and unlocking it with my index finger that is already in the right position on the back of the phone based on how I hold it. But they've also been chasing the gimmicky - they have 2 different varieties of multiscreen phones, the Wing, which has a screen that rotates 90° while leaving the lower half of the vertical screen available for simultaneous use, and a second screen case for the V60, which opens like a book to reveal 2 screens. Moving parts to break, more screens to use battery and get broken, limited case options, and generally thicker to accommodate all of this for questionable benefit. It all seems a bit "shark jumpy" to me in light of the above mentioned rumor.  

I don't think this set of wants and needs is that unique to me, but maybe it is. Fortunately I have time to do some more research and not rush to make a decision since my current phone is functional.