I’ll admit it — I’m super eager for the first edition of Tesla “Full Self Driving” to roll out. Before Tesla critics and skeptics start freaking out, I should make clear that I don’t expect to nap, work, or watch movies while using initial iterations of Full Self Driving. I currently use Autopilot a ton, and I’m well aware that even though it is handling most of the driving just fine while I’m essentially a simple passenger, I am 100% in charge of the car and need to be vigilant and cautious. I do keep my hands on the wheel almost all the time and constantly scan the road — as if I’m a pretend driver intent on acting out the role so well that I could win an Oscar (or at least a Teen Choice Award).
When Autopilot is engaged, I’m convinced that my family and I are much safer — it’s easier to scan all directions and watch out for unexpected challenges as the car drives itself smoothly. I look forward to supervising the car in this way as it drives itself in more scenarios and possibly even from parking space to parking space on some trips. But I do have some cases I’m curious about since Autopilot consistently can’t handle them. I’ll come back to those in a moment.
First, though, for a bit more background, I’ll note more about my history with Autopilot. I had a 2015 Model S with version 1 Autopilot hardware in Poland (it was a shared vehicle for our regional Tesla Shuttle services). That was Mobileye hardware. After Mobileye and Tesla split, it took a while for Tesla’s fully in-house Autopilot suite to catch up to that earlier tech. Though, to many people’s surprise, Tesla’s in-house solutions caught up to and surpassed the Mobileye tech in a relatively quick period of time (“relatively quick” in the grand scheme of things). By the time I test drove a production Model 3, it was clear Tesla had made strong advancements in Autopilot abilities and smoothness.
Perhaps most notably was that Autopilot had become much better at changing lanes. With the 2015 Model S, I drove from Poland to Paris and back and used Autopilot on much of the trip. Autopilot made the long journey so much better than it would have been otherwise. However, changing lanes was more of a gimmick than a helpful feature — after a while, I stopped using it. When I test drove a Model 3 (or three of them) in 2018, it seemed Autopilot’s automatic lane change feature had become genuinely useful. By the time I got a Tesla Model 3 in August 2019, Autopilot had surely improved more, and not long after that purchase, we got an update that really boosted the car’s skill and smoothness changing lanes. Now, I seldom prefer to change lanes myself. The car is better than me at it and I personally have the ability to look around much more (scan the area for threats) while the car does the core work. This is the case in city traffic as well as highway traffic.
Long story short: I’ve seen Tesla Autopilot improve considerably over the years. Tesla Autopilot today is worlds better than Tesla Autopilot from just 2–3 years ago.
My top questions and thoughts about Full Self Driving as I somewhat impatiently wait for it concern 1) areas where Tesla Autopilot still has trouble and 2) areas where I haven’t gotten to see its skills at all since Tesla hasn’t let owner cars attempt those tasks.
1. Large intersections that curve a bit.
My biggest problem with Tesla’s current version of Autopilot is at a couple of intersections I routinely cross. There are two possibilities that are rather common and shouldn’t be. One is that the car will drive all the way through but with some sudden jerking of the wheels back and forth. The car gets confused but then quickly resolves its problem and gets back on track. Another possibility is that the car gets confused and can’t correct itself. The car will then either abort Autopilot (at the easier intersection) or try to turn into a lane that goes in the opposite direction (the harder intersection). There are always a lot of cars at the red light in the second case, so I’m confident the Tesla would just stop there rather than try to drive down the wrong lane, but stopping in the middle of an intersection at an odd spot when cars are routinely driving quite fast behind you is not at all advised and could easily lead to an accident.
When I go through these or similarly challenging intersections, I have two options. One is that I’m just hyper-vigilant and quickly right the car if it starts going in the wrong direction. The other is that I just disengage Autopilot before going into the intersection. I’ve started doing the latter much more because the former has a tendency to scare people both in and out of the car, and it’s just not fun.
I haven’t noticed any progress on these cases since August, so I wonder when and how they will be addressed. When will the car be able to go through these intersections without getting a bit out of line or even abandoning its self-driving tasks completely? (This is a genuine question since I honestly have no clue at all — perhaps an update this week could include the necessary intelligence, or perhaps it could take years due to the oddities of these cases.)
The one thing that gives me a lot of hope is one of the ways Navigate on Autopilot works differently from Autopilot. Navigate on Autopilot knows what lane to be in when a turn is coming up. It can be better at planning for those than me. The thing is, it has an actual route programmed in. Based on maps, it thus knows what lane it needs to be in for an upcoming turn. I presume that also means that it should know what lane to stay in from the beginning to the end of a big, curving intersection. I presume that means it knows where the lane is headed. I presume that from paying attention to the map, it will know to keep its butt going straight rather than giving into some urge to dart in another direction to follow some lane markings. We’ll see.
Autopark is not exactly my favorite feature in a Tesla. Maybe if I used it more, I’d like it more, but I don’t use it much because it freaks me out and it often doesn’t complete its task. My fear of the feature may be completely irrational, but when I’m sitting in the driver’s seat and the wheel starts spinning really quickly and turning the car, it can be too nerve-racking to let it keep going when there are two cars on either side of the space that could get smashed with one simple mistake. It’s just not worth it to me to let the car keep going when I’m not 100% certain it sees everything as it should and I wouldn’t have time to avoid an accident if it made a wrong move.
The part that I know is not ideal comes at the end of parking the car. It has a tendency to not back up as far as it should because it spots the parking block, thinks it’s a wall or obstacle, and then completes the task while sitting a couple of feet further up than it should.
That’s the end of the parking experience. The beginning would basically be Smart Summon, which does the work of moving through the parking lot. As far as I can tell, Smart Summon would be quite good if given the chance to operate independently. After hours of testing it, I’m not worried about its capabilities or caution. It’s just that last 1–5% where I don’t yet see Tesla approaching perfect. (Or perhaps a few tweaks of the software code will magically make it smoother than me parking myself. We’ll see.)
3. Turns, stop signs, and red lights
This is an area where Tesla Autopilot still doesn’t mess around. The car won’t even try to make a turn or stop at a stop sign or red light. So, I have no real insight into how well Tesla vehicles will navigate these tasks. This is what I’m most eager to see — the car attempting to turn on its own and stop in precisely the right spot in front of a stop sign or red light beforehand.
As soon as the car can turn at an intersection, I think I will basically be able to sit and watch the car drive itself from a starting point to an ending point without my interference (other than touching the wheel to show that I’m paying attention).
Again, I don’t expect that I’ll be able to sit back and rest my eyes or check my phone any time soon. However, I will be thrilled and smiling big if I can witness an A-to-B self-driving trip in our Tesla Model 3.
I’ll be honest — I don’t have much hope on this one. However, it’s sort of a biggie that I’d really like to see resolved. The most common reason I have for not using Autopilot is potholes. We’ve got a decent number of large potholes in the area, and if I leave it to Autopilot, the car will slam into many of them and leave all passengers a bit too jarred. Also, I doubt it’s good for the car itself to hit so many holes. So, I disengage for large portions of an average trip and have to drive myself (the horror!) to avoid the bad spots on shoddy roads.
I don’t have much faith in Tesla resolving this issue any time soon, since I think it’s too difficult for Tesla hardware to notice these potholes, or notice them early enough to move around them without spazzing out and risking road safety. But we’ll see. I’d be more than happy to be proven wrong!
From a daily user perspective, these are the big edge cases in my life that I’m eager to see progress on. As it stands today, I appreciate and use Autopilot more than I expected to when I bought the car, and I’m just excited to see the next stage. I’d highly recommend Full Self Driving if you are on the line, as I was to some degree before placing the order. It’s amazing what the car is capable of today, and I’m convinced it makes driving both more enjoyable and much safer. I know the tech will continue to improve, but it’s already a genuine wonder and better than anything else you can buy in a passenger car.
In the coming few weeks, I plan to interview a handful of autonomous-driving experts on a variety of robotaxi matters. I want to ask them about other topics than what I discussed above, but I may also slip into these topics a bit too. If you’ve got robotaxi questions you’d like asked, or if you’re intrigued by specific edge cases like the ones described above, pass them along and I’ll see what we can uncover.
If you’d like to buy a Tesla Model 3, Model S, or Model X and get some free Supercharging miles, feel free to use my special, magical, unicorn-blessed referral code: https://ts.la/zachary63404. You can also get a $100 discount on Tesla solar with that code. There is currently no use for a referral code when putting down a reservation for a Cybertruck or Model Y.
Have a tip for CleanTechnica, want to advertise, or want to suggest a guest for our CleanTech Talk podcast? Contact us here.