Tesla Full Self Driving- Isn't

TL&DR: full self driving, isn't. It's at best a co-pilot, and one who has had a few too many drinks to be trustworthy.

We bought a Model Y dual motor long range, just about a year ago. And we LOVE the car. We have nearly 30,000 km on it already, and 92% of our kWh were recharged at home using Ontario's ultralow overnight rates, currently 2.8 cents Canadian per kWh plus 1.9 cents delivery. Unbelievable energy economy, and unbeatable GHG and toxic emission reductions (by 94% relative to our gasoline Prius), in a car with considerable comfort and unbelievable performance. And the supercharging network makes it practical to drive basically anywhere we've wanted to go. We're never going back to an IC engine car- by never, I mean not ever again.

Our only quibbles so far are with the software and sensor package. Like the irritating fact that the windshield wipers false trigger, wiping the windshield during days without a cloud in the sky. While we've seen this problem mitigated to some degree via software updates, it still happens. And when the driving assist features are engaged, there's no way to keep the wipers off. That this simple flaw persists is a warning that the software isn't really up to snuff. Who knows, maybe they'll fix it one day, but online, people have been complaining about the false triggering for several years.

Our car also lacks the ultrasonic sensors present on prior models. It uses the cameras only for parking assistance. To say that the camera system is unreliable at judging distance to damaging objects is an understatement. In fact it even says so on the car's display- "Parking assist features may be degraded". Yeah, that's no lie.

Tesla offers two levels of basic driving assist features: "Autopilot", which consists of basically adaptive cruise control (a feature available on cars 1/2 the cost of the Model Y) and some emergency steering interventions, and "Autosteer", a product seemingly perpetually in "beta" which provides basic lane maintenance. As far as we can tell, you need to spring thousands more for "advanced Autopilot" to get the features of automatic parking and lane changes. As long as you are in the middle or left lane on a divided highway, the adaptive cruise control, automatic safety interventions and lane maintenance features work reasonably well, though "phantom braking" is still a problem. A bumper sticker saying "Caution- this car brakes for ghosts!" would be appropriate. But the safety benefit of having these features versus driving without them is not lost on me. I have no doubt that these features will save lives.

Full Self Driving Trial

We were offered the 1 month free trial of Supervised Full Self Driving, and while the performance of the basic package was craptastic enough to make me quite sure I wasn't going to pay the $16,000 adder (now reduced to $11,000 or $99/month) for this feature, I figured we'd try it out.

And the results were a mixture of pleasant surprise and ultimate disappointment.

The notion that this software implementation is anything close to "full self driving" is just wrong. At best, this might be referred to as "co-pilot", and for sure, the co-pilot has been drinking.

The car handled some things extremely well in a way that I was not expecting. We encountered a situation where we were on a 2 lane residential street, with the lane ahead blocked by a parked concrete truck. The FSD system managed to get us around the concrete truck, and then carefully poked its nose out to see past the building on the left to know when it was safe to make the required right hand turn. It did that without needing to touch the steering wheel.

It also very aggressively and accurately managed lane changes on the highway. It does have a tendency to hog the left lane, to the irritation of drivers behind- but I guess it might mitigate guilt for the driver a bit. It also signals lane changes before vehicle being overtaken is out of the blindspot, which might cause defensive drivers to brake. But this is all minor quibbling- it handles divided highway driving very well, and for some people that's all it would need to do to be worth the money.

But to say that it's anything you could trust for trips door to door without a full time driver being there, aware and ready to intervene at a moment's notice, is definitely untrue. We'd have been in the ditch or in collisions many times had a driver not been ready to intervene.

The navigation system chose to navigate down gravel roads in the country, for instance- which is no problem. But its lane maintenance system isn't capable of handling roads without lane markings properly, as evident by several occasions where I had to intervene to keep us out of the steep ditch. It doesn't seem to realize that these roads aren't really a full 2 lanes wide. It similarly was very nervous making turns on unmarked roads, to the point that intervention by applying the accelerator was nearly always required to avoid getting a "honk" from anyone behind you.

Navigation also (needlessly) suggested a route requiring a U turn, but then was unable to complete the U turn without assistance. With no driver, this would have been a disaster.

The software was also ridiculously "nervous" the moment that weather conditions weren't perfectly clear. And when we really could have used driver assistance- when snow made visibility poor- the reliance on cameras alone basically made our electronic co-pilot as blind as we were. The original car models had radar and lidar (pulsed laser range detection) which would have been a major help.

We didn't do the trial in the harshest part of winter, but given the fact that the cameras were occluded every time we set out for a trip in winter, and needed to be manually cleaned off to be of any use at all, I would imagine that winter performance would be less than stellar.

It also wouldn't accept route changes via use of the turn signal, about 4 out of every 5 times we tried. This worked well enough in our residential neighbourhood, but didn't work at all out on the regional roads near our farm.

The notion that this is "full self driving" is just false advertising. And the notion that software anything like this is ready to navigate a pilotless "robotaxi" is pure nonsense.

To be fair, the trial was really about gathering huge amounts of data for Tesla to use in training the "self driving" system. They provided a feature where you could record a message to explain why you'd taken over from the self driving feature, which would help them to examine the failures and upgrade the software. So offering the free trial was a good business decision on Tesla's part, and may greatly improve upcoming versions of the software.

As to value for money: to us, this feature isn't worth even a fraction of the cost it's being offered for. Perhaps a 1 month subscription during a long road trip lasting a couple weeks would be something to splurge on, but paying $11k for this feature the way it is, is just really bad value for money in my view. Others are of course free to differ- perhaps money comes more easily to them than to me.

So if your investment in Tesla stock is predicated on the notion of robotaxis, you may want to reconsider. My bet is that this feat is well out of Tesla's reach, unless they have something far more powerful than they're selling for $11k a pop to Model Y owners.