I think it's important to make a distinction between R&D and the realities that Tesla owners live with on a daily basis. Maybe
@Aikouka would like to chime in on his experiences with FSD.
It can be a bit hard to truly comment since Tesla has greatly shifted the goalposts as to what constitutes "FSD", and I'm talking about strictly from a feature set. This came with the switch from them offering Enhanced AutoPilot and Full-Self Driving to just offering AutoPilot and Full-Self Driving where a few features from the former (AutoPark, etc.) were transferred to the latter.
To be blunt, I pretty much never use the features. Going through the list of them from one of the articles that you linked:
- Navigate on Autopilot: My problem with this feature is that it arguably works great in a bubble, but given bad drivers and congestion, NoAP doesn't have the understanding to take things into account that we know how to handle. For example, what would happen if NoAP was stuck in a clump of cars and it needed to move over to take an exit?
- Auto Lane Change: Takes too long and doesn't consider making adjustments to timings based upon other vehicles. In other words, I think it's too rudimentary.
- Autopark: Autopark is something that has to be presented to you when going to park, and it barely ever presents the option to me to begin with. I think that's partly because Autopark seems to prefer spots where you're already bordered by two other vehicles and not empty spots in an open area. In other words, it likely isn't looking for lines so much as spaces between vehicles.
- Summon: It's pretty useless outside of simple scenarios. The problem is that parking lots can actually be really complex, which is often made worse by poor or very worn markings. For example, in a Walmart parking lot the other day, which uses directional aisles, it was very difficult to see the direction arrows at the end of the aisle. The car has no understanding of these directions; it's pretty much a shortest path algorithm.
- Traffic Light and Stop Sign Control: In the current version, there are some major limitations. For example, the traffic light detection routines have no ability to handle combined traffic lights. For example, you may have one traffic light with five lights: green arrow, solid green, yellow arrow, solid yellow, and red, but these only show up as a single light.
- Speed Sign Detection: This also has a major issue in that it has no comprehension beyond reading the numbers, and just imagine how annoying that would be if you routinely pass through a school zone. I've gone through school zones where it simply lists the time of day in which the reduced speed takes effect, but the car changes the speed regardless of what time it is. It also has an issue with sometimes getting confused as to which road you are on. For example, I was actually using TACC the other day, and all of a sudden, it started slowing down for no reason. I looked at the screen and it had me listed at 55 MPH where the speed limit in that area is 65 MPH. I've also had it drop cruise control before, which causes the car to immediately start braking.
The history of broken promises is pretty irritating to a lot of people. I have a number of friends who aren't too happy that Tesla wants them to shell out $1,000 for a computer that they were promised was FSD-capable
What's pretty awkward is that if you would have just paid for FSD from the start, that $1000 upgrade would have been completely free. My car was produced in 2018, and I received my V3 upgrade at no-cost due to having the FSD package. In fact, if you compare the costs now, I paid $5,000 for AP and $3,000 for FSD, which is $2,000 less than the $10,000 cost for FSD now. (AP is included now, so we can't factor in its cost.)
Frankly, if they're hellbent on charging for the computer upgrade, why not just mandate a minimum subscription duration? For example, if a user is not at V3 hardware, then they must stay subscribed for 6 months or pay $1000 for an upgrade. (At that rate, you'd be better off just doing the subscription as $1000 + $199/mo is the same as six months @ $199/mo.)
I like Tesla's quest, but living with their machines, build quality, service, beta software, etc. is another matter entirely, which is one of the reasons I don't currently have a Tesla in my garage. So future growth, living with a machine, and investment opportunities are all separate, compartmentalized things. I think the CNET video was a pretty fair assessment of "where things are at today, right now" in terms of the technology.
I find Tesla rather frustrating with their approach to automobile design. It's like... they're designed by engineers that never actually use the car. They think of an interesting idea that applies well to the idea of automation, but they completely ignore that people are still driving their car manually. I saw an article pop up recently about how they want to bring smart shifting to the Model 3 and Model Y, and frankly, I don't want it. As noted earlier, I've seen Tesla's decision-making fail while driving, and I don't want to trust it with such things.
I've seen other users talk about their upcoming FSD beta, which is supposed to be far, far closer to the idea of full self-driving. However, even Musk has admitted that it will not handle certain situations such as construction, obstructions, or inclement weather. Frankly, as to the latter, I have no idea what Tesla will do, because they're transitioning to camera-only detection, and whenever I'm driving in any decent amount of rain, I get warnings about cameras being obscured. Ultimately, their upcoming beta feels like more of a "happy path" version that is incapable of handling the more ideal situations. I've had to dodge a ladder in the middle of the road on my morning commute; what is the upcoming FSD going to do there?
Another facepalm-inducing incident was when Musk stated that he had no idea how hard Autopilot would be. From the perspective of a software engineer -- albeit, not in the AI space -- I was a bit taken aback by that remark. I mean... most of what we do is try to break a problem down into computationally capable components. Now, imagine if you were to take all of the driving-related decisions made during today's morning commute, and then consider that a computer needs to be able to take in the proper stimuli and make a comparable determination. Another aspect is that you're dealing with software that needs to be able to operate properly among erratic human drivers.
Engineering Explained posted a video recently talking about mileage and travel time reported in a Car & Driver issue across a wide array of EVs including Porsche, Audi, Nissan, Tesla, Ford, etc. One point that he mentioned that I strongly agree with is that Tesla doesn't do a good job of noting their car's range when operating at mostly higher speeds. For example, ICE-powered cars will list in-town and highway mileage, but with Tesla, you get an eMPG that doesn't represent highway mileage
at all. If I remember correctly, Engineering Explained put his estimated range (for a Model 3 Performance) at around 240 miles on the highway, and that seems to correlate with what I've seen. I've mentioned before how I took a round-trip that was ~130 miles in each direction, and I wouldn't have been able to get home even though I started at 95% charge. (My LR RWD has a higher range too.)
shop.tesla.com
Facebook
Twitter