I'm still really curious about how they plan on doing FSD with no Lidar, especially in the rain & snow. In theory, the physics calculations should be even better in the snow than human driving, especially for recovery in a spin or a slide, but who knows...it's all vaporware until it happens. I'm a huge, huge fan of Tesla & really love the concept of Autopilot, but tbh it feels like a really nice party trick right now. When I test-drove Kia's Stinger, the lane-centering technology coupled with the TACC was actually really fantastic...and, I mean, if you do a lot of highway driving, that or EyeSight pretty much fits the bill for the majority of highway driving.
It really comes down to how they can interpret the data that they're given. To be honest, I'm leaning toward a bit of over-selling on the part of Tesla. It sort of feels like that uneasiness that I get listening to Peter Molyneux talk about a game. I just have to take everything with a hefty grain of salt, and a dash of "we'll see...".
With that said, I still really like the Y. Really fun to drive, roomy, lots of cool features. Downside is fully-loaded, in my state, after taxes, it's $78k with everything. With zero down, that's $1,300 a month on a 60-month loan. That's an awful lot for what is basically an electric Nissan Versa lol. That's like a mortgage payment, haha! And that's why a $40k-base Cybertruck sounds so good!
On a recent Rich Rebuilds video, he was talking about aftermarket upgrades for the Model 3 that unlock more power for the Dual Motor (i.e. non-Performance) AWD models. They tested an unlocked AWD vehicle against a Performance Model 3, and the unlocked vehicle was as fast as if not slightly faster than the Performance version. I guess what I'm trying to get at is that you could potentially go with a non-Performance Model Y and eventually get an upgrade. The only thing I'd worry about is how persnickety Tesla would be about the upgrades. Would they do something like disable Supercharging if they determine that your car has been modified?
The more I try to use Autopilot, the more I regret paying for it. The only FSD feature that I use is the auto lane change since the Y lacks basic blind spot monitoring in the mirrors.
Hm, that is a good point. I wish they would've added a real blind spot indicator on the Model Y rather than continuing with the awkward red vehicle function on the display. It just comes across as a bit hackneyed due to needing to keep your eyes on the road and not down at your screen. Ultimately, the Model 3 and Model Y really just need a
simple HUD for the driver. I think if it had speed information and a MFD similar to what Ford uses, it would work fine. Ford's right screen can be swapped between things like Multimedia, which displays the current song, or Navigation, which displays the next turn information.
To your point, the nuances of driving are not something that the car can accurately measure. Maybe if the majority of cars were self driving and the roads where monitored in real-time, but a couple 100k cars trying to make it in chaos isn't realistic. I can't imagine it in a city when you throw in pedestrians, people on bikes and the random jackassery of people.
One thing that came to mind is just a sort of intuition that you have when driving. For example, you said that you live around Atlanta. Let's just say that if the Tesla tried to drive around the beltway at 50 MPH, I don't imagine that you'd be making very many friends. In that situation, the only thing the car could do is go on an understanding of "keep up with the flow of traffic" and speed up regardless of the posted speed limit.
During my trip, I dealt with some construction, and it made me wonder how the car would deal with things like lane shifting. Lane shifts don't always involve repainted lines, but are typically done with cones or barriers. How would the car deal with that? I could see it prioritizing cones over lines, but then again, it interprets quite a few things as cones that are definitely not cones. I've seen yellow fire hydrants classified as cones.
Tesla's AP is still learning. It has the best inputs/db to learn from and growing. Think of this as a 'kid' now. Eventually, it'll grow beyond any human capability.
I'm a bit curious to know how exactly they train the AutoPilot algorithm. Typically, AIs are trained on datasets that involve simulations using the aforementioned data that result in a pass/fail state. For example, Google's DeepMind trains to play against humans in StarCraft II by playing against itself, and if I recall, the winner's actions are compiled into the "hive mind" (very Zerg-like!). Do they use a driving simulation with AutoPilot (feeding it the applicable camera and distance data) to help train it?