Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register

Why AP 2.0 Won't Be Here Soon, and It Won't Be What You Think It Is

This site may earn commission on affiliate links.
Yeah, we noticed this with the 17.17.4 update that we received in mid-May. We were on a road trip on 5/12 and 5/13. The functionality wasn't there on 5/12 but after updating the firmware to 17.17.4, we noticed it on the drive back the next day. We didn't really notice it on regular cars but it was definitely happening when we'd pass semis.

I absolutely have to agree. Since the past half year or so, AP 1.0 and 2.0 both ‘lean’ away from semis when we pass them. I find it comforting, though it’s almost creepily human-like (the uncanny valley of self-driving? :p).
 
  • Helpful
Reactions: davidc18
Is there perhaps a disconnect WRT passing speed? I only recall a shift away from other vehicles a few times in the past, but only when passing speed was not too fast, where you were almost pacing the other vehicle. I don't recall ever seeing it when passing much slower traffic.
 
  • Helpful
Reactions: davidc18
I was one of a handful of people to "love" this original post when it came out, so I'm going to take a bit of a victory lap. My concern remains the same: We have AP2.0 hardware in cars that is getting rapidly outdated while we wait (likely years) for FSD. Is it even universally agreed that there's true feature parity with three year old AP1.0?

Assuming the software is ready within two years, that's a three year hardware gap. Look at the differences between an iPhoneX and a three year old iPhone 6. The X can do 3840x2160 4K at 60fps compared to 1920x1080. For FSD, that resolution is a dramatic difference. Geekbench single core is 4200 on the X vs 1382 on the 6 (over 3x the power). Multicore is 10100 vs 2300 on the 6 (over 4x the processing power). Graphics is 15000 vs 4200 (over 3x the speed).

These are the real world advances we're seeing in three years. Cameras with higher resolution, better dynamic range, and higher refresh rates. Camera quality is critical to Tesla's vision of FSD. CPU/GPU processing tripling. I suspect we'll see even bigger advances in the next two years with more custom hardware dedicated to FSD (Hello, Jim Keller?).

I'm just really concerned they pushed AP 2.0 (and even 2.5) out too soon before the software was ready. (I suspect this fell in their lap because of the falling out with MobileEye.) Granted, it may be a necessarily evil for shadow testing... and I'm sure we'll see some cool new features before FSD Level 5, but there's no way this is the final sensor suite and CPU system that will unlock the full potential of Level 5 at the time it's released.

Did we ever get the promised radar-assisted safety features (with whitelisting database of signs/bridges) promised for AEB on AP1.0 and 2.0? You know, the ones enable by updated firmware from Bosch?

That said, they've promised to upgrade all AP2.0 vehicles to be compatible with Level 5, so it's all good. I just hope that it includes all the latest advancements (including cameras, if necessary).
 
I was one of a handful of people to "love" this original post when it came out, so I'm going to take a bit of a victory lap. My concern remains the same: We have AP2.0 hardware in cars that is getting rapidly outdated while we wait (likely years) for FSD. Is it even universally agreed that there's true feature parity with three year old AP1.0?

Assuming the software is ready within two years, that's a three year hardware gap. Look at the differences between an iPhoneX and a three year old iPhone 6. The X can do 3840x2160 4K at 60fps compared to 1920x1080. For FSD, that resolution is a dramatic difference. Geekbench single core is 4200 on the X vs 1382 on the 6 (over 3x the power). Multicore is 10100 vs 2300 on the 6 (over 4x the processing power). Graphics is 15000 vs 4200 (over 3x the speed).

These are the real world advances we're seeing in three years. Cameras with higher resolution, better dynamic range, and higher refresh rates. Camera quality is critical to Tesla's vision of FSD. CPU/GPU processing tripling. I suspect we'll see even bigger advances in the next two years with more custom hardware dedicated to FSD (Hello, Jim Keller?).

I'm just really concerned they pushed AP 2.0 (and even 2.5) out too soon before the software was ready. (I suspect this fell in their lap because of the falling out with MobileEye.) Granted, it may be a necessarily evil for shadow testing... and I'm sure we'll see some cool new features before FSD Level 5, but there's no way this is the final sensor suite and CPU system that will unlock the full potential of Level 5 at the time it's released.

Did we ever get the promised radar-assisted safety features (with whitelisting database of signs/bridges) promised for AEB on AP1.0 and 2.0? You know, the ones enable by updated firmware from Bosch?

That said, they've promised to upgrade all AP2.0 vehicles to be compatible with Level 5, so it's all good. I just hope that it includes all the latest advancements (including cameras, if necessary).

FSD hardware software is complimentary. The more advanced the hardware, the less advanced the software has to be. So full FSD with AP2.0 HW is possible, but would demand software that way beyond what we have today (or be practical in any foreseeable future). It is possible that the SW limit will be lower with AP2.5/3.0 HW and the time will come sooner.

As long as we are eligible for no cost hardware upgrade to enable level 5 whenever that comes, that part of the HW promise holds true.
 
FSD hardware software is complimentary. The more advanced the hardware, the less advanced the software has to be. So full FSD with AP2.0 HW is possible, but would demand software that way beyond what we have today (or be practical in any foreseeable future). It is possible that the SW limit will be lower with AP2.5/3.0 HW and the time will come sooner.

As long as we are eligible for no cost hardware upgrade to enable level 5 whenever that comes, that part of the HW promise holds true.

True. One can argue you just need a camera on a pan/tilt gimble and good enough software for Level 5. That’s really all we use. (Any of our other senses are already in the car, like knowing its speed.)

To your point, the question becomes that sweet spot between sensors and hardware and software.