FSD Beta 10.69

[AlanSubie4Life]

I made a video to purge some of my frustration. It didn’t help. This video ended up truncated because the GoPro crashed (all GoPros suck AFAICT). Meant that you will miss some more of my witty repartee. On the upside it kept the video short.

You can see we have a LONG LONG way to go. If you can’t see the mistakes there is nothing more I can do for you. The evidence is clear.

 

[sleepydoc]

I really disagree with this assessment. As Tesla’s release notes show, the correct driving behavior is objective and quantifiable.

Some people drive wrong, no question about it. And everyone makes high jerk moves (which are typically related to mistakes). But let’s be clear that correct behavior is measurable and knowable. And hopefully Tesla can achieve it.

This isn’t some sort of impossible goal. We’ll see massive improvement in the coming years, I am quite sure. Nothing unreasonable here. Just would like to see it ASAP and one way to provide feedback is to disengage.

to be sure, some behavior is flat out wrong - wandering outside of your lane, stopping 30 feet before the stop sign, etc. Other behavior is much more subjective - rate of acceleration, rate of braking, following distance, when to turn on a turn signal before turning, etc.

Let's take a lane change as an example - MN requires you to signal 100 feet before chaining lanes. Ok, signal on, 100 feet later you change lanes. Do you slowly drift over taking a quarter mile? Do you jerk over like a slalom skier? Something in-between?

How about following distance - what would be considered tailgating here in MN is SOP in New Jersey and if you leave more room people get annoyed.

A while ago you were complaining about how slowly FSD accelerated. What is the 'right' amount of acceleration?

Tell me, what is 'correct.' I guarantee you I can find someone to disagree with you no matter what you answer.

 

[AlanSubie4Life]

What is the 'right' amount of acceleration?

About 0.25g in good conditions would be fine. Please keep the jerk low. (This can also be quantified - roll it in over 1 second or so.)
Configurable and scalable would be fine; not too hard once you can actually nail the profile, which is the problem.

 

[sleepydoc]

About 0.25g in good conditions would be fine. Please keep the jerk low. (This can also be quantified - roll it in over 1 second or so.)
Configurable and scalable would be fine; not too hard once you can actually nail the profile, which is the problem.

and there lies the problem...

 

[AlanSubie4Life]

and there lies the problem...

I’m not sure what you mean. Clearly quantified. You asked me to provide a value. I don’t think adding significant figures is going to be noticeable.
I am not sure how it can be more clear that this is something that is well understood and not subject to personal preference, beyond simple scaling.

You can just measure it and then we know the answer.

In the video above I got to 40mph in 7 seconds, on a turn. That is 0.26g on average. Totally normal driving. Estimate not far off. Can be scaled. Make it a slider. Some like more pep some like less.

Notably this is not the problem right now. Gotta nail the profile first! Just can’t do it. 😢

 

[Yelobird]

About 0.25g in good conditions would be fine. Please keep the jerk low. (This can also be quantified - roll it in over 1 second or so.)
Configurable and scalable would be fine; not too hard once you can actually nail the profile, which is the problem.

By my made up calculations I believe it should be .2745g and roll in is Proven fact that I made up to validate my statement at 1.218 seconds. Luckily you don’t have to show your math or facts on a forum!

 

[AlanSubie4Life]

By my made up calculations I believe it should be .2745g and roll in is Proven fact that I made up to validate my statement at 1.218 seconds. Luckily you don’t have to show your math or facts on a forum!

They’re not made up. See above. Simple division!

As we know, a common limit for insurance purposes for longitudinal control is about 0.3g.

Not sure why these numbers are claimed to be made up when the math is all right there.

Anyway these numbers aren’t the real issue; they are a distraction.

We all know the issue.

 

[Phlier]

Off topic, for a pilot, @Phlier , I figured you should be able to spell "Flier" correctly

I was going to reply to this, but I got a fone call.

 

[Phlier]

When I pull into my garage, on the driver side of my car there is one 1979 Yamaha SRX 440 and one 2012 Yamaha Apex XTX. They show up as motorcycles. Pretty fitting, as snowmobiles do look kinda like motorbikes. My 1 series (vomit, time for a 3039R) John Deere tractor with a blower displays as an suv. With the blower off, it displays as a car, so it definitely knows length (as the blower adds like 3ft to the overall length of the machine) although not represented “correctly”. It knows cars are shorter than most suv’s.

This reminded me to post about a recent experience...

I have a sink in my garage that sits directly in front of where I park Bluebird.

Remember when Elon carried a sink into Twitter's HQ? The man must have some sort of a sink fetish going on, because guess what my car shows the sink as?

Other guys that have sinks within detection range of your car in your garage, please chime in here.

 

[nvx1977]

I was going to reply to this, but I got a fone call.

tbh, before I knew you phlew planes, I was mentally pronouncing your handle as "flee-er" instead of "fly-er"

 

[nvx1977]

I made a video to purge some of my frustration. It didn’t help. This video ended up truncated because the GoPro crashed (all GoPros suck AFAICT). Meant that you will miss some more of my witty repartee. On the upside it kept the video short.

You can see we have a LONG LONG way to go. If you can’t see the mistakes there is nothing more I can do for you. The evidence is clear.

Thanks for posting your vid. I actually don't disagree with what you are labeling as mistakes, though I'm not sure how many others would be so strict for some of them. But this has always been true even for the AP1 days. Our expectations for how "good" it should be was all over the place. Some people called it garbage and refused to use it while others considered it heavenly.

Speed limits are also defined based on known rules, so I also agree there could exist an objective standard of what defines perfect driving (the center of the bell curve) based on safety, efficiency, and comfort, and us humans can fall anywhere along the entire bell but ideally near the center. This is actually how I think about driving efficiently when I'm really trying: minimizing my deviation from the most efficient speed/accel/decel possible for a given situation**. I can't wait for FSDb to attempt to do this as well; it's just nowhere ready to drive at this level of proficiency yet.

** in many situations, it's not possible to be maximally efficient. For example, the most efficient acceleration is the slowest acceleration. It's a limit that cannot be physically achieved; at the limit, you're literally not accelerating. This would seriously piss off everyone behind you (and would mean you being unable to stop for a sign or light in the deceleration case). So we have to pick an accel/decel that is safe and reasonable. The "objective standard" would then be the lowest accel/decel that maintains safety and conforms with nearby driver expectation. Same with speed. The ideal speed is one where momentum does the vast majority of the work while wind resistance is minimized. That speed is somewhere just below 30mph based on TY vids from early hypermiling attempts with Model 3. But going under 30mph in a 45mph zone would also piss off other cars. So we have to deviate from the efficiency ideals to satisfy the safety and local driving culture aspects.

From this perspective, I can see why people disagree that there is some objective standard of perfect driving. But it would be nice for the car to take all of this stuff into consideration (safety, comfort, efficiency) and attempt to hit a theoretical standard where those three things are as optimized as possible, where safety comes first.

 

[aronth5]

I’m not sure what you mean. Clearly quantified. You asked me to provide a value. I don’t think adding significant figures is going to be noticeable.
I am not sure how it can be more clear that this is something that is well understood and not subject to personal preference, beyond simple scaling.

You can just measure it and then we know the answer.

In the video above I got to 40mph in 7 seconds, on a turn. That is 0.26g on average. Totally normal driving. Estimate not far off. Can be scaled. Make it a slider. Some like more pep some like less.

Notably this is not the problem right now. Gotta nail the profile first! Just can’t do it. 😢

Example:
I want FSD to take my elderly parent to the doctor. Safely without affecting the safety of other cars. Nothing more nothing less. I also know if I asked 5 different friends to drive my parent to the doctor they would all drive very differently. Would I care, no.
Many of the video "mistake" comments just aren't important to this goal even if they are mistakes which isn't clear. You're gotten to far down into the weeds and FSD will likely never meet your expectations. Don't let perfect (whatever that actually is) be the enemy of good.

 

[ACPark]

Was it flickering RED GREEN or just RED and NOT RED? I’ve had this issue for well over six months, I used to be able to reproduce it at specific lights and intersections. Now, it’s just random which of course keeps one on their toes.

It’s really the type of thing that VERY much concerns me for a “wider roll out” because I doubt that every person who has paid for FSD is going to be able to react as quickly when their car decides to just proceed into a red light indicated intersection. Left side t-boned would be the outcome which is one of the worst outcomes. TWT

It was red green flickering back and forth rapidly. Same issue at both intersections.

I haven't see it since.

 

[Rikster]

Example:
I want FSD to take my elderly parent to the doctor. Safely without affecting the safety of other cars. Nothing more nothing less. I also know if I asked 5 different friends to drive my parent to the doctor they would all drive very differently. Would I care, no.
Many of the video "mistake" comments just aren't important to this goal even if they are mistakes which isn't clear. You're gotten to far down into the weeds and FSD will likely never meet your expectations. Don't let perfect (whatever that actually is) be the enemy of good.

I agree with this assessment. Many of the things that were labeled as "mistakes" in the video were simply personal preferences (acceleration rate, slowing rate, etc.). I think once FSD achieves consistently "good" driving (safe, relatively comfortable, somewhat efficient), hopefully there can be some customizable features for FSD that will allow an individual driver's preferences to be implemented by the car (within certain set parameters to ensure safety).

Some of my drives with FSDb have nearly gotten to the "good" status, and I can definitely imagine that FSD will get better on its consistency with future updates.

 

[cbdream99]

It was red green flickering back and forth rapidly. Same issue at both intersections.

I haven't see it since.

I have seen strange behavior, 3+1 solid red lights in the front but one is blinking in the visualization. Also, I found it runs a yellow light (and therefore red) every single time even if it can stop in time, this is behavior since 10.69.2.4 and 3.1 is the same. It's kind of stupid, in normal traffic, it's over-cautious tapping on the brake unnecessarily (ease of throttle is sufficient), but seeing a yellow light but making no attempt to stop.

 

[Phlier]

Also, I found it runs a yellow light (and therefore red) every single time even if it can stop in time, this is behavior since 10.69.2.4 and 3.1

Same.

but seeing a yellow light but making no attempt to stop.

They had to program *some* California into it.

It was red green flickering back and forth rapidly. Same issue at both intersections.

I can repeat this exact behavior at four different traffic lights.

If only there was some way I could report it.

 

[Ramphex]

I made a video to purge some of my frustration. It didn’t help. This video ended up truncated because the GoPro crashed (all GoPros suck AFAICT). Meant that you will miss some more of my witty repartee. On the upside it kept the video short.

You can see we have a LONG LONG way to go. If you can’t see the mistakes there is nothing more I can do for you. The evidence is clear.

What model SD cars are you using? I’ve had 3 GoPro versions and currently own 2, only issues I had were at first when I used cheap low speed SD cards, then the lesson was learned, and no issues since.

 

[nvx1977]

What model SD cars are you using? I’ve had 3 GoPro versions and currently own 2, only issues I had were at first when I used cheap low speed SD cards, then the lesson was learned, and no issues since.

sounds just like all the issues we had with dashcam recordings before Tesla started issuing a USB stick standard with the car.

 

[AlanSubie4Life]

What model SD cars are you using? I’ve had 3 GoPro versions and currently own 2, only issues I had were at first when I used cheap low speed SD cards, then the lesson was learned, and no issues since.

SanDisk 256GB Extreme microSDXC UHS-I Memory Card with Adapter - Up to 160MB/s, C10, U3, V30, 4K, A2, Micro SD - SDSQXA1-256G-GN6MA

Bigger problem I have is actually overheating. Usually the card is not corrupted. I could probably wipe it and that issue might improve. Or get a new card, who knows. I haven’t heard of a single person with a GoPro that has not overheated. (Obviously this includes on very hot days! Definitely it should not overheat then.). Overheating is not a problem right now.

 

[kabin]

I made a video to purge some of my frustration. It didn’t help. This video ended up truncated because the GoPro crashed (all GoPros suck AFAICT). Meant that you will miss some more of my witty repartee. On the upside it kept the video short.

You can see we have a LONG LONG way to go. If you can’t see the mistakes there is nothing more I can do for you. The evidence is clear.

Even if a rare FSDb driver is comfortable with all those mistakes it wears on passengers. It might be similar to comparing Disney's worn-out 1959 Matterhorn ride to one of their newest roller coaster rides. The difference is night and day.