Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register
This site may earn commission on affiliate links.
Its got a very long way to go - but I'm happy to be part of the journey.
That's exactly how I look at it. I want it to succeed as soon as possible so my parents can stay mobile as their health and vision and reflexes inevitably deteriorate. Oh--and so my colleague can let his car drive him home safely from the bar.
 
The thing to remember is humans learn. Computers don’t. The ‘learning’ is done by the programming team adjusting algorithms.
Not strictly true. Even with classic procedural programming, it's possible to write programs that "learn", accumulating knowledge in a database and making decisions that change based on the accumulated knowledge. A classic example is a program that "learns" to play Tic-Tac-Toe, which they had many decades ago.

With neural networks, learning in baked into the architecture. The entire paradigm is designed to emulate organic brains' abilities to learn. Now, it's true that our _cars_ don't learn; the neural nets they run are trained (learning) on Tesla's "Dojo" supercomputer. The whole point of this architecture is that the programmers don't have to try to tweak algorithms to account for every possible road scenario; instead, the system is given a "goal" (get from point A to point B), and "constraints" (don't break traffic laws, don't physically run into things, etc.), and then trained with real-world and simulator data.

An oversimplification, largely due to the fact that I retired from programming before neural networks were more than a laboratory curiosity.
 
I think the average human driver is better in the first week of driving than FSD beta. I don’t remember any of my friends having a collision in their first week of driving. I think there’s a higner chance that FSD beta would have a collision.
chance. It’s odds and depends on the driver - a friend’s daughter was trying to park and drove the car over the curb, over the sidewalk and into the median of the street. Luckily she didn’t hit anyone.
 
chance. It’s odds and depends on the driver - a friend’s daughter was trying to park and drove the car over the curb, over the sidewalk and into the median of the street. Luckily she didn’t hit anyone.
In the first week of driving?
Anyway, my intent was to say it's all odds! I'm just saying I think the odds of a human driver having a collision in their first week are lower than the odds of FSD Beta having a collision in the same number of miles.
I realize this comparison isn't really fair because the human driver has someone in the passenger seat telling them what to do and I'm comparing them to unsupervised FSD Beta. I suspect it would be true for an unsupervised new driver as well though. There are always news stories of kids stealing their parents cars and driving improbably long distances...
 
  • Like
Reactions: sleepydoc
Continued distinct lack of smoothness overall of course whenever there are required speed changes. Definitely keeps you alert. I've noticed a definite "jump off the line" when lights turn green, or the car in front moves. Otherwise it seems to handle speed changes pretty well. When the limit increased on a road from 40 to 50, it gradually accelerates to 50. Similarly, when the limit changes from 50 down to 40, it slows down pretty smoothly for me. I've read some people say their cars brake pretty hard when the limit changes downwards. That has not been my experience.
This is an area that has been a constant concern for me through several versions of the beta, and not because the car slows too aggresively. Quite the opposite.

My experience has been on two lane highways that enter a small village or town. The speed limit drops from 55 to 45 to 35 over a short distance, but the Tesla responds much too slowly to these changes. If I'm going 60 in a 55 zone, I may find myself still in the mid 50s when it changes to 35, and I may be a half mile down the road before the car reaches a speed where I would not be ticketed if clocked by police.

It would help a lot if the car would read and respond to the advance warnings (35 zone ahead), but it ignores them and changes its behavior only when the speed limit actually changes, and then it takes much too long to reach an acceptable speed.

I deal with it by dialing back the speed myself when I know what's coming, but even then the Tesla responds slowly to these changes.
 
This is an area that has been a constant concern for me through several versions of the beta, and not because the car slows too aggresively. Quite the opposite.

My experience has been on two lane highways that enter a small village or town. The speed limit drops from 55 to 45 to 35 over a short distance, but the Tesla responds much too slowly to these changes. If I'm going 60 in a 55 zone, I may find myself still in the mid 50s when it changes to 35, and I may be a half mile down the road before the car reaches a speed where I would not be ticketed if clocked by police.

It would help a lot if the car would read and respond to the advance warnings (35 zone ahead), but it ignores them and changes its behavior only when the speed limit actually changes, and then it takes much too long to reach an acceptable speed.

I deal with it by dialing back the speed myself when I know what's coming, but even then the Tesla responds slowly to these changes.
It's a very subjective action. For some people, slowing more rapidly is uncomfortable and they misinterpret it as a phantom brake, until they realize the speed limit changed. For others it's too gradual, like yourself. It's hard to find a happy medium that will make everyone satisfied.
 
  • Funny
Reactions: AlanSubie4Life
If I'm going 60 in a 55 zone, I may find myself still in the mid 50s when it changes to 35, and I may be a half mile down the road before the car reaches a speed where I would not be ticketed if clocked by police.
This behavior was introduced in recent versions of FSD beta. It's obviously a bug caused by fixing something else that was broken. Even using the scroll wheel to reduce the maximum speed doesn't help. The problem is so obvious that I would be surprised if it isn't fixed in the next version of FSD beta
 
For some people, slowing more rapidly is uncomfortable and they misinterpret it as a phantom brake, until they realize the speed limit changed.
It’s really not though. It is nearly entirely the jerk that matters to basically everyone. Unfortunately that is not minimized. There’s no reason jerk cannot be kept very low when adjusting to changing speed limits. Unfortunately it often feels like a step function. Tesla knows how to do this. Just disable autopilot and as it blends in to manual control (with no input) you get a minimized jerk profile.

This isn’t really a personal preference thing. In control theory, the comfort of control systems is pretty well understood.
 
It's a very subjective action. For some people, slowing more rapidly is uncomfortable and they misinterpret it as a phantom brake, until they realize the speed limit changed. For others it's too gradual, like yourself. It's hard to find a happy medium that will make everyone satisfied.
I posted about this some time ago - it’s more than just ‘subjective.’ There's a road I routinely drive where the speed limit drops from 50 to 35 and it takes a full quarter mile for the car to slow down when using FSD. Of note, it takes longer than it does if using plain TACC and if you watch the energy bar, you can see the car is actually sending energy to the motors to maintain speed (the bar is not even neutral, it's black, on the right side.)

As others have pointed out, you are required to slow down before the speed limit sign, not start slowing down at the sign, and you can be pulled over and cited a block afterwards and taking a quarter mile to slow 15 MPH is completely unacceptable.

It's odd that the car will slow 10MPH quite aggressively for nothing, but when there's a legal reason to do so it couldn't care less.
 
Instead we are subjected to completely bizarre jerk profiles.

Ideally jerk would ramp slowly from zero to small negative values, then gradually decrease, pass through zero, and go to small positive values for a while. And then finally ramp smoothly to zero again.

Instead jerk rapidly goes negative, then around 20-30 mph goes significantly positive, with acceleration increasing from negative values to briefly near zero (so significant deceleration transitioning to nearly zero deceleration), before swinging rapidly negative again, and then finally positive again, followed apparently by a final negative step at the end. It’s very mysterious.

I should get an app but seems like the only ones that record these data are $23. I guess I could extract it from video footage but speedometer resolution might be too poor; I guess SMT might give fractional speed, not sure…
 
Gonna be tough to get zero intervention drives as long as spastic steering movements cause disengagements, and as long as I am constantly overriding the accelerator to achieve the desired deceleration (and acceleration) profile.
They can filter those out in their data. This could also be why you no problems with the attention monitoring, they're actually using the spastic movements to weed out the drivers who think it's acceptable.
 
  • Informative
Reactions: AlanSubie4Life