-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Elon's Thanksgiving gift: Beta 5 coming in a few days
- Thread starter DanCar
- Start date
diplomat33
Average guy who loves autonomous vehicles
11.3 mile drive on FSD Beta 5 with some issues:
My summary of the issues:
- Disengagement when FSD was going to hit the curb on a left turn.
- Driver initiated auto lane change on highway to move over for emergency vehicle on the side of the road
- Car ping pongs between two lanes after a right turn
- 2 disengagements when car almost hits construction barrels. It seems FSD was confused about what was driveable space in the construction area.
My summary of the issues:
- Disengagement when FSD was going to hit the curb on a left turn.
- Driver initiated auto lane change on highway to move over for emergency vehicle on the side of the road
- Car ping pongs between two lanes after a right turn
- 2 disengagements when car almost hits construction barrels. It seems FSD was confused about what was driveable space in the construction area.
diplomat33
Average guy who loves autonomous vehicles
"Reverse Summon" or "Smart Park" instead of the crippled and utterly useless Auto Park would be beneficial to without the context of having city street AP.
Auto Park still only uses sonar only for parking hence non-existent recognition of parking spots besides parking between 2 vehicles in a fake detected parking spot that it very often fails to do as it only uses the geometry of the other cars to park exactly in between even though those cars could and likely be parked incorrect.
There hasn't been any sign or progress related to parking and Elon didn't hint about it too so I don't think we will see any improvements related to that in 2020.
EVNow
Well-Known Member
I've found that to be useful. It makes it easier to get out/in - rather than parking very close to one of the cars, even if perfectly aligned with the parking spot lines and in the middle of the parking spot.
powertoold
Active Member
And it's only the beginning.
I remember a few years ago. Elon said that the "silky smooth" AP update was coming soon, and 6 months later, it still wasn't very silky smooth. It seems FSD development is 100x faster than AP development back then.
Tesla Autopilot: Elon Musk says next update features new 'silky smooth' control algorithm - Electrek
EVNow
Well-Known Member
Yes that was my take as well. This suggests to me that FSD beta being in the holiday release is unlikely at this point, which is too bad.
Yea "end of year" is proof old Elon time is NOT dead.
I bet it is closer to March before it is released and that may be a little optimistic. While it is looking promising it is still far too dangerous to allow in the hands of less attentive and down right irresponsible (lets make a sleeping in the car video) people.
Also keep in mind that we are now in the SLOW holiday session and updates will be less often anyway.
I bet it is closer to March before it is released and that may be a little optimistic. While it is looking promising it is still far too dangerous to allow in the hands of less attentive and down right irresponsible (lets make a sleeping in the car video) people.Also keep in mind that we are now in the SLOW holiday session and updates will be less often anyway.
As of now Tesla calls you and goes through a verbal checklist of dos and don'ts before you are allowed to get it. Also it appears all participants are "hand" selected for approval. I bet you have to sign off on a specific NDA too.
Last edited:
powertoold
Active Member
If they're taking this much precaution for the beta, I have trouble seeing how they'd do a wide release without achieving at least human level+.
EVNow
Well-Known Member
I think it is as much about controlling the narrative as it is about safety.
Depends on what you mean by "Human level+".
To get a statistically significant human level + on accidents you have to run on FSD hundreds of millions of miles. That is not possible with just EAP, let alone current beta tester. You need a wider GA release to be able to even figure out whether it is human level +.
They are more likely to go with # of disengagements (per unit of time driven or miles) - and some other metrics (like achieving a particular pass rate in their test suit).
Last edited:
powertoold
Active Member
I don't think that's true. It'll be clear to us once the FSD beta gets to a "human" level or better. We'll be able to see it in the videos. Right now, it's very obviously not there yet.
You're talking about two different things. The hundreds of millions of miles may be required for wide robotaxi approval, but the testing can be done with a much smaller group. Plus, Tesla uses data from everyone for testing, not only those who activate FSD beta.
EVNow
Well-Known Member
Not talking about subjective observation - objectively to prove that something is Human level + needs a lot of miles. Checkout my older posts - where I've provided links to white papers on this. For eg. to compare fatality rates, you need Billions of miles.
powertoold
Active Member
I agree with you. I was more referring to the subjective observation, and yes, that's not a good way to measure FSD.
I was trying to point out that Tesla has ways to measure FSD safety and performance without the driver activating it or even having it. So Tesla doesn't necessarily need to widely release it to continue improving or test it. As such, I think it's not prudent to widely release it until they think it's human level+.
SunCharged
Member
Similar threads
