Elon: FSD Beta tweets

[jeremymc7]

I do drive in areas where the car requires me to take over or disengage. Roundabouts, bike lanes, bus lanes and the such cause me to take over quite often.

That said there was none of those on my recent drive with repeated nags.

So does the car keep an individual history and factor that into the nags. Or is it
More standardized across the group.

 

[Supcom]

I do drive in areas where the car requires me to take over or disengage. Roundabouts, bike lanes, bus lanes and the such cause me to take over quite often.

That said there was none of those on my recent drive with repeated nags.

So does the car keep an individual history and factor that into the nags. Or is it
More standardized across the group.

Probably a long-term moving average. One easy drive does not reduce the nags.

 

[clydeiii]

I mean AP has trained me to hold the wheel such that I don’t get nags ever (or very rarely). I’ll try to make it nag more tomorrow and report back!

Sure enough, ya'll are right! I let go of the wheel on the highway around 70mph and it took it about a minute before it wanted steering wheel torque! Prior to that was maybe 10-15s, but these are all estimates. Wish we knew the mathematical formula that determines when a nag occurs.

 

[dramsey]

Daytime, clear I assume? Both suburban and highway?

Yes to all three.

 

[EVNow]

I have noticed less nags for me on my daily route.

Same here. Far fewer nags.

 

[EVNow]

I mean AP has trained me to hold the wheel such that I don’t get nags ever (or very rarely). I’ll try to make it nag more tomorrow and report back!

In Soviet Russia Musk's America .... Neural Network trains you

 

[AlanSubie4Life]

Same here. Far fewer nags.

Yep.

On the freeway I can go for miles, at least 10 minutes, without nags. In general. Probably depends on conditions.

Today in medium traffic traveling at 70mph, I experimentally took my hands off the wheel (while leaving them on the wheel of course). Went for at least 10 minutes with my hands held that way, just a couple mm off of the wheel.

As soon as the car exited the freeway for a rather tight high speed right hander, the car asked for hands again.

If you want to avoid nags, just be sure to keep your hands on the wheel and eyes dead ahead. It’s not a good test if you take your hands off and put them in your lap - the car can easily tell. (It cannot see your hands or the wheel but it can definitely tell.) Also avoid surface streets.

I do feel like the first couple drives were worse but it may have been random variability.

 

[eli_]

As long as you look straight ahead at the roadway and your eyes aren't darting over to the screen or whatever else, you get almost a minute of no-nags on the highway.

 

[AlanSubie4Life]

As long as you look straight ahead at the roadway and your eyes aren't darting over to the screen or whatever else, you get almost a minute of no-nags on the highway.

More like ten minutes but yes. As long as you have you have fake on-the-wheel hands. Through an interchange and various splits and merges too! I am sure it can be less or more depending on confidence and conditions.

 

[EVNow]

I still think Tesla should release a device that goes on the wheel and detects the hand. I’ve no problem with having the hand on the wheel but getting nags while having hands on the wheel is not pleasant.

 

[willow_hiller]

I still think Tesla should release a device that goes on the wheel and detects the hand. I’ve no problem with having the hand on the wheel but getting nags while having hands on the wheel is not pleasant.

The most ironic thing for me is that I get nags more often with two hands on the wheel than when I just use one. The weight of each hand cancels the other out, and registers as no hands to FSD Beta.

 

[boonedocks]

The most ironic thing for me is that I get nags more often with two hands on the wheel than when I just use one. The weight of each hand cancels the other out, and registers as no hands to FSD Beta.

Glad I am not the only one. I get less nags with no hands on the wheel than woth one or two on the wheel.

So annoying because part of me says let loose and enjoy the nag free ride and the other part says hang on for Mr Toad’s Wild Ride!

 

[DanCar]

TeslaFi.com Firmware Tracker

11.3.6 is out.

https://twitter.com/x/status/1644830920170434560

 

[AlanSubie4Life]

https://twitter.com/x/status/1644659294153977858

 

[jeremymc7]

Just went from 11.3.4 to 11.3.6 in only a week.

Nice work Tesla.

It seeing any differences in release notes. Given the minimal point updates just bug fixes or tweaks.

 

[JHCCAZ]

Just went from 11.3.4 to 11.3.6 in only a week.

Nice work Tesla.

It seeing any differences in release notes. Given the minimal point updates just bug fixes or tweaks.

Usually the secondary point releases have very little or no specific documentation differences in the release notes. Probably we won't see much of that until 11.4.x.

When that comes, it would be nice to see some references to:

• City Streets improvements in turn lane recognition and reduction of uncertainty leading to appearing and disappearing turn flasher intent.
• Highway improvements in merge opportunity recognution.

Regarding 11.3.6, so far tonight I see only one published YouTube video, possibly revealing a somewhat less timid approach to stop signs.

 

[DanCar]

Tesla needs to do a better job predicting pedestrian intent.

https://twitter.com/x/status/1644942345383673856

Diffusion seems to be more compute-efficient than transformers for vision.
From 2019:

https://ieeexplore.ieee.org/ielaam/76/8786781/8425080-aam.pdf

Develop an effective variant for estimating
visibility statuses of objects while tracking them in videos. Dealing
with partial or full occlusions is a long standing problem in
computer vision but largely remains unsolved. In this work,
we cast the above problem as a Markov Decision Process
and develop a policy-based jump-diffusion method to jointly
track object locations in videos and estimate their visibility
statuses. Our method employs a set of jump dynamics to change
object’s visibility statuses and a set of diffusion dynamics to
track objects in videos. Different from traditional jump-diffusion
process that stochastically generates dynamics, we utilize deep
policy functions to determine the best dynamic for the present
state and learn the optimal policies using reinforcement learning
methods. Our method is capable of tracking objects with full or
partial occlusions in crowded scenes.

 

[EVNow]

Tesla needs to do a better job predicting pedestrian intent.

https://twitter.com/x/status/1644942345383673856

This shows why FSD is so cautious around VRU. It only has a vague idea about them. They just need to do what Kinect did a long time back !

We need to predict what pedestrians will do based on their behavior, including limb angle & direction of sight. FSD currently sees all pedestrians as cuboids, so is overly cautious.

 

[Supcom]

This shows why FSD is so cautious around VRU. It only has a vague idea about them. They just need to do what Kinect did a long time back !

I have plenty of issues around uncontrolled intersections with no pedestrians, or even cars, in the area.

 

[EVNow]

I have plenty of issues around uncontrolled intersections with no pedestrians, or even cars, in the area.

"uncontrolled" ?

As I wrote somewhere else ... FSD seems to have trouble making sure an intersection is empty. It takes more time to figure this out than if an intersection has vehicles.

Thats a different issue though. Here we are only talking about how FSD treats VRU.