Tesla Software updates - Australia

[paulp]

Not sure about the 3 but with s/x the steering wheel buttons can answer the phone

 

[paulp]

It is an interesting thought. I think that the AI will need to predict behaviour of other road users from fairly subtle cues rather than just respond to the immediately observable events.

I suppose once the density of AI driven reaches a sufficient threshold, especially if there is vehicle to vehicle communication then that issue will disappear.

Things will be some time away. My main interest is to have some increasingly vigilant copilot with the ability to intervene as needed to help keep me out of trouble.

There are some interesting papers on trust in AI beginning to be written.

I hope its a long way off. Driving my tesla is way more fun for me then being a passenger.

 

[SandyP]

New OTA update with side cameras – expected soon (with 2020.24.5 already in some early adapter vehicles)

Now you can see the side rear view in the Big screen. And FSD preview is coming in China btw. : teslamotors

Tesla releases 2020.20.16 software update packed full with highly requested features - Drive Tesla Canada

Update: It appears EAP owners in the US are seeing this update too, but it is numbered 2020.24.5

Backup Camera Improvements
Dashcam Viewer Improvements
Walk-Away Door Lock Improvements
On-Route Battery Warmup
Cabin Camera

 

[GrimRe]

It starts with knowing how to have a rational conversation without making yourself feel superior.

The training thing is marketting talk.

Take your own advice mate.

Its all he knows. Its what those with an inferiority complex do. Makes them feel better.

Nothing to me screams inferiority complex more than being dismissive of things you don't understand.

I would agree with the argument that Tesla's positioning of the FSD option over the years leaves a lot to be desired (some cars are 5 years old at this point which have been promised FSD). However to say that Tesla do not perform meaningful ML training is patently wrong.

The cut in detector is a great example of this. It used to be hand written code and quiet jerky on the road. When Karpathy took over he designed a self supervised training pattern whereby they would collect all examples of cars transitioning across lane lines and then rewind back a few seconds to see the positional attributes of that vehicle prior to the cut in. This means the NN model in charge of predicting cut ins is practically superhuman. It can pick up small changes in other car's yaw, speed and trajectory that would indicate an impending lane change well before a human would predict it. Thus the car now smoothly adjusts its own speed to prevent a collision or hard braking.

FYI - Another good option when talking about AI/ML is to watch The Good Place and how they handle the trolley problem in the 2nd series (which makes me feel we'll always need the option to intervene). I'm just glad I program software that has little chance of killing people.

The thing I've never understood about the "trolley" problem is that it assumes the vehicle is somehow instantaneously transported into a situation where the only two options are kill through action or inaction. In the real world this would never happen. How many times has a human been in this situation? Now imagine an autonomous car that is ten times more aware of its surroundings and react 100 times faster than a human is it really going to realistically ever going to get into the "trolley" situation?

Take a more realistic example of say an undivided highway and an oncoming car suddenly veers over onto your side of the road with less than 1 second before head on impact at 100 km/h. Human has zero chance of even "thinking" let alone deciding on which course of action will cause least harm. An autonomous car on the other hand firstly has 1,000ms of time to think and can process about 30 predictions in that time frame. Secondly, same as with the cut in example above, the veering car will in all likelihood have displayed some leading indicators that would infer its impending actions (slight lane ping, ponging, wheels at a slightly different angle etc) probably giving the car 5 seconds in time to respond.

 

[moa999]

Update: It appears EAP owners in the US are seeing this update too, but it is numbered 2020.24.5

Still hasn't shown up in TeslaFi

 

[SandyP]

Still hasn't shown up in TeslaFi

Not yet in wide rollout, as it was first seen less than 24 hours ago.
2020.24.5 (Software Update) | Teslascope

 

[SaturnV]

Take your own advice mate.



Nothing to me screams inferiority complex more than being dismissive of things you don't understand.

I would agree with the argument that Tesla's positioning of the FSD option over the years leaves a lot to be desired (some cars are 5 years old at this point which have been promised FSD). However to say that Tesla do not perform meaningful ML training is patently wrong.

The cut in detector is a great example of this. It used to be hand written code and quiet jerky on the road. When Karpathy took over he designed a self supervised training pattern whereby they would collect all examples of cars transitioning across lane lines and then rewind back a few seconds to see the positional attributes of that vehicle prior to the cut in. This means the NN model in charge of predicting cut ins is practically superhuman. It can pick up small changes in other car's yaw, speed and trajectory that would indicate an impending lane change well before a human would predict it. Thus the car now smoothly adjusts its own speed to prevent a collision or hard braking.



The thing I've never understood about the "trolley" problem is that it assumes the vehicle is somehow instantaneously transported into a situation where the only two options are kill through action or inaction. In the real world this would never happen. How many times has a human been in this situation? Now imagine an autonomous car that is ten times more aware of its surroundings and react 100 times faster than a human is it really going to realistically ever going to get into the "trolley" situation?

Take a more realistic example of say an undivided highway and an oncoming car suddenly veers over onto your side of the road with less than 1 second before head on impact at 100 km/h. Human has zero chance of even "thinking" let alone deciding on which course of action will cause least harm. An autonomous car on the other hand firstly has 1,000ms of time to think and can process about 30 predictions in that time frame. Secondly, same as with the cut in example above, the veering car will in all likelihood have displayed some leading indicators that would infer its impending actions (slight lane ping, ponging, wheels at a slightly different angle etc) probably giving the car 5 seconds in time to respond.

Wow, this sounds great! When is this update release coming?

 

[GrimRe]

Wow, this sounds great! When is this update release coming?

I have no idea. I was responding on a hypothetical scenario with an equally hypothetical one.

 

[SaturnV]

I have no idea. I was responding on a hypothetical scenario with an equally hypothetical one.

I should have /S in my previous. So many, (and every edge case) hypothetical needs to be covered, let alone every day driving scenarios for "FSD" to be truly functional. It will come....one year.

 

[meloccom]

A lot of people seem to think that FSD was promised for Autopilot since the beginning in 2013. That is not my recollection and is backed up by the Wikipedia entry that quotes:
“Elon Musk first discussed the Autopilot system publicly in 2013, noting "Autopilot is a good thing to have in planes, and we should have it in cars.".[7]

All Tesla cars manufactured between September 2014 and October 2016 had the initial hardware (HW1) that supported Autopilot.[8]On October 9, 2014, Tesla offered customers the ability to pre-purchase Autopilot capability within a "Tech Package" option. At that time Tesla stated Autopilot would include semi-autonomous drive and parking capabilities,[9][10][11] and was not designed for self-driving.[12]“

It is my recollection that at the time Elon said that cars capable of full self driving would be produced in about 3 years from that time so 2016-2017 which closely aligned with when HW2 was introduced.
I can say that I purchased my 2014 Model S understanding that it would never be Full Self Driving.

 

[SaturnV]

I can say that I purchased my 2019 Model 3 with the expectation that it would never be Full Self Driving.

 

[paulp]

A lot of people seem to think that FSD was promised for Autopilot since the beginning in 2013. That is not my recollection and is backed up by the Wikipedia entry that quotes:
“Elon Musk first discussed the Autopilot system publicly in 2013, noting "Autopilot is a good thing to have in planes, and we should have it in cars.".[7]

All Tesla cars manufactured between September 2014 and October 2016 had the initial hardware (HW1) that supported Autopilot.[8]On October 9, 2014, Tesla offered customers the ability to pre-purchase Autopilot capability within a "Tech Package" option. At that time Tesla stated Autopilot would include semi-autonomous drive and parking capabilities,[9][10][11] and was not designed for self-driving.[12]“

It is my recollection that at the time Elon said that cars capable of full self driving would be produced in about 3 years from that time so 2016-2017 which closely aligned with when HW2 was introduced.
I can say that I purchased my 2014 Model S understanding that it would never be Full Self Driving.

Here is the first online page offering autopilot and its capability in 2015 on version 1 hardware. In the 2014 online pages tesla offered a tech package but no mention of autopilot.

 

[paulp]

Then in 2016 they added to the wording still the original front nose shape. Everyone can decide for themself whether these features were delivered on that model;

“Autopilot combines a forward looking camera, radar, and 360 degree sonar sensors with real time traffic updates to automatically drive Model S on the open road and in dense stop and go traffic. Changing lanes becomes as simple as a tap of the turn signal. When you arrive at your destination, Model S will both detect
a parking spot and automatically park itself. Standard equipment safety features are constantly monitoring stop signs, traffic signals and pedestrians, as well as for unintentional lane changes.”

 

[GrimRe]

Then in 2016 they added to the wording still the original front nose shape. Everyone can decide for themself whether these features were delivered on that model;

“Autopilot combines a forward looking camera, radar, and 360 degree sonar sensors with real time traffic updates to automatically drive Model S on the open road and in dense stop and go traffic. Changing lanes becomes as simple as a tap of the turn signal. When you arrive at your destination, Model S will both detect
a parking spot and automatically park itself. Standard equipment safety features are constantly monitoring stop signs, traffic signals and pedestrians, as well as for unintentional lane changes.”

reading between the lines a bit, they were probably promised a lot from MobilEye at the time. Elon probably got pissed at the lack of progress and decided to go it himself. That’s when they switched to the internal AutoPilot program and used NVIDIA gpus in the Cars. A few years into that program they realised they would run out of Computational headroom and decided they would need to build their own inference optimised chips.

Now almost 8 years later I think they finally have the right scaffolding in place to execute on the FSD vision. I would guess that it might require HW4 but nonetheless paying $8.5k now will be a bargain come 5 years time (even with HW3). The car itself will obviously depreciate 30-50% but the FSD option will appreciate 2x in that time, more than making up the difference. This is assuming only linear progression. It seems to be entering the logistical growth curve phase which is very interesting...

it does suck for those early adopters whose cars will be too old to really see FSD applied but such is life when you are on the first wave of an adoption curve — can’t have it both ways.

 

[HockeyMonkey]

Really looking forward to 2020.24.5.1 which will continue through green lights if there is a lead vehicle. Losing this capability definitely felt Iike a regression.

2020.24.5.1 (Software Update) | Teslascope

 

[Brissienew]

paying $8.5k now will be a bargain come 5 years time (even with HW3). The car itself will obviously depreciate 30-50% but the FSD option will appreciate 2x in that time, more than making up the difference.

Seriously doubt that.
Never mind the misleadingly named FSD, I just want my auto wipers to work properly, like how they do in most other cars. They still go crazy in light rain after the most recent update.

 

[Hairyman]

Seriously doubt that.
Never mind the misleadingly named FSD, I just want my auto wipers to work properly, like how they do in most other cars. They still go crazy in light rain after the most recent update.

I’m really interested in the variability of the function of the auto wipers. We all have the same software and hardware, but I’m not seeing the behaviour you are seeing. I’m wondering what the trigger is.

 

[Brissienew]

. I’m wondering what the trigger is.

In my case pretty much any amount of rain. Others have described the same experience. Unfortunately, the auto setting’s speed is not adjustable, unlike in most other cars.

 

[RareEarth]

Very conscious of wiper operation since reading other people's experiences here. Last night had to go out, light rain falling as I backed out from under the carport, first wipe after about 10 seconds. Stayed intermittent until rain intensified as I went from a 50 zone to 80 then 100, wipers responded accordingly until going pretty fast. Only had to intervene once for the whole 50km trip.
I have had to get the wipers adjusted previously as the passenger side wiper was stopping on top of the camera, leaving water and debris right in front of the lens. This was causing havoc with autopilot. Now it goes a little way past the camera.

 

[malc01]

Something observed and welcomed on version 2020.20.13 is loss of connection issues seem to be fixed. Great for Spotify users.
LTE would drop out in an underground car park as an example and remain unconnected even 20 minutes or more after going back to street level again. Now connection resumes in about a minute.