Speculation: HW4 at Tesla AI Day 2 and already being installed

[n.one.one]

Interesting speculation from Dave Lee regarding HW4 at Tesla AI Day 2 and the possibility it is already being installed in production vehicles.
The follow-up comments in the Twitter thread add additional context and makes this very interesting.

https://twitter.com/x/status/1575536233870041089

 

[MY1stEV]

Tesla demoing HW4 tonight at AI Day is plausible, but I'm not so sure I buy-in on the speculation that recent cars already have HW4 installed. If I remember correctly, Elon said that HW4 will debut on the Cybertruck.

 

[diplomat33]

Dave Lee was completely wrong. There was nothing on HW4 at AI Day 2.

 

[aronth5]

Dave Lee was completely wrong. There was nothing on HW4 at AI Day 2.

Is there a difference between "wrong" and 'completely wrong"?
Dave only said "I could be wrong but decently high likelihood" so gave himself some wiggle room.

He wasn't alone though in thinking HW4 would be mentioned.

 

[Daniel in SD]

Elon said the robotaxi model would be shown at AI day but I guess it wasn’t as far along as the bot. I’m guessing HW4 will be shown once they figure out what hardware is required for an actual robotaxi.

 

[Krash]

…I’m guessing HW4 will be shown once they figure out what hardware is required for an actual robotaxi.

I’m guessing HW7 will be shown once they figure out what hardware is required for an actual robotaxi.

 

[JulienW]

I’m guessing HW7 will be shown once they figure out what hardware is required for an actual robotaxi.

HW6.9 will be unveiled on April 20.

 

[diplomat33]

Elon said the robotaxi model would be shown at AI day but I guess it wasn’t as far along as the bot. I’m guessing HW4 will be shown once they figure out what hardware is required for an actual robotaxi.

That does not bode well for Tesla's robotaxi efforts.

 

[Daniel in SD]

That does not bode well for Tesla's robotaxi efforts.

Or it means they're really trying to finalize a product (versus showing something in its infancy like the bot).

Really I don't see any point in releasing HW4 until they can be sure that it will allow driverless operation. Of course I've always been a huge skeptic of "door to door" supervised automated driving. I think you actually need a certain error rate to keep people attentive (or maybe some limitations that require frequent driver input).

 

[aronth5]

Elon said the robotaxi model would be shown at AI day but I guess it wasn’t as far along as the bot. I’m guessing HW4 will be shown once they figure out what hardware is required for an actual robotaxi.

Would you please share the quote from Elon since I must have missed it. Thanks.

 

[DanCar]

Or it means they're really trying to finalize a product (versus showing something in its infancy like the bot).

Really I don't see any point in releasing HW4 until they can be sure that it will allow driverless operation. Of course I've always been a huge skeptic of "door to door" supervised automated driving. I think you actually need a certain error rate to keep people attentive (or maybe some limitations that require frequent driver input).

I was driving on FSD. I woke up and wasn't aware where I was. I'm disappointed I'm able to fall asleep with my eyes open, facing forward with a hand on the wheel. I think I tuned out for less than 10 seconds.

 

[Soda Popinski]

For what it's worth, Dr. Know-It-All speculated Optimus is running on a HW4 chip in his latest video. He noted the board only had one chip, instead of the usual two, and when he asked a Tesla engineer if they were using a new chip for it, they gave a somewhat surprised and guilty look, but no answer.

 

[diplomat33]

Or it means they're really trying to finalize a product (versus showing something in its infancy like the bot).

That's never stopped Tesla in the past. And why would they be ok with rolling out the bot in its infancy, but not be ok with rolling out a robotaxi prototype that is still in its infancy?

Really I don't see any point in releasing HW4 until they can be sure that it will allow driverless operation. Of course I've always been a huge skeptic of "door to door" supervised automated driving. I think you actually need a certain error rate to keep people attentive (or maybe some limitations that require frequent driver input).

First, Tesla needs to be clear about their goal. The sensors and hardware will depend on whether the vehicle is designed to be driverless robotaxi or "door to door" supervised automated driving. That makes a big difference. Tesla needs to decide which type of automated driving they are designing and then tailor the sensors and hardware for that. If they are doing L2 door-to-door, Tesla needs to update the cameras and add cameras to help see fast cross traffic, and upgrade the computer to handle the extra processing with more redundancy. And they need proper, reliable camera based driver monitoring. If they are doing driverless robotaxi, they need 360 degree HD cameras, 360 degree HD lidar, 360 degree HD radar and the computer to handle the extra processing with enough redundancy.

 

[Krash]

…If they are doing driverless robotaxi, they need 360 degree HD cameras, 360 degree HD lidar, 360 degree HD radar…

I’m not sure you and Elon are on the same page.

 

[diplomat33]

I’m not sure you and Elon are on the same page.

No, we are definitely not. But I am simply stating what a driverless robotaxi should have to be super safe and reliable.

 

[Dan D.]

I was driving on FSD. I woke up and wasn't aware where I was. I'm disappointed I'm able to fall asleep with my eyes open, facing forward with a hand on the wheel. I think I tuned out for less than 10 seconds.

I'm disappointed in that too, but how would HW4 or HW3 help though? If you're sleeping with your eyes open facing forward for 10 seconds with hand on the wheel how would any of the car companies' systems notice?

Glad you woke up alive.

 

[diplomat33]

I'm disappointed in that too, but how would HW4 or HW3 help though? If you're sleeping with your eyes open for 10 seconds with hand on the wheel how would any of the car companies' systems notice?

I believe many of the camera based driver monitoring systems can detect drowsiness or signs of sleepiness. So they can detect when our head tilts down because we are tired or when we shake our head to try to stay awake. So if @DanCar showed any signs of tiredness, a good camera based driver monitoring system could have alerted him to pay attention and maybe prevented him from falling asleep for a few seconds.

 

[DanCar]

I'm disappointed in that too, but how would HW4 or HW3 help though? ...

Will HW5 allow for L3?

 

[Dan D.]

I believe many of the camera based driver monitoring systems can detect drowsiness or signs of sleepiness. So they can detect when our head tilts down because we are tired or when we shake our head to try to stay awake. So if @DanCar showed any signs of tiredness, a good camera based driver monitoring system could have alerted him to pay attention and maybe prevented him from falling asleep for a few seconds.

Would they catch the zoned out 1000 yard stare? I guess if your eyes aren't moving enough. Maybe we need more biometrics, HR, O2 sat, to tell if we're alert or sleeping upright.
L2/L3 need to be better monitored, and with less time delay between nag checks.