Middle of the Road

  • SUPPORT THE SITE AND ENJOY A PREMIUM EXPERIENCE!
    Welcome to Tesla Owners Online, four years young! For a low subscription fee, you will receive access to an ad-free version of TOO. We now offer yearly memberships! You can subscribe via this direct link:
    https://teslaownersonline.com/account/upgrades

    SUBSCRIBE TO OUR YOUTUBE CHANNEL!
    Did you know we have a YouTube channel that's all about Tesla? Lots of Tesla information, fun, vlogs, product reviews, and a weekly Tesla Owners Online Podcast as well!

tencate

Top-Contributor
TOO Supporting Member
Joined
Jan 10, 2018
Messages
652
Location
Los Alamos, NM
Country
Country
Tesla Owner
Model 3
#1
This is more of something to think about, not sure it belongs here but anyway... I drive the same two lane stretch of road to work every morning. I frequently let the car do the driving for me, mostly to see how well it's handling the curvy parts. I'm curious to see if it's learning or getting better at it. But, what got me thinking this morning was just how well it was keeping me in the exact middle of the lane. So much so that I'm hitting the exact same pothole in exactly the same way day after day. Doh! :confused: But does this mean, eventually, with a road full of Model 3s on autopilot, are we going to get wagon wheel ruts in our roads? :)
 

garsh

Dis Member
Moderator
TOO Supporting Member
Joined
Apr 4, 2016
Messages
13,054
Location
Pittsburgh PA
Country
Country
Tesla Owner
Model 3
#2
I'm curious to see if it's learning or getting better at it.
Everyone seems to get confused when Musk mentions using "machine learning" to help improve self-driving abilities. No individual car learns anything. And they're still at fairly basic stage of trying to teach a car to correctly recognize the boundaries of a road. They're nowhere near the point of pothole detection yet. Heck, they can't even reliably keep the car on a basic two-lane road.
 

SoFlaModel3

Moderator
Moderator
TOO Supporting Member
Joined
Apr 15, 2017
Messages
10,175
Location
Florida
Country
Country
Tesla Owner
Model 3
#3
Yeah @garsh is right. Your car isn’t learning and should handle the same drive at the same speed the same way every time (assuming conditions remain static). With software updates the machine learning is applied to the fleet with new logic to improve how it handles these various conditions.
 

tencate

Top-Contributor
TOO Supporting Member
Joined
Jan 10, 2018
Messages
652
Location
Los Alamos, NM
Country
Country
Tesla Owner
Model 3
#5
I dabble in machine learning at work. Bad choice of words on my part: I realize it's not actually learning to drive better. But I am curious to know why, when my daughter first engaged autopilot, the car was having to do something before it'd let us do anything with EAP. What's it doing exactly? Anyone know? Some sort of calibration? I didn't have the car when she turned it on but she reported that Max wasn't very good at anything at first. By the way, the software itself is getting lots better. I've seen it go from getting seriously confused with highway entrances and the like (early 2018) to now being able to follow some incredibly twisty roads around here with ease. That said, I never really take my hands off the wheel either.
 

garsh

Dis Member
Moderator
TOO Supporting Member
Joined
Apr 4, 2016
Messages
13,054
Location
Pittsburgh PA
Country
Country
Tesla Owner
Model 3
#6
But I am curious to know why, when my daughter first engaged autopilot, the car was having to do something before it'd let us do anything with EAP. What's it doing exactly? Anyone know? Some sort of calibration?
I believe it's calibrating the sensors (cameras). Each car is going to have the cameras aimed *slightly* differently, so it has to first figure out exactly where each one is pointing.

Sorry, I don't recall where I heard that information. I need to track that down.
 

Brett

Well-known member
TOO Supporting Member
Joined
Jul 31, 2017
Messages
235
Location
Brentwood, CA
Country
Country
Tesla Owner
Model 3
#7
I believe it's calibrating the sensors (cameras). Each car is going to have the cameras aimed *slightly* differently, so it has to first figure out exactly where each one is pointing.
Personally I don't buy the "calibrating" answer. The cameras are going to be seeing slightly different views all the time, when the road changes, when the weather/lighting changes, when you you change lanes or steer the car, etc. The light intensity on the camera and the placement of objects within the frame aren't constant enough for a calibration to be meaningful.

I think the system is running in shadow mode to validate the hardware and software are working correctly. Basically if the software is going to crash or the hardware is going to overheat (etc) they want to notice it when the system isn't actually controlling the car.