Autowipers have stopped working entirely

[TigerNinety]

Hey all,

I know, it's an eternal issue. But this is unusual for me: usually, they misbehave a bit then seem to get "unstuck." Now, for the past week or ten days, they don't work at all. Like, today, I tried twice to see how long I could go in the rain before they'd start. Twice, I was basically driving blind and they didn't pipe up at all, even after manually wiping or using some of the specific, non-auto settings. Even more strange: a friend who just took delivery of a brand new 23MYLR also reports that his autowipers don't work at all.

Is there something with the latest software?

Thanks,

Todd
v2023.44.30.8

 

[OxBrew]

Mine did that for several days, wiped maybe 2x in 3-4 days of lots of rain. Then yesterday after noon the worked perfectly for 2 miles, in a light to medium rain, late afternoon, plenty of light out. Same sw version.

IDK. I want to buy a rain sensor and hook it up to a solenoid with a little finger that pushes the end button. I bet I can make it work faster than Tesla can.

 

[TigerNinety]

I'm willing to bet Tesla could make this work faster than all of us; they just aren't trying!

 

[OxBrew]

Yeah, if they installed proper rain sensors. But can't have that, might add $4 to the BOM cost. Heck, the replacement units I'm going to use are $7 retail from a Lexus, can't be more than $2 cost.

The idea that you can do this reliably with a camera behind the glass, across all weather and lighting conditions is laughable. It would be shot down as ridiculous in a no bad ideas brainstorm session. Guy with the pen at the whiteboard would pause, and say "ok, wait guys, you really want me to write that down? I mean I know, no bad ideas, but can we all agree right here this won't work and we're not wasting our time on it?"

And a couple people would laugh, and the guy who suggested it would say no, no, I was just making a joke!

 

[TigerNinety]

Yeah, if they installed proper rain sensors. But can't have that, might add $4 to the BOM cost. Heck, the replacement units I'm going to use are $7 retail from a Lexus, can't be more than $2 cost.

The idea that you can do this reliably with a camera behind the glass, across all weather and lighting conditions is laughable. It would be shot down as ridiculous in a no bad ideas brainstorm session. Guy with the pen at the whiteboard would pause, and say "ok, wait guys, you really want me to write that down? I mean I know, no bad ideas, but can we all agree right here this won't work and we're not wasting our time on it?"

And a couple people would laugh, and the guy who suggested it would say no, no, I was just making a joke!

Is it really that laughable? We count on AI for all kinds of pattern recognition, you’d think this one would be not that hard.

 

[OxBrew]

Is it really that laughable? We count on AI for all kinds of pattern recognition, you’d think this one would be not that hard.

Solve the case where the glass at the camera is perfectly clear and dry but the glass at the driver's head is smeared with mist and impossible to see through. A rain sensor at the base of the window is much closer to the drivers view.

Surprisingly often, mist blows up from the hood and only covers the lower 2/3 of the window.

Simply impossible to fix that one.

Or it's so dark out on a back road that no light is hitting the camera area. I see that a lot in winter.

And if AI could fix this with pattern recognition, why hasn't it? It's been what, 5 years? --looked it up: 8 years. lol.

There are enough common "edge" cases it will never cover that it's a waste of time to try.

Edit to add:

Plus, you've got a wide variety of cameras across models, so you have to calibrate for each one. Then add in a range of color drift even on the same batch of cameras (some look more blue, some more red). Then add even more variety around the chaotic mix of -whatever chips they could get- for different boards that have to process the images, all impacting what the "AI?" really sees. It's a chaotic s---show that any decent PM would shut down in 15 minutes.

Choices:

1. commit to a hellacious long term development and maintenance nightmare that would make good software engineers quit on the spot, using up who knows how much future labor.

2. Buy an off the shelf solution for $5 (looked it up: $1. ONE DOLLAR. JFC.) that's already solved and only needs occasional updates to deal with new connector designs or stuff like that, that you already implemented in the past.

Sorry to rant about this, it's just crazy how Tesla has lot the plot on some of this stuff.