Automated Cars Are Not Able To Use the Automated Car Wash (thetruthaboutcars.com)
schwit1 shares a report from The Truth About Cars: [T]he simple task of washing a self-driving car is far more complicated than one might expect, as anything other than meticulous hand washing a big no-no. Automated car washes could potentially dislodge expensive sensors, scratch them up, or leave behind soap residue or water spots that would affect a camera's ability to see. According to CNN, automakers and tech firms have come up with a myriad of solutions to this problem -- though a man with a rag and some water appears to be the most popular. Toyota, Aptiv, Drive.AI, May Mobility, and Uber have all said they use rubbing alcohol, water, or glass cleaner to manually wash the sensors, before carefully finishing the job with a microfiber cloth. While it's more than just a little ironic that these automated vehicles require gobs of attention and pampering from human hands just to function correctly, some companies are working on a way around it. General Motors' Cruise has said it will design and implement sensor-cleaning equipment in production vehicles.
If it can't handle a car wash at low velocity that lasts a few minutes, how's it going to handle hours and hours of rain with entrained road grime at highway speeds?
I deal with Lidars a lot, on self-driving loaders in an underground mining environment. It's pretty much the worst place for them - dusty, wet, hot, lots of vibration, you name it. Even though they are sealed to IP67, with o-rings on sealing surfaces & etc, they get water in them on a regular basis - IP67 is no match for even the mild pressure from a garden hose, let alone a pressure cleaner. The recommendation from the manufacturer is to send them back to the factory when they get wet, we generally take them apart and dry them out because a visit to the factory costs upwards of $5,000 (and six weeks delay) for an $10,000 device.
They mostly need to be cleaned about once a shift if the conditions are average. They need to be cleaned hourly if conditions are terrible. Failure to clean them gives us missing portions of scans if a mud splatter hits the lens, or a general loss of distance if it's just grime. Both of those things upset the self-driving software eventually, and then it's tedious manual control until someone can go clean it.
A dirty lens used to give us a "pollution error", but we changed the settings in the firmware of the lidars to turn that function off because we were sick of regular halts for errors that had yet to make an impact on the machine's operation. That is, what the manufacturer thinks is a critical pollution fault is actually about halfway to being unusable.
Lenses on our machines typically last about a couple of thousand hours of operation - probably a year or so if you translate that to a passenger vehicle. And of course, when cleaning them the instructions say to use a mild detergent and a clean, lint free cloth, gently buffing to a sparkling result. In reality, that is usually windex (or contact cleaner if there's grease on the lens) and any sort of material that can be found to wipe it with - paper towel, the sleeve of your shirt, a thumb, etc. Needless to say, this generally transforms the finely polished plastic lenses into a hazy scratched mess fairly quickly, especially if people spray and then wipe the lens without actually rinsing the crud off. So expect this to happen to consumer gear as well. And you can't just directly hose them, because hey, they aren't that waterproof either.
As long as there's plastic lenses in use, there's not really much manufacturers can do about this, other than have a secondary, cheap, external covering that can be unclipped and swapped out quickly. Or something like peel-off stickers like motorbike riders have for their helmets. They could shift to proper glass lenses, but even though they'd be much more durable, they would also be much more expensive to make.
(And what's going on with your backend, Slashdot? Heaps of timeouts and errors today.)
You are in a twisty maze of processor lines, all alike.
There is a lot of hype here.