Wondering if “manual adjustment” means to change from Auto to the I, II, III or IIII setting - or also includes pressing the stalk for single extra wipeIn the release notes for 2019.40, Tesla confirms that it is using their first production deep neural net trained with over 1 million images of rain drops.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Tesla now using deep neural net for auto wipers (2019.40)
- Thread starter diplomat33
- Start date
I would assume both. Pressing the stalk for a single wipe would definitely be a manual adjustment.
Tesla should pay us for having trained this neural net. Each time auto didn’t work and we did a manual wipe, we taught it.
Still a bull *sugar* implementation. Auto wipers were perfected decades ago and all this to save a few cents on a traditional sensor.
To boast about it takes great arrogance.
Still a bull *sugar* implementation. Auto wipers were perfected decades ago and all this to save a few cents on a traditional sensor.
To boast about it takes great arrogance.
alstoralset
Man from another place
Hurrah for Deep Wipe! Can't come soon enough. Here in the Seattle area, we have 57 different words for "rain."
What does percentage have to do with it? Think CO2 emissions.
Even just the amount of energy wasted training the AI to spot rain... All for the sake of not fitting a $5 optical sensor that is more reliable anyway.
The optical sensor is not more reliable at detecting occusions in a different section of the windshield, it is useless for that.
Edit dropped a decimal.
$5 = 30kWh = 8,000 hours of operation at 3.6 W.
Or, at 300k cars per year, Tesla saves $1.5 million to put toward other CO2 reducing initiatives like solar in GF1. Since solar is $1 a watt, that is a net reduction in emissions over an off the shelf sensor.
But, again, why pick on the rain sensor NN (which has a necessarily role) when there are much more power hungry drains on electricity which don't?
Radio - optional
Heated seats - optional
Heater steering wheel - optional
Clear view of the road - required, unless you want to drop the optional AP entirely which would save a lot more energy than 3.6W.
Last edited:
Though I am very new to the Tesla model 3 I do know about neural networks. The training is done on a big computer somewhere. The error backpropagation algorithm for a deep NN is computationally intensive. But, once trained, the neural network can be deployed on low processing power devices, such as our Tesla 3 computer.
This assumes the rain in Spain is the same as the rain everywhere else, I suppose.
This assumes the rain in Spain is the same as the rain everywhere else, I suppose.
SabrToothSqrl
Active Member
Too bad it's still crap. My 10 year old Tahoe had 100x better auto wipers.
Jumping through hoops to save a $5 (maybe more?) sensor is actually pretty common in the auto industry. For example, General Motors has very sophisticated algorithms to estimate control air fuel ratio in their gasoline engines, so that they can use cheaper oxygen sensor (narrowband sensors). As you said it's millions of dollars and a lot of upfront work. If Tesla makes half a million cars next year it adds up.
The difference is, these kinds of neural nets need tons of real world data and the customer inevitably become beta testers.
You have version 2019.40 and tested it? Most everyone agrees it’s crap. We’ll see when everyone has it. I’m Cautiously optimistic.
Kid had a GM Flex Fuel car. Needed a tank of E85 every fall due to losing calibration.
Tesla isn't the only one with not reliable sw out there. At least T has OTA to fix it.
“Deep rain” is too little too late. As a custumer I had never imagined my brand new 60kUSD car would not have such standard equipment.
For those times when it's raining on one side of the window but not the other?
Most cars get around that extremely rare problem by having "tap to wipe" on one of the steering wheel stalks, along with the windscreen washer. No illegally messing with the touchscreen just to clean your window.
Except they probably wasted 20x that amount getting it to work in the first place. How many man-hours and how many compute-hours on the training, which let's not forget they botched the first time and had to repeat and it still doesn't work very well.
alstoralset
Man from another place
Similar auto wiper sensors retail for $100-$300 on Amazon, so I suspect they cost OEMs at least $20. If correct, the ongoing BOM savings to Tesla is pretty significant.
I suspect that in hindsight they wish they'd included the rain sensor. The sensor omission is just one of many cost cutting measures employed in the M3, in order to build an "affordable" electric car. This one blew up in the test tube, but overall the strategy has been successful, IMO.
I suspect that in hindsight they wish they'd included the rain sensor. The sensor omission is just one of many cost cutting measures employed in the M3, in order to build an "affordable" electric car. This one blew up in the test tube, but overall the strategy has been successful, IMO.
Uhh have you read the owner's manual? Press the button on the left stalk of the Model 3 and it is "tap to wipe" , and if you really press it in until it bottoms out, it turns on the windscreen washer.
That's not how it works. There's a saying in the automotive industry (where I work), "Software is free." It's a bit facetious but basically reducing piece cost has a much higher benefit financially than you think, because of the accounting. Software man hours are fungible, $5 per car at half a million a year is a big deal. It improves cash flow and feeds into the calculation of the profit margin a lot, something that public companies have to pay attention to. I've seen entire development programs (albeit smaller scale) to cost save sensors that cost a few dollars.
In the meantime, if the wipers annoy you, hit the button on the stalk.
Similar threads
- Article