[bigdang]

Giving optimistic estimates based on generalized vehicle information, then giving more precise estimates after the 50% mark—and after having collected usage information based on the user’s actual environment—sounds to me like a decent algorithmic solution to a hard problem.

[542354234235]

It sounds like a terrible solution to an easy problem. And calling it an "algorithmic solution" is being generous, considering that our dead simple decade old Mazda 2 gives us a closer than 25% range estimate just based on the mpg for the previous X number of miles driven. That's not an algorithm, that is a simple math calculation with only two inputs; gas consumption rate and miles driven. Tesla, with thousands of data points on previous usage and driver behavior, could give you an almost dead accurate estimate, but chooses to give a basically useless estimate because it looks better. Then people come around and make ridiculous excuses for it and why its actually a “decent solution” (it isn’t) to a “hard problem” (its not).

[simondotau]

Tesla vehicles do provide an accurate estimate. The estimates provided when you navigate to your destination are usually within 1-2% of correct, and take into account road speed, elevation changes, regenerative braking, wind direction, etc.

Like many EVs, the "fuel gauge" number is based on the EPA estimate and the navigation range is based on accurate predictions. It's like this in the Polestar too.

https://www.youtube.com/watch?v=U3P32TyLMMM&t=644s

[ke88y]

That only makes sense if there's strong correlation between the first 50% of a charge and the second 50% of a charge, but NOT an equally strong correlation between the previous charge cycle and the current charge cycle.

That can be the case (eg on road trips) but usually isn't.

Why not just always give the more precise estimates?

[barbegal]

Because then you get people complaining it doesn't meet the range as specified in the marketing material which is based on the EPA (or equivalent) test.

[tekla]

How does the car know what conditions its going to be driving in? I don't believe Elon has installed a time machine yet into a Tesla.

[ke88y]

I'll literally repeat myself, since I think the point was abundantly clear:

>> That only makes sense if there's strong correlation between the first 50% of a charge and the second 50% of a charge, but NOT an equally strong correlation between the previous charge cycle and the current charge cycle.

But "That", I mean using optimistic estimates at the beginning of a charge cycle but a model based on historical driving about half way through a charge cycle.

There's no reason, prime facie, that future discharge rate should be better-correlated intra-charge than inter-charge, especially for recent charges, and especially if you have lots of historical charge/discharge cycle data for both the machine, the geographic area, and the individual driver.

The question is not "why not be perfect". The question is "why not do inter-charge forecasting the same way you do intra-charge forecasting".

[wredue]

Er? Perhaps the car could make use of sensors that every ice and ev vehicle has had installed for decades, along with previous driving habits?

[tekla]

Yes the sensors on the car are now time machines that can figure out what the cars future conditions are going to be.

Why would you want your previous driving habits to affect the base level estimation? Your driving today could be vastly different, and then you would be here complaining about how wrong it was and the range anxiety you got when the car vastly underestimated your range because it assumed you were in 40 mph traffic when today you were late and speeding 80 mpg down the highway.

[542354234235]

>Why would you want your previous driving habits to affect the base level estimation?

Because I could still just look at how much charge I had, the same way I look at how much gas I have left in the tank of my car. My old Mazda 2 gives estimated miles left in the tank based on a dead simple math calculation of average mpg for the previous X number of miles driven with only two inputs; gas consumption rate and miles driven. No time machine needed because it is an estimate and I am an adult with a functioning brain. I already know how big my gas tank is. I already know what my "EPA" mileage is. Telling me those things is not giving me an estimate, its just reminding me of what's printed in the owners manual. It is pathetic that my old cheap car gives me an actual estimate using actual data from my actual car and Tesla doesn't...until its under 50% and then suddenly its ok.

[mrguyorama]

Funny, my ICE car doesn't have to look into the future to know that I've been driving like a grandma for the past 100 miles and I have four gallons of gas left so I can probably go another hundred miles at my driving style.

Which is how it calculates it's range, and also how my 2004 toyota calculated range. The maximum range figures consistently lowered over the course of time I owned it as the engine got old and less efficient! It went from consistently having a max range of 420ish miles to consistently having a max range of 380 miles over twenty years!

Funny that old stuffy mid 2000s toyota could manage that on a car that cost $25k, but legendary tech luxury car maker Tesla can't

[clintonb]

Navigation data may include altitude changes. Weather data along the route gives you temperature. The driver’s history tells you how likely they are to speed versus drive the speed limit.

[marcosdumay]

Hum... ICE vehicles have given their range in litters for decades, and nobody every had a problem with it.

They recently stated giving estimates in distance, but it's clearly marked as an unreliable estimation.

Looks like Tesla has another huge communications and UX issue, and not a mechanics (electric?) one. They have to get their designers in a room, fire the management, and ask them to actually design stuff for humans.

[mrguyorama]

ICE cars have been ACCURATELY estimating remaining range based on current driving habits since at least 2004 in my toyota.