Performance not getting 310 miles promised

[Perry]

Tesla rates the mileage of their cars more accurately than almost any other car on the road today. For instance, the 2013 Toyota Prius that I sold to get my Tesla had a rated MPG of 51. It actually got around 46. With a 10 gallon tank that's a difference of 50 miles.
Or the 2019 Nissan leaf that says it gets 124 miles of range but you only really get 60. There are 3 major standards to rate mileage that I'm aware of, most are a joke.

That's not my observation. My car is absolutely horrible at estimating my consumption and it would be understandable if it did so for a while while it learns my driving habits but it is as dumb as a post. It has learned nothing over the 10,000 miles I have driven it. Not once has it even been close so instead I have to do my own pen and paper calculations because if I followed it I'd always be stranded.

Here is an example. Estimate in gray. Actual in green. On this trip I got home with 27% when it estimated 38%

 

[AlanSubie4Life]

That's not my observation. My car is absolutely horrible at estimating my consumption and it would be understandable if it did so for a while while it learns my driving habits but it is as dumb as a post. It has learned nothing over the 10,000 miles I have driven it. It's so bad that it is useless. I have to do my own pen and paper calculations because if I followed it I'd always be stranded.

Here is an example. Estimate in gray. Actual in green. On this trip I got home with 27% when it estimated 38%
View attachment 441822

That's super interesting how bad that is. I'll have to experiment with using the Trip page on some of my commutes - normally I wouldn't even look at it.

I could see something like this happening if you were stuck in traffic on a hot day blasting the AC. Any details on whether this particular trip ended up taking place at a typical speed, according to what the Navigation would think was normal? I could see it being pretty dumb in those types of situations (though it should be able to handle it, as it knows about traffic and estimated arrival time, and external temperature...it may not do a good job). I assume you are routing with traffic data (it's possible they just are not sophisticated enough to take this into account though).

It looks to me like this trip was a net downhill, as the nav predicted nearly zero energy use. Furthermore it was a lot of downhill, followed by some uphill. But 10% of your battery is 7kWh, which would be like 3-7 hours of AC usage...for a 10 mile trip. I know the traffic in San Jose is bad...but...

Details on your wheels & tires, etc.? If you're running something non-standard that could confuse it I guess.

What was the Wh/mi on this trip? Looks like about 700Wh/mi for a trip that goes downhill about 1000 feet over 10 miles? Not sure whether their algorithms are tuned to adapt for that sort of driving.

It looks to me like it did adapt a little bit to your driving style (look at the slope of the gray line prediction over the last 2.5 miles of the trip vs. the green line slope between 7.5 and 10 miles - it's slightly steeper), but it may have a capping assumption at some point on how inefficiently a person could possibly drive.

 

[diamond.g]

That's not my observation. My car is absolutely horrible at estimating my consumption and it would be understandable if it did so for a while while it learns my driving habits but it is as dumb as a post. It has learned nothing over the 10,000 miles I have driven it. Not once has it even been close so instead I have to do my own pen and paper calculations because if I followed it I'd always be stranded.

Here is an example. Estimate in gray. Actual in green. On this trip I got home with 27% when it estimated 38%
View attachment 441822

Don’t you leave yours in track mode all the time?

 

[Perry]

I invite other P3+ owners to update their Screen Shots on what they are getting, I think someone leaves nearby me in Diamond Bar... only way I can tell if my car is a lemon.

On my P3D my range is 120 miles if driving normal (my normal) and 60 if I push it on the track. It sucks, I know, its not even close to rated, but if you get more than 120 then the battery is probably fine because I've never been able to get more.

 

[AlanSubie4Life]

On my P3D may range is 120 miles if driving normal (my normal) and 60 if I push it on the track.

Yeah. Curious: how many sets of tires have you gone through at this point? How many miles on the car?

How many times have you had to replace the brake pads?

 

[Perry]

Don’t you leave yours in track mode all the time?

Now about 50% of the time yes but that doesn't do much for consumption. I have not seen it be any more accurate in its estimation of miles in or out of track mode. Its bad all around. The problem is that there is no learning algorithm to adapt to your driving. It should calculate the estimate based on you life time per mile consumption which is a great indicator of how the car will be driven under given profile.

 

[AlanSubie4Life]

It should calculate the estimate based on you life time per mile consumption which is a great indicator of how the car will be driven under given profile.

You could argue that, but it's fairly clear that you don't care about your range. Kind of a less important corner case for Tesla. It is interesting that it appears to have some limits to how much it will adapt, though. It's possible your wider wheels also throw off their estimation algorithms, even if you were driving efficiently, I guess.

Would be interesting to know the exact route that was shown above just to see what baseline assumptions it was making about your driving. We should be able to back-of-the-envelope calculate how efficiently it thought you would drive.

It definitely does make some adjustments in the estimate for speed, so it has some ability to adapt the estimate. But I guess there are limits. As I understand it, most people find the estimates to be pretty accurate once they get underway for a few miles.

 

[Perry]

Yeah. Curious: how many sets of tires have you gone through at this point? How many miles on the car?

How many times have you had to replace the brake pads?

My stock tires became no tread slicks at 6k miles -- just from city driving; did not track car with these. I have 10K now. I got wider tires (265) as my second set so these are lasting a little better because of better grip so they are not sliding all over the place leaving rubber on the streets. The car should have come with 265 stock for how much power and g-forces it put out.

Still on my stock brake pads.

 

[AlanSubie4Life]

just from city driving

"City" driving*



*Example of sedate city driving provided herewith


How are your brakes doing?

better grip so they are not sliding all over the place leaving rubber on the streets

Yeah those stock PS4S tires aren't known for their grip.

 

[Perry]

What was the Wh/mi on this trip?

~430

Not sure whether their algorithms are tuned to adapt for that sort of driving.

Yes they don't take past driving user characteristic into account at all which I think they should. It is only accurate if you drive exactly how they expect you to drive with an unmodified car.

It looks to me like it did adapt a little bit to your driving style (look at the slope of the gray line prediction over the last 2.5 miles of the trip vs. the green line slope between 7.5 and 10 miles - it's slightly steeper), but it may have a capping assumption at some point on how inefficiently a person could possibly drive.

But why does it have to adapt on each trip. Instead it should use past user profile consumption as one element of the estimating function that way the estimate would be accurate from the beginning rather than having to adjust. By using past consumption it would automatically take into account the mods done on the car like wide wheels and such. The way they do it, the computer always gets shocked every time as if they've never seen this car nor this driver and have to adjust in an irrational way.

 

[ibGeek]

That's not my observation. My car is absolutely horrible at estimating my consumption and it would be understandable if it did so for a while while it learns my driving habits but it is as dumb as a post. It has learned nothing over the 10,000 miles I have driven it. Not once has it even been close so instead I have to do my own pen and paper calculations because if I followed it I'd always be stranded.

Here is an example. Estimate in gray. Actual in green. On this trip I got home with 27% when it estimated 38%
View attachment 441822

Wow, well all this does is reinforce how glad I am that I purchased the SR+ M3. I always beat expected mileage in it and I'm not one to poke around at anything less than 80 MPH on the highway, always with the AC on. (set to 72)

But as for your reply, I was referring to the window sticker.

Here is some charging data from my car: (note the 100% charges were during trip)

 

[ibGeek]

I find that the more you use autopilot, the better the mileage estimates are as you would expect.

 

[AlanSubie4Life]

But why does it have to adapt on each trip. Instead it should use past user profile consumption as one element of the estimating function that way the estimate would be accurate from the beginning rather than having to adjust.

The time the Trip page is most important is on a road trip. Introducing longer term behavior into that would introduce a long decay time constant, and result in potential inaccuracy on the % available upon arrival. This is because you might do 30 miles of city driving, then the next day head out on a freeway road trip - the baseline numbers will be very different. You want the estimate to adapt fairly quickly to current conditions, while taking into account future terrain & expected speed.

I agree it would be kind of nice if it would take into account wider tires or something on a continuous basis, allowing you to dial in a rolling resistance correction or something. But that gets complicated and it's useful for an extremely small % of owners. And it's safe to say that if you're running 265 width tires you don't actually care about your range.

Because of your driving style, you are way outside the bounds of any estimate they might be able to do. Again, that was a downhill trip you did and the projected usage was ~0 Wh/mi. You ended up using 400Wh/mi. That's a massive deviation from "conservative" driving behavior.

If someone cares how far they can go, they wouldn't be driving that way, typically. It doesn't sound like you've ever been stuck, and you pay close attention to your % because you know you're burning it rapidly. This is a use case that Tesla doesn't really need to focus on.

I'm not saying the Trip Estimate couldn't be better. It probably could be.

 

[Perry]

... you don't care about your range.

I care about range but as a second priority to performance. Its not fun constantly having range anxiety because even here is California there aren't SC stations close enough for the short range of my car. I have to plan very carefully not to get stranded.

It's possible your wider wheels also throw off their estimation algorithms, even if you were driving efficiently, I guess.

Except it was poor at calculating even before I got my new wheels. Not as poor but poor.

Would be interesting to know the exact route that was shown above just to see what baseline assumptions it was making about your driving. We should be able to back-of-the-envelope calculate how efficiently it thought you would drive.

This was the route:

 

[ibGeek]

I care about range but as a second priority to performance. Its not fun constantly having range anxiety because even here is California there aren't SC stations close enough for the short range of my car. I have to plan very carefully not to get stranded.

Except it was poor at calculating even before I got my new wheels. Not as poor but poor.

This was the route:

View attachment 441841

I ran the numbers per "Abetterrouteplanner.com" which has always been spot on for me and provided stick to 125% of the speed limit you should arrive with 85% of your battery left. But just in case I entered data about your car incorrectly, I've added a link to go to the shared route so you can make the edits you want. I use this all the time and it's fantastic.

A Better Routeplanner

I understand that you want to go fast and turn sharp. As the child of a race car driver I appreciate the need for speed. But I'm older and like the chill stress free experience that my car provides me when I can let it do the driving for me. The benefit of which is spot on range and no anxiety.

 

[AlanSubie4Life]

This was the route:

That's a 30-mile route. You posted a 10-mile Trip estimate. That's a 1600-foot descent and you got 400Wh/mi on it. Impressive!

It's hard to be definitive/accurate because the picture is incomplete. There are no mysteries here though.

 

[Perry]

The time the Trip page is most important is on a road trip. Introducing longer term behavior into that would introduce a long decay time constant, and result in potential inaccuracy on the % available upon arrival. This is because you might do 30 miles of city driving, then the next day head out on a freeway road trip - the baseline numbers will be very different. You want the estimate to adapt fairly quickly to current conditions, while taking into account future terrain & expected speed.

I agree. But start out with an estimate where one element of the equation is past consumption behavior (as a base line) and then adjust from there based on road conditions, terrain, and all other factors you mentioned. I don't think Tesla looks at that at all which I think is a flaw because in my case, and probably others, this parameter they are leaving out is the one that happens to be the most influential if the goal is a correct estimate.

I agree it would be kind of nice if it would take into account wider tires or something on a continuous basis, allowing you to dial in a rolling resistance correction or something. But that gets complicated and it's useful for an extremely small % of owners.

I don't think it would be that hard. I am sure they start out with a seed base line consumption value in their equation upon which they apply their other variable calculations. It is obvious that this seed value is constant. All they would have to do is make that seed a variable instead based on life time consumption per user profile. Or if not life time, in case the user changes behavior or hardware, make it last 3000 miles. This simple value would take into account all behavioral and physical changes because it is one indicator that catches all.

Because of your driving style, you are way outside the bounds of any estimate they might be able to do. Again, that was a downhill trip you did and the projected usage was ~0 Wh/mi. You ended up using 400Wh/mi. That's a massive deviation from "conservative" driving behavior.

It would not have been as bad if their base line seed consumption value to start the calculations from would have been 430Kwh and then calculated up or down from there based on terrain, weather, and all the other variables.

It doesn't sound like you've ever been stuck, and you pay close attention to your % because you know you're burning it rapidly. This is a use case that Tesla doesn't really need to focus on.

No I have not but it takes a lot of effort on my part and often I can't take certain routes because I am low and there are no SC stations in certain areas so it limits my driving.

Its an easy fix. I could fix it if they'd let me. I am a senior software engineer and work on things similar to this. It is an order of magnitude easier to fix this than doing silly things like the fart app and lame game ports I would never play in my car.

I don't think the focus is in the right place and that is making the car's basic operational roles function properly.

 

[ibGeek]

 

[AlanSubie4Life]

It is obvious that this seed value is constant.

Maybe they should use a neural net for the estimator.

 

[AlanSubie4Life]

A Better Routeplanner

You ran his route the wrong way around. He was obviously going downhill! Also his route appeared to be between the pass and Gilroy (10 miles), not the 30 miles to Danna Court.

I get an average of 86Wh/mi to get to Gilroy with ABRP (think that might be a bit pessimistic). He managed to do 400Wh/mi. Could happen to anyone.