Energy / battery - why no adaptive range?

[macleodal]

Folks,

As a relative newcomer to Tesla, but as an engineer who understands the power dynamics, I am very curious about something and wonder if I am alone.

My Model S Long Range gives me a "miles" number for the equivalent of my fuel tank - but it is never right, it always over-states. Equally the energy burndown display when on a trip is never right. And when the car tells me I should make it home with 3% to spare, I already know I won't make it home! All of which is fine, I can choose to drive like a monk (which you need to do to achieve the indicated range) but I don't choose to.

What I am curious about is this. With such a smart car, with so much compute power available to it, with so many options and adjustments... why does the car not give me the chance to see a REAL range based on it's learned understanding of my driving style? I understand we all drive differently and I also understand that averages aren't always helpful but this car is monitoring my every move: steering inputs, break pressure, acceleration behaviour in different weather and road conditions, etc. I am 100% certain that, if they wanted to, Tesla could allow my car to give me a highly accurate range: "Alastair if you drive like you normally drive you have a real range of 265 miles not 320" for example; in the same vein the energy burndown chart could set the baseline for a trip to be what the car expects ME to achieve, rather than Manny the Monk.

I do get that Tesla maybe don't want to advertise that the real range is lower than the published, marketed range but unless we're stupid we all knew that before we got the car in the first place. Now it's nothing but annoying that the range is just wrong all the time and you have to manually calculate the offset. Ironically the range indicator on my diesel BMW was much more reliable!

Any thoughts?

Thanks

Alastair

 

[MrBadger]

Use the energy graph. And if you put in a destination, it will predict your arrival statistics based upon the planned route topography and adapts as you drive.

 

[tonyj01]

Hi Alastair,

I too came to the Model S from a BMW diesel that did 600 motorway miles on a £100 tank!

You will get used to the actual range you will achieve, but there are other variables.

I used to be puzzled why some days a 100 mile journey in my BMW would be at say 35mpg, the return trip, same day, would be 41mpg. In the Tesla, I now know this might be due to wind direction, and/or rain. In an ICE, you might see the variation in range yet shrug it off, we analyse it more in an EV because, I think, the range is lesser.

The variation is also large between winter and summer.

After 4 years I still think the benefits of an IT based EV outweigh the messing about with charging that is necessary with limited range. I think this remains a deterrent for ICE drivers.

Enjoy!


Tony

 

[Adopado]

What I am curious about is this. With such a smart car, with so much compute power available to it, with so many options and adjustments... why does the car not give me the chance to see a REAL range based on it's learned understanding of my driving style?

The only truthful answer is "nobody knows"! My speculation is simply that most drivers vary their style a fair bit so it won't know whether to base the estimate on when you were in a sporty mood or when you were needing to calm down after a stressful work meeting. It therefore picks a somewhat idealised usage pattern that it can use all the time and the driver can then get used to that consistent prediction and how it will relate to their own driving ... which is a reasonable compromise. I also think you are right that they would never want to predict a lower range than the car is stated as being capable of.

 

[Jeeves]

ABRP Premium does a pretty good job at estimating complex trips with waypoints taking into account weather, terrain and load. I use that when planning an unfamiliar journey.

As you get used to the car, you might start to become less concerned about this. At the beginning, I was as puzzled as you. Now all I’m concerned about is whether or not I have enough juice to get me to the next charging point in 2-3 hours worth of driving. The answer is almost always yes for me unless I’m going somewhere not well served by superchargers. In that case, you might consider arriving at a public charger with quite a bit of reserve in case it doesn’t work when you get there.

 

[SSedan]

The energy graph will tell you remaining range at current consumption.
I have found NAV to be pretty good at estimating destination arrival range once it gets an idea what the drive will be like.
It can be confused if for instance the temp suddenly plummets and it hasn't been working heater use into consumption till you leave that morning but by half an hour in it generally has things figured out.

You will adapt to the car. Just enjoy.

 

[macleodal]

Thanks to all the respondents above - I still think this is a very interesting (and curious) topic!

I do totally get the variability of this: mood I'm in (sporty? chilled?), the weather, the road conditions... get all that. But somehow my diesel car told me (say) 152 miles remain and I knew with certainty that I could do 152 miles or slightly more.

This doesn't stress me at all with the Tesla. I just wonder why they don't try to do a better job and/or give you an option ("Do you want the idealised range or the modelled range based on your last [X] miles of driving?" - could even let us put in the X). I am at the point now where I just knock off about 15%... if the Tesla says I will be able to do 200 miles, I mentally assume not more than 170. But I just figure, if within a month of owning it and driving it I've worked out that -15% is a pretty good estimate for me... why not have the car apply a few Tb of data points, shove them through some machine learning, and come up with an adaptive range estimate based on either preset parameters or based on user-input parameters?

Personally I'd rather see it tell me 250 miles because I can very probably achieve 250 miles in the real world driving as I do every day, rather than 300 miles which I know and the car knows will never happen. The energy chart is even more silly because there - connecting a trip - it not only knows the mileage, it knows which roads (and how I usually drive on them), it knows the topology (which drives energy consumption), it knows the time of day (which determines traffic - and anyway it knows traffic!), it could even know the temperature and weather conditions. For TRIPS specifically there is no reason for the car to get it any less accurate than 1%... and yet it still presents the idealised figure!

Ranting now! But I'm very happy with it. Just think (for example) the time spent creating the code so I can have a picture of the suspension activity - which is pretty but also pretty pointless - could instead have been spent on something actually valuable to the driver.

Alastair

 

[Gatsojon]

I have my battery set to % and so don’t have this frustration. Think I’ll leave it there!

 

[Joelly]

Set battery to %, use A Better Routeplanner.
I suspect they don't do it because when you put your foot down it'll show half the expected range and they would get slaughtered for it

 

[Tony Hoyle]

I find it's more or less right in summer and an overestimate in winter, which is what I'd expect.

More importantly though it's consistent so you can predict range based on your own habits. and knowledge of the journey ahead. I had a leaf that tried to predict range dynamically. They call it the guess-o-meter because what it said was just a guess that changed based on seemingly random things.. you just can't predict the variability of driving with any accuracy.

 

[Glan gluaisne]

For at least 30 years car manufacturers have been wrestling with ways to better provide range information to the driver, and none have done it very well. As I mentioned here before, my old XJ-S from 30 years ago tried to display range, mpg etc on the computer display, but that had exactly the same problems that any modern car has, in that it doesn't have a clue as to what's going to happen in the future. Put your foot down for a few minutes and the range could drop by 50 miles. Tesla's system tries to guess range based on what it thinks is the state of charge of the battery, what it knows of the past driving style of the driver and things like the current outside temperature, speed, recent variations in terrain, etc. At best all it can do is assume that the car may be driven in future in much the same way as it's been driven in the past.

It's easy to test this. Just do a few spirited drives and the car will assume that's the new "normal" and base future range guesstimates on that. Conversely, drive very carefully for a few trips and the range estimate will start to increase a bit. The biggest single factor seems to be temperature, a very cold night can significantly decrease the guesstimate of range, and a very warm day can increase it.

I'm not convinced that any amount of code tweaking would get it much better, given the big variations in energy use with speed, temperature, gradient, air density etc. The Tesla guesstimate seems a lot more accurate that most ICE vehicle range predictions systems. When I used to keep records of mpg I found that my old Prius would have around 20% or so more range in summer than in winter, but that the onboard display didn't seem that good at predicting this, even thought the seasonal correlation with fuel consumption was very tight.

 

[jmaddr]

It's been said above, but that's why I have the battery icon in the main display show percentage. Just like a gas gauge. Then, when you want know how many miles you realistically have left, open the energy app. It does everything you stated above. It uses either the past 5m, 15m, or 30m as a baseline and gives you your estimated mileage left based on your previous driving history for that distance. When you add a trip, it adds a tab called trip which shows your trends of it's prediction vs your actual over the course of the trip. In an ideal world, I would want the (30 mile) range displayed there but Tesla is in the business of selling range so they use the EPA estimate.

Bottom line, use percentage in the battery gauge like a gas gauge. You will learn than 50% means 120 miles and not 150 and 25% means 60 miles and not 75. Use the energy app when you want to get nerdy and see real expected range based on driving habits, range, terrain, etc of the past 5/15/30 miles.

 

[MrBadger]

Tesla's system tries to guess range based on what it thinks is the state of charge of the battery, what it knows of the past driving style of the driver and things like the current outside temperature, speed, recent variations in terrain, etc. At best all it can do is assume that the car may be driven in future in much the same way as it's been driven in the past.

It's easy to test this. Just do a few spirited drives and the car will assume that's the new "normal" and base future range guesstimates on that. Conversely, drive very carefully for a few trips and the range estimate will start to increase a bit. The biggest single factor seems to be temperature, a very cold night can significantly decrease the guesstimate of range, and a very warm day can increase it.

Definitely not how ours behaves.

TeslaFi shows a rated range based upon EPA and battery condition, and an estimated range, which varies according to actual consumption during recent travel time.

Guess which one the car displays?

Yep, rated range. So not based upon recent driving.

However, TeslaFi does show a 'cold' battery rating, which if in force (its not in this screen shot), would have showed in blue along side the 76 above. The cold rating appears to be the % that the app displays, but I'm not entirely convinced that it is consistent. That will I believe affect the rated range, which is probably what you are seeing with cold/warm day differences.

 

[Glan gluaisne]

I must admit that I leave the car set on SoC % and only ever notice range changes as reported by Teslamate now. There is definitely a change in the remaining range as reported by Teslamate, but I've not bothered to see if that tallies with the car display when set to distance. Teslamate reports a couple of projected range numbers, one based on battery level the other based on usable battery level, and those numbers seem to either be identical, or up to about 35 miles different, for no obvious reason I can see from looking back through about a years worth of data. Dropping the temperature by about 15° seems to drop the range by around 16 miles, according to Teslamate.

 

[Yachtsman]

I do totally get the variability of this: mood I'm in (sporty? chilled?), the weather, the road conditions... get all that. But somehow my diesel car told me (say) 152 miles remain and I knew with certainty that I could do 152 miles or slightly more.

I totally agree with macleodal tesla could do a much better job.
My last 3 bmw/Alpina costing similar to my model 3 could all predict my range accurately enough that i never gave it a second thought and never run out of juice.
Weather, road conditions and my driving style were equally as varisble, so theres no excuse Tesla!

 

[MrBadger]

They do calculate it (energy graph and TeslaFi for example show it), but choose not to show it on the main screen for what ever reason. Its a UI choice rather than an omission. Anyone that has had the car for a reasonable amount of time seems not to be bothered by these decisions. Plenty of other real design decisions to focus our attention on.

 

[macleodal]

For at least 30 years car manufacturers have been wrestling with ways to better provide range information to the driver, and none have done it very well. As I mentioned here before, my old XJ-S from 30 years ago tried to display range, mpg etc on the computer display, but that had exactly the same problems that any modern car has, in that it doesn't have a clue as to what's going to happen in the future. Put your foot down for a few minutes and the range could drop by 50 miles. Tesla's system tries to guess range based on what it thinks is the state of charge of the battery, what it knows of the past driving style of the driver and things like the current outside temperature, speed, recent variations in terrain, etc. At best all it can do is assume that the car may be driven in future in much the same way as it's been driven in the past.

It's easy to test this. Just do a few spirited drives and the car will assume that's the new "normal" and base future range guesstimates on that. Conversely, drive very carefully for a few trips and the range estimate will start to increase a bit. The biggest single factor seems to be temperature, a very cold night can significantly decrease the guesstimate of range, and a very warm day can increase it.

I'm not convinced that any amount of code tweaking would get it much better, given the big variations in energy use with speed, temperature, gradient, air density etc. The Tesla guesstimate seems a lot more accurate that most ICE vehicle range predictions systems. When I used to keep records of mpg I found that my old Prius would have around 20% or so more range in summer than in winter, but that the onboard display didn't seem that good at predicting this, even thought the seasonal correlation with fuel consumption was very tight.

Thanks - you said some things I've not heard before and have also not personally experienced yet with my own Tesla but now you've said this I will look again.

I do understand the variables very well, which perhaps doesn't come across. Of course if the modeled range assumes "normal" driving (whatever that is) and then you start trying to impersonate Lewis Hamilton, or prove to every car in the county that you can out-accelerate them, of course it won't get it right! However the comparisons with older cars are less relevant to my point; obviously an old Jag or my first BMW back in 1992 that had "range", these things were totally rudimentary -- average MPG over a fixed period of time, projected forward based on fuel remaining (which, in itself, was not very accurate given the sensing technology of the day).

Where you and I depart is that I am 100% convinced that the massive stream of data from the car available in real time (speed, acceleration, turning force, grip, temperature, humidity, gradient, control inputs, etc etc) should be capable of a really good and accurate forecast range. Any computer model that bases a prediction on the past is at risk of getting it wrong when you introduce a discontinuity into one of the highest-weighted parameters of course; but even then, if you do it habitually, the machine can learn... hence the term. The computer in the Tesla should be able to take literally hundreds of parameters into account to come up with an accurate reforecast range -- from your comments, you think (or maybe you know) that it does this? My impression is that the range calculation - whether the one on the dash or the burndown chart - uses the same totally-idealised "Manny the Monk" driving model and that it completely ignores my own driving style. I drive in a fairly consistent way, I have had the car for over a month, and the error range between the energy forecast and the actual on arrival has not improved at all since I owned the car... typically 4-5% of total battery power difference on a journey of 2hrs or more.

I am starting to get the feeling on this forum that the real reason this doesn't get looked at is that Tesla owners like it the way it is or don't care. If that's the case then fair play to Tesla, they are giving the market what it wants - I just don't understand really why the market wants substandard information which is far far below what the car is capable of. I've just finished working for an advanced analytics / machine learning company dealing with much bigger data than this and much harder models to build and solve than this - I know, for sure, it would be a fairly small task to build a really good model and give me the option to use it; and given I choice of that or a silly suspension graphic, I know what I would choose!

Anyway. As I appear to be in a category of one here, I will now leave this topic and keep my thoughts to myself until Tesla do, at some point, fix this as I'm sure they will.

 

[macleodal]

I totally agree with macleodal tesla could do a much better job.
My last 3 bmw/Alpina costing similar to my model 3 could all predict my range accurately enough that i never gave it a second thought and never run out of juice.
Weather, road conditions and my driving style were equally as varisble, so theres no excuse Tesla!

Exactly. But we seem to be alone in our view here! Maybe early adopters of EVs are already reconciled to the limitations and shortcomings, I am a new (late) adopter and I think they should have done a better job here and - if they had - a lot of the real world range anxiety problem would be resolved. There is literally no way my wife will ever use this car for a journey if I have to brief her first on how to calculate the true range offset compared to what the car tells you!!

 

[Adopado]

Exactly. But we seem to be alone in our view here! Maybe early adopters of EVs are already reconciled to the limitations and shortcomings, I am a new (late) adopter and I think they should have done a better job here and - if they had - a lot of the real world range anxiety problem would be resolved. There is literally no way my wife will ever use this car for a journey if I have to brief her first on how to calculate the true range offset compared to what the car tells you!!

I don't think it's so much to do with early adopter vs later adopter ... it's just having a longer experience of the issues you describe (which is slightly different). Most of us swap to percentage view and rarely have to think in terms of overall miles. The energy graph in trip mode gives a real-time updated percentage prediction which I have found to be remarkably accurate. I know that if I drive moderately the initial prediction will be very close. I also know that if I drive in a more spirited style that the predicted percentage for arrival will reduce progressively en route. It's always sensible to leave an extra "buffer" if in any doubt. We are now getting into the colder weather period so short journeys will be particularly inefficient and will not be at all representative of the range on a longer trip or of the short journey efficiency in the summer months. Winter does highlight range variability and though ICE vehicles are impacted similarly you have a lot more leeway so it's not so noticeable.

 

[Jibjab]

Exactly. But we seem to be alone in our view here! Maybe early adopters of EVs are already reconciled to the limitations and shortcomings, I am a new (late) adopter and I think they should have done a better job here and - if they had - a lot of the real world range anxiety problem would be resolved. There is literally no way my wife will ever use this car for a journey if I have to brief her first on how to calculate the true range offset compared to what the car tells you!!

Totally agree.

Every car I’ve had with a trip computer, that estimates miles remaining, I’ve barely noticed the fuel gauge and always gone by the miles remaining view. I, and I think the vast majority of other users, understand that if, for example, the estimate is 50 miles remaining when you’ve been doing 50 mph, and you then decide to drive like you stole it, that the range will dynamically reduce. Fingers crossed for a future update...and this is before I’ve actually collected the car