Kurt Kinetic - Virtual Power

(Dan Morley) #1


I have completed 4 rides on Zwift Island now. I am using a Kurt Kinetic trainer and Zwift Virtual Power.

On my first three rides, I kept feeling that the power numbers (and speed) being show were unrealistic. I do not have a power meter, I have not had my power tested. I am basing my feelings on my effort while riding outdoors at comparative speeds and on readings from Trainer Road.

For my last ride, I setup one PC to use Zwift, another PC to user Trainer Road, and also recorded the ride on my Garmin 920XT. I then imported all three rides into SportTracks. The Zwift and Trainer Road fit files contain the power tracks, and I use the ST plugin Trainer Power to calculate the power for the Garmin fit file.

I then compared the power tracks for all three. The Zwift power track is consistently higher than the other two. It also seems that the higher the power, the larger the difference between the numbers.

See chart at https://dl.dropboxusercontent.com/u/62482348/Power%20Comparison.PNG

FYI, the Heart Rate and Cadence charts tracked very closely.

Before each trainer ride, I make sure my tire pressure is the same, and I perform my own calibration for the wheel against the trainer to try ensure I get consistent results.

I love the concept of Zwift, I think it has great potential!

(Clint Westhoff) #2

Maybe Zwift can tell us what formula they are using for the KK. It should be the same one TrainerRoad uses so virtual power should have matched up.

(Nigel Van de Velde) #3

I noticed the same thing for a Kurt Kinetic Road Machine II.

Today i did another Zwift ride and simultaneously captured the data with my garmin device (1s recording, no autopause). After the ride i editted the data of the garmin’s ride and calculated the virtual power for every datapoint with the formula stated on Kinetic’s website (https://kurtkinetic.com/technical-information/kinetic-power-tech/).

That way I ended up with 2 files; 1 from Zwift containing virtual power and 1 from my garmin containing virtual power. Following link shows the respective power curves of the 2 rides.


As you can see the profile is indeed a lot higher for the Zwift recording…
I realise virtual power calculation is a lot harder on the Zwift course, as they need to compensate for elevation as well, so they can not use the formula from Kinetic’s website… Still I hope that this can be solved during the beta testing.

(Clint Westhoff) #4

Nigel van de Velde: “I realise virtual power calculation is a lot harder on the Zwift course,”

Actually it isn’t. The power calculation has nothing to do with what the little guy on the screen is doing. The power calculation is determined by the speed your real world wheel is spinning while attached to your trainer. The results of that formula is your power. That power affects what the guy on the screen does just like with a power meter, but the guy on the screen’s environment doesn’t affect the power computation (again just like a power meter).

(Nigel Van de Velde) #5

@Clint Westhoff

Ah yes, that would be easier to implement. Probably just a typo in their used formula then, should be addressed soon I guess…

(Clint Westhoff) #6

There was a thread on trainerroad in 2012 about what was, or was not, the correct formula: http://blog.trainerroad.com/kurt-kinetic-power-curve-virtualpower/

Apparently KK had published something that left out a number. At higher wheels speeds the difference can be around 10 watts or more.

(Jon Mayfield) #7

Hey guys, hopefully I can clear up a few things. We’re not doing “virtual power” in the classical sense for premiere basic trainers like a Kurt Kinetic. What we are testing here is zPower - our own little creation.

It is not sufficient for our game to simply take speed and convert to watts, because then your sprint efforts would be nullified…With virtual power a sprint might look something like this: 300w,350w, 400w, 440w, 480w, over some period of time as you slowly put energy into accelerating the trainer, when in reality you might instantly hit with 1000watts right from the start.

We’re still dialing this in, and it’s tricker than it sounds due to how seldom ANT+ sensors broadcast their data to us, but hopefully you guys at least notice that the data is somewhat responsive and that you could actually react to a competitor in the world.

(Nigel Van de Velde) #8

@ Jon Mayfield

Thanks for the clarification. Now I understand the logic why there is a devation from the classical virtual power for short bursts. You need to be able to react immediately to accelerations of other competitors and should not wait untill your trainer gets up to speed, right?

But shouldn’t this difference even out after let’s say 10-15 seconds. Even the most unresponsive trainer will be up to speed by then, so the relation between power on the pedals and rear wheel speed should be back to normal by then and z-power should be identical to normal virtual power.

I made the same power curve comparison as I did in a previous post between data gathered by a Garmin Edge 800 and the Zwift data. This time the activity did have some shorter sprints and higher power outputs. The sprints always started from low power, so they were typical accelerations.

So based on your explanation I would expect to see a big difference in power output for shorter bursts between the Garmin data (low responsivity of rear wheel speed) and the Zwift data (immediate responsivity of z-power). The difference should get smaller as time increases. This is the curve I found (green curve are garmin-derived data, purple is Zwift):


Strangely enough it seems to be the opposite. For short times (up to 10 seconds, so a typical acceleration/sprint) both curves line up perfectly and for longer times (when the rear weel speed already catched up with power output) the gap opens up… So far I only have one ride that contains sprint efforts, so I can’t check that this is a recurring power profile.

(Jon Mayfield) #9

Nigel, I sort of agree with your assessment, however latency doesn’t really show up in a mean maximal graph (which is what you’ve got there) and latency is what we are trying to optimize for.

Once we get this working as we hoped it would I suspect I do a blog post or article showing the advantages (and possible disadvantages) it has over standard virtual power. Another advantage is the opposite of a sprint - a rider could stop pedaling after spinning up the trainer to 40mph and vPower would still have the rider putting out watts for 15-20 seconds during the spindown. With our solution the power should go to (near) zero, even if you didn’t pair a cadence sensor.

We’re changing the formula a bit over the weekend and next week it’ll be slightly different and hopefully better. It’s a work in progress.

(Nigel Van de Velde) #10

This curve might be more useful in that case (again green is garmin and purple is zwift):


This shows acceleration (power delta in W/s). Here you can clearly see the intended acceleration difference for short bursts. Yet you also see the big difference for longer accelerations (e.g. at 1min, garmin versus zwift: 6W/s vs 12 W/S). Curves should allign better after +/- 10s I guess. I’ll test again for my coming rides, it’s quite addicitive anyway :smiley:

(Tim McCallum) #11

I have the Kurt Kinetic Road Machine trainer and also use TrainerRoad with virtual power calculations based on an Ant+ speed sensor.  

However the discrepancy I experience between Zwift and Trainer Road is unfortunately the other way:

My TrainerRoad wattage is 10-15% higher than what is output on Zwift (Would much rather this was the other way around)

I have the files to compare, however not the software to graph the comparison… Open to suggestions…