Fix physics simulation on the flat

Thanks for that info. I think some runs where you only change 1 variable at a time would provide even more insight. For example:
-different weight, same power, same height.
-same weight, different power, same height.
-same weight, same power, different height.

1 Like

I don’t think your conclusion is supported by the evidence, and using the evidence I reach the opposite conclusion.

Comparison of Zwift with Olympic TT data

In the Olympics the W/kg were (5.75, 5.74, and 5.70). Here the spread is (5.5, 5.42, and 5.36) - almost three times as wide as spread and smaller. So we’d expect the heavier rider to perform better on the flats compared to the Olympics (because the are producing relatively more power ber kilo).

Also the Olympic course is hillier - again we’d expect the heavier rider to perform relatively better in this Zwift trial agaisnt the Olympics.

The absense of wind speed effects in Zwift would also be expected to favour the heavier rider.

These suggest that in this comparison the heavier rider in Zwift will go significantly faster than his lighter competitor relative to the performance in the Olympics

The measured speeds in Zwift are (46.68, 45.51, and 44.76) kph - first goes 1.17 kph faster than second, and 1.92 kph faster than third. In the Olympics the differences in average speeds were 2.19 kph, and 2.31 kph.

Therefore on a scenario where you’d expect the performance of the heaviest rider vs the lightest rider compar, it is measured to be significantly worse based on the difference in avarage speeds.

Preliminary Conclusion: Zwift simulates heavier riders more harshly when compared to an example of an external event…

Comparison of Zwift and the Kreuzotter Model (KM).

KM used the triathalon setting and an 8kg bike for this comparison.

Speed difference between actual Olympics and KM are (-1.46, -1.45, -1.87) - pretty consistent, and in line with the hills and other real world effects for the Olympics leading to an expected slower speed for all riders.

For Zwift vs KM-Model the differences are (-0.82, 1.81, 0.86) - heavier riders go substantially slower than KM predicts for these rides, and lighter riders go substantially faster than KM predicts for these rides.

Conclusion:

Using this limited sample of rides to compare Zwift rides against both a real world event and a model of the cycling power equation suggests that Zwift gives a substantial advantage to lighter riders with their algorithms.

4 Likes

The point was not to see if Zwift simulate speed the same as IRL the point of the test was to see if the comment that Zwift would simulate the opposite to the Olympic data is true or false.

Those Olympic numbers was taken from normalized power, I used average power.

The Olimpic riders did not ride at constant power, look at Cancellara he did a peak 5 min power of 547w.

There are to many differences to draw any conclusion apart from the fact that the comment made above is not true.

If I wanted to test IRL speed compare to Zwift I would use better sample data. :rofl: But as I said that was not the point of the test.

Not even KM can simulate a route with so many changes in elevation. You will need to use the raw KM formulas and apply it for every second using the power and grade as input.

1 Like

The sole conclusion in your post was

“Zwift does a good job modeling bike speed for different weight and length riders.”

I don’t think the evidence presented by you supports this conclusion

2 Likes

If you read it out of context yes you are correct the conclusion should be changed so that it can be read out of context.

I will update that. Thanks

Edit: I updated that conclusion so It can stand alone.

I’ll just add this here:

Please go to the 13:00 minute mark onwards for the relevant comments and insight that Zwift modelling feels incorrect.

Theres also another video he does where he specifically discusses one of his real world friends that he races against and there is no way IRL that this person would beat him because of the watts difference (I’m trying to find the video and the timestamp)
UPDATE: found it here:


go to the 40:00 minute mark

This is Alex Dowsett, current world tour pro, former world hour record holder and time trial specialist:

6 Likes

Once Alex buys a fan I’ll pay more attention to his complaints.

5 Likes

:rofl::rofl::rofl: Fair enough!

1 Like

In real life, I always Tt faster than my lighter friends, which 10-20 kg lighter than me.
Even in my draft they always suffer.
But in zwift lighter guys can keep up no problem, as long as they can give same watt per kg.
50kg @ 4w/kg (200 watt) can stay with 90kg @4w/kg who put out 360 watt.
In real life it hard to do that on flat road even in draft

6 Likes

So this thread was really interesting to me as a) I like stats and b) I feel like I am way slower in Zwift than IRL with the exception of hills which feel about right.

So I recently did two laps through the ocean tunnels at somewhat consistent power (using garmin vector 2s as the power source) on the basic TT bike with enve ses 3.4 wheels.

I then pulled the average power, speed, elevation and cadence (why not) from Strava for an 1,100m section that cut off any extra speed from the descent.

So for me (180 cm, 89 kg) using the calculator site from above with measured cadence, 20C temp, -5m elevation , 0 slope, no wind, and assuming a 9.5kg triathlon bike with “narrow racing tires”

Power 157W @ 81 rpm
Zwift 2:10 @ 32.2 kmh
Kreuzotter.de 2:08 @ 31.0 kmh

Power 275W @ 94rpm
Zwift: 1:43 @ 40.5 kmh
Kreuzotter.de: 1:43 @ 38.3 kmh

Despite being a very short interval (and the lack in precision on length from strava), that is surprisingly close, suggesting that any discrepancies in expected speed (for me at least) could be coming primarily from slopes (up or down) or the alternate road surfaces (why so much dirt zwift?). Or that after a drastic reduction in fitness from parenting and a long time off the bike I’m just way slower than I wish I was :wink:

How can the model predict a quicker time with a lower average speed? Is there a difference in assumed route length?

Has to be the rounding error from what Strava tells me I selected (to the nearest 100m) and the actual distance in zwift. As its a short segment I expect some error, but it’s the only true flat I could find.

Today we can see how each one of the light and short climbers is going to literally destroy Fabian Cancellara on a completely flat TT, even putting average 100W less:

Today we will see the evidence of the ridiculous and absurd advantage of being short and light in Zwift, which is huge on all terrains in addition to the obvious (ascents) that IRL does occur.
The reason so many cyclists do the “Zwift diet”.

The model should be corrected because it is absurd that the aerodynamics become lower and lower as a rider gets short and light, such that 155cm and 48kg could feel even tailwind… :joy: :joy:

Please Zwift programmers: ASK FABIAN.
Sure he will help.

3 Likes

really like your effort in this thread here @Susun_Corda !
this w/kg bogus should be unnerved.

next if this is fixed is the powermeter/trainer mess so many monitoring devices measure anything but real values …

Of course, next would be display power device in classification as Rouvy or Bkool actually DO.

Sorry to be all grumpy but as a lighter rider of average height (60kg/176cm) my distinct impression is that Zwift unfairly favours heavier riders, particularly on any kind of rolling terrain.

Ultimately the fact that we’ve come to opposing conclusions probably means that the physics model isn’t actually that bad.

Never seen IRL that if on the flat you (60kg) put 50W less than a 70kg rider, you will be faster, as is happening on Zwift.
Try yourself in real life, riding next to a cyclist heavier than you and then put 50W less.
Everybody knows what will happen!
You sure know this.

Never seen that in a 12km descent, average -8.5% slope, without braking,… a 7kg lighter rider (75kg VS 82kg) will be only 7 seconds slower thatn the heavier, as actually is happening on Zwift, and Zwiftinssider have demostrated.

So, lighter are CLEARLY favoured compared with how things are in the reality. :wink:

2 Likes

wake me up before you gogo

Hi, I just wanted to add my two cents of empirical data gathered on Zwift races held by my squad over two consecutive days.

The first is a TT held on Greater London Flat (no drafting) and the second one is a climb up Ven-Top on regular bikes.

I have plotted average speed vs average watts/kg for all racers, and included a linear regression line for each chart.

As can be seen, there is quite a bit is dispersion along the trend line for the flat course, while there is an almost perfect correlation for the climbing course.

Being that this is the same group of riders, my conclusion is that average power can be misleading as a speed predictor on flat terrain, due primarily to momentum effects, which penalizes variance in power.

Weight does not play a big role on a flat road. same as in real life.

1 Like