Initial feedback - computrainer


(W ish I was Outside) #1

Hi, just did my first ride and thought I’d provide my initial feedback. I mainly use Ergvideo, and occasionally use real course videos for spinscan time, so that’s my basis for comparison.

First of all, initial reaction is that the graphics are great and the speed/framerate is pretty good. I like the little touches like the butterflies and deer at the side of the road.

The overall experience is less immersive than the Ergvideos, I think perhaps this has to do with the difference between real video and the rendered 3d graphics, and the animation of the rider and rider POV (there should be some natural sway to the camera point and it should increase with effort, because the rider will rocks and move slightly and to a greater extent as their effort level increases; should be relative to rider FTP so that weaker rider would lose form and sway more at a lower wattage).

The animation for drafting really needs some fine tuning, it was really awkward and clunky feeling and I didn’t feel like I could really cooperate with the AI riders. They rode fast downhill and slow uphill, there should be a mix of ‘riding style’ within the AIs to have climbers, descenders, riders who will push on the flats, as well as different styles for attacking and sitting in etc. There should be some ‘Cato’s’ amongst the AIs like Cato in the Pink Panther movies, who will hide in a pack of AIs and attack when you least expect it! The AIs were unagressive, nobody hiding ready to attack when you come by, and basically were no fun at all. There needs to be some wolves in with the lambs or they might as well not clutter up the road.

The lack of support for calibration via the computrainer head unit is a show stopper for commercial use in my opinion - fine for beta to calibrate first outside the application but the competition allows full control from the head unit which minimally means on the fly calibration, starting/stopping the simulation. Bonus for being able to skip forward/backward within the course so that you can easily repeat a hill (or skip one as the case may be), just add a bit of hysteresis to the response to a button push (e.g. plus to skip forward minus to skip back) before resuming the simulation so that multiple presses can skip back e.g. 200m at a time.

I have a pretty good idea of my HR for a given power level as I’ve had lab VO2 max testing and regularly ride with power, HR on Zwift was high for the power output indicated in the simulation. The effect is particularly noticeable at low power levels e.g. 200W where HR should be low it was still quite high, this makes me think that perhaps something about the calibration compensation wasn’t working quite right. As watts got higher, the HR became closer to what it should be, but still reading high. I usually ride at about 3.00 calibration setting since at lower clamping force wheel slippage starts getting to be a problem whenever watts are high and wheel speed is low. For example, my slowest lap was 8:02 at 257W and my average HR on the lap was 172. That is crazy high HR for 257W, 250W is ‘ride al day’ sort of wattage level should be more like a HR in the low 150s not 172 which is low theshold for me. The speeds themselves look reasonable, what seems off is the wattage as compared to the speeds and heart rate. And since I train mainly based on wattage, everything just seems off.

Some of the rider animations appear to be floating above the ground level rather than on it; this is noticeable when cresting a hill and observing oncoming riders that are cresting in the opposite direction.

All in all, a good start but really the support for controlling the simulation from the computrainer head unit is not a nice to have, it’s a must-have.

I will definitely be doing more rides on this, keep up the good work.


(W ish I was Outside) #2

Setting lower tension ~2.3 seemed to produce a better result in terms of the power/HR correlation. Today in trying to figure out a reasonable protocol for starting zwift and being able to calibrate without leaving the bike to interact with the computer to start Zwift, I started Zwift but did not have it scan for the computrainer, I rode and calibrated the Computrainer then (using a windows IR remote control) scanned for the computrainer. It found it but somehow got confused and before I could start the simulation it completely lost track that the computrainer was there and would not find it on subsequent scans. I had to stop and restart Zwift (but not the Computrainer) and then it all started OK. I did not attempt to calibrate between starting Zwift and scanning the second time. Having to get off the bike to go over to the computer is a real PITA. I’m sure mine is a very common setup, desktop computer connected to a short throw projector displayed on a big screen in front of me, so the computer is not in front of me or in arms reach in any way. I think this would be typical at least for a lot of people and definitely for anyone running multiple trainers off the same computer (I’m just running one in my basement).

Experimented a bit more with drafting, it was hard to tell whether the difference it made was realistic, my sense is that this area needs work, that the effect in the sweet spot is less than in real life and the effect ‘not in the sweet spot’ is perhaps exaggerated.

The simulation seems to have a hard time detecting the difference between sitting and standing based on cadence and power; I would be standing climbing at around 400W and it would show me sitting down even though because I was standing the cadence dropped down into the 70s. Usually if I’m sitting and climbing sitting down with a deliberately low cadence, I’d still be at 80rpm not down in the 70s. If I’m just coasting along then my power would be low, it would never be under 300W if I’m standing so that would be one way to tell the difference between standing and sitting if the cadence is low i.e. >300W and cadence below 78 = standing. On the other hand I think it did a pretty good job of detecting on the rivet sprinting vs just regular riding.


(W ish I was Outside) #3

Oh, and some of the heuristics that determine the rider animations should probably be scaled based on the rider’s declared FTP vs actual current power output.