I've arrived at a similar problem with my (non-smart) trainer using TrainerRoad - I seem to not be able to do ramp tests well, or at least well enough to arrive at an FTP figure which when accepted provides taxing enough training rides in the base and build plans. I wonder whether TrainerRoad's virtual power curve for my trainer overestimates at low power and underestimates at higher power (this would tie in well with the fact that I really struggle to complete VO2max sessions, but find myself upping the intensity for FTP and sub-FTP sessions.
It might be that I'm just bad at the psychological aspect of the ramp test, specifically that they're open ended - at least with a 20 minute FTP test, there's a defined end point, which I can mentally work towards. With a ramp test, even if I force myself through one step, there's always another one, and then another one, so I think I end up quitting early for that reason.
I'm going back to 20 minute tests as my chosen measure of FTP, but having said that I won't always accept the figure it spits out as gospel, unless it aligns with my estimated FTP based on how I finished the previous training plan i.e. if I finish a training plan at a given FTP, and feel I have another 10 or 20 more watts in the tank, then unless the 20 minute test indicates that I'm 20 watts up, I will probably manually set my FTP to FTP+20 and go from there. My reasoning is that there might be plenty of factors which adversely affect an FTP test, and I know from using TrainerRoad for a few years now to trust my feel on whether a workout is too easy, too hard, or about right based on my FTP.