Hi @Airon,
If your motors are all positioned and rotating as expected, and you’re not setting a pitch target for the vehicle, then my main thought for a cause for unexpected upwards pitching would be if the weight is balanced too far backwards relative to the centre of buoyancy. There’s some more discussion on that here, but you may want to try moving some of your ballast weights further forwards if possible.
ArduSub’s Extended Kalman Filter (EKF) sensor fusion algorithm resets its state if it gets too uncertain, which can happen if the sensors are poorly calibrated or if there’s too much integration drift from a noisy sensor (e.g. that is common when the ROV is sitting still instead of moving, especially if the gyroscope calibration was performed while the vehicle was on a moving platform or boat). If it’s happening frequently during operation then it’s likely worth redoing your sensor calibrations, and making sure not to be too close to large magnets or metallic structures when doing so.
Hmm, I’m not aware of something specifically showing this, but it does seem like it would be useful. I’ll raise it internally as something we should look into providing, but don’t know how long it will take us to do that. Thanks for the suggestion/question 
Depth hold mode operates in the vehicle’s frame of reference (rather than the “world frame”), so I’m aware that if you’ve explicitly set/commanded a non-flat pitch angle and move forwards then the depth target will change to match the current depth when the controls are released. That said, if the pitch is set to flat (which is normally the case) then I believe the depth target should only get changed when a depth increase or decrease is commanded, not from horizontal motion input commands.