After arming and throttling up, they run smoothly. I tried increasing the deadband without result. What can be causing this?
I noticed that the effect is present when using a 2.4GHz RC radio for control inputs to the Navio2, but the effect is much worse when attempting to use a joystick and using telemetry for the control channel from Mission Planner.
I even added an extra ground with no result.
When I check the “Status” view I can see the roll and pitch values shifting about even when the quad is sitting still. Could this be a clue? Maybe it thinks its flying?
Are you using the oneshot protocol on BLHeli ESCs?
The signal wire to the ESC should always be twisted with a signal ground wire.
In Cleanflight/Betaflight you can lower the min command after calibration, but I don´t think its possible on AC.
Try to calibrate the ESCs with a higher min RC throtlle value maybe 1080µs.
The gap between the disarmed value (I think 1000µs) and the calibated min ESC value could be too small.
Gregor
Interesting speculation regarding the signal wires picking up interference. I will try moving them farther away from other wiring (mostly power wiring). I will also try twisting them as much as I can being that they are flat three-conductor cables.
I tried various spacing between throttle off and MOT_SPIN_MIN, MOT_SPIN_ARM, etc. without success. It got better when I set the ESC endpoints manually using the Castle Link - I’m using Castle Creations DMR ESCs. I did find that the chatter is minimized when the system is set to use oneshot125. The DMR ESCs auto-detect the signaling type.
i’ve used oneshot spider esc with blheli , i’ve just increase min pwm value just a bit 
and it’s gone