Not so much that they’d override, but rather that there’s no way to tell the autopilot to prefer control from one source over another when it’s receiving conflicting control commands. You could potentially create a proxy program on the RPi to do that for you (or on your topside computer if both QGC and your control program are running on the topside), but that may not be trivial, especially if you’re wanting to maintain the existing communication latency.
Yes, and telemetry from the autopilot (running on the Pixhawk) also gets forwarded by the onboard computer (RPi) to the topside computer.
Fair enough. Logic wise that should be reasonably straightforward with a combination of the existing depth hold mode and a process like the one described here. The control switchover is a separate issue though, as discussed above.
Ooh cool - sounds fascinating! 