Given Randy’s sentiment from the other topic, I think my approach to a similar design would be to keep the autopilot in the loop during manual control. I’d write the microcontroller code to take input from the onboard controls and output the same RC protocol as used by whatever receiver I chose (ExpressLRS, most likely). Then I’d simply put a switch on the RC receiver signal line(s) to toggle between the actual RC receiver and my surrogate microcontroller “receiver.” I’d make sure to keep the “flight mode” channel switched to MANUAL mode in the microcontroller code and likely include a couple of layers of failsafes, including a big emergency stop button to kill all motion/power.
That approach avoids the need to write code for servo drivers and switch all of their signal lines to another source, with the added benefit of having datalogging for all manual control, which could be a helpful tuning tool for the automated side of things. It does have the drawback that the autopilot becomes a single point of failure.