I'm very rusty on this but I did set up a PID a few decades ago. In that case we modeled the system dynamics and calculated the PID parameters from that, but as a control we also set it up by trial and error and that seemed to work OK (and gave me much less of a headache).
Is the system small / light enough that you can hold it and move it around and feel how the control system behaves? In that case I think you should be able to set it up by simulating a disruption from equilibrium, starting with all gains set to zero and then introducing proportional, derivative and then integral feedback in that order. The crucial thing is to have the vehicle attitude under your control while you do this, not letting it move around unrestrained.