Flight-controller tuning for a survey quadcopter in simulation

Skyline Survey flies a 450 mm X-frame quadcopter weighing 2.1 kg for mapping farms and building sites. The company has fitted a new 450 g sensor payload under the frame. Since then, the drone wobbles in roll at hover, overshoots on every turn and drifts about 2 m in gusty wind, which ruins the mapping overlap. The pilot has been lowering gains by trial and error in the field, and on one flight the drone nearly hit a mast.

The operations lead has banned field tuning. All tuning must now happen in simulation first, with written targets and logs. The targets are: - rate loop phase margin at least 45° and gain margin at least 6 dB in the linear model; - a 10° roll attitude step with no more than 10% overshoot and settling within 0.5 s, in simulation; - position hold within ±0.5 m in steady 8 m/s wind with gusts, in simulation; - no visible oscillation at hover with the payload fitted, and no motor output saturating during the test manoeuvres.

You will write the requirements and test plan, build a linear model of the roll axis from the drone's physical data, design the rate and attitude controllers with Bode and step analysis, set up PX4 software-in-the-loop simulation with a vehicle model changed to match the heavier drone, fly a scripted test sequence, analyse the logs, tune the loops in order, test in wind, and produce a parameter file and a report. You need a computer that can run Ubuntu or WSL2 with PX4 and Gazebo. No drone is needed, and none should be flown as part of this project.

Done looks like a model, plots and logs that show each target met, a parameter file with every change explained, and a first-flight checklist the pilot will follow before the new parameters ever touch the real drone.