diff --git a/autotune/README.md b/autotune/README.md index 6f767c6..640ca7c 100644 --- a/autotune/README.md +++ b/autotune/README.md @@ -29,3 +29,90 @@ python3 autotune.py ![image](https://github.com/user-attachments/assets/fcdf5c25-d92d-4487-9736-e77f6576d180) +# Tuning Worflow + +## Fixed-Wing Rate Tuning + +**Prerequisites**: + +- Requires valid airspeed measurement on vehicle +- Correctly set FW_AIRSPD_TRIM + +### Flight Maneuvers + +Using default PX4 gains. Perform these maneuvers separately for each axis: roll, pitch, and yaw. + +1. Requires valid airspeed measurement, and a correctly set FW_AIRSPD_TRIM. +2. Fly in stabilized mode and start from level flight. +3. Apply a manual sine chirp input (sine wave with increasing frequency) on the selected axis. Avoid any other control inputs simultaneously. +4. For pitch and yaw: The amplitude of the input should not be too high, but should be enough to overcome noise. +5. Repeat the maneuver a few times if necessary to capture good data. + +### Using the Tool + +#### Load the flight log and select the relevant tuning loop (roll, pitch, or yaw): + +![image](https://github.com/user-attachments/assets/f962a004-eae8-433f-9b7e-adc618dcbeae) + +#### Window selection: + +1. Select the portion of the log where the maneuver occurred. +2. Ensure the window starts from level flight and covers the full maneuver. +3. (Experimental) Use the coherence plot to check whether your input-output data have a strong enough linear relationship at the desired frequencies. For fixed-wings: \ + i. We are interested in frequencies between 0.5-10Hz. \ + ii. Values > 0.6 are sufficient. \ + iii. If the coherence function is oscillating dramatically, the identified model may be unreliable at those frequencies. +4. Once you have identified the window, load the selection into the tool. + + + + + + +
+ Good window selection
+
+ Starts from level flight and covers the full maneuver. Coherence is sufficient. + 		+ Bad window selection
+
+ Does not isolate the maneuver. Coherence is insufficient. +
+ +#### Model Verification + +1. Check that the estimated dynamic model correctly tracks the actual aircraft output. +2. If not, adjust the system order (number of poles and zeros) and/or delay parameters. +3. Verify that the system is minimum-phase (i.e., zeroes lie inside the unit circle). + + + +#### Tuning Gains + +1. Set all gains to zero +2. Increase the P-gain until the step response shows oscillations. Then reduce the P-gain by approximately 50%: + + Increase P-gain
+ + + Decrease by 50%
+ + +3. Adjust the I-gain to achieve a good step response with acceptable steady-state error. + + ![image](https://github.com/user-attachments/assets/78f3bca5-2642-44e2-b3f2-7444fc4ffa40) + +4. Fine-tune both gains as needed. At the 1 second mark, a disturbance is injected into the system. The tuned response to this disturbance is visible in the step response plot. + +5. Use the Bode plot to confirm that the resonant frequency peak remains below 0dB. + + ![image](https://github.com/user-attachments/assets/86d99295-43a8-47a9-b052-324ab5ac3082) + +### Applying the Gains + +1. The gains shown in the parallel form correspond to the PX4 rate controller parameters. + ![image](https://github.com/user-attachments/assets/2db77238-bd33-4454-af1f-a38901571e88) + +2. If the identified parameters are strongly different to the currently set ones (more than 30% for multiple axes), an incremental application while flying is recommended (not prior to takeoff, during takeoff, or during landing). + +3. For all loops tuned with the strategy outlined in this guide: increase the respective integrator limit parameters to 1 (FW_RR_IMAX, FW_PR_IMAX, FW_YR_IMAX).