AutoTune Logic

Executive Summary

AutoTune is a state machine that iteratively tunes the PID controller by injecting "Twitches" (step inputs) and measuring the vehicle's response. It balances Performance (fast rise time) against Stability (minimal overshoot/bounce-back).

Theory & Concepts

1. The Twitch Test

The core mechanic is the Twitch.

1. Demand: Request a sharp 90 deg/s rotation.
2. Measure: Record the maximum rate reached and the "Bounce Back" (overshoot) when the stick is released.
3. Adjust:
   • If rise time is too slow $\rightarrow$ Increase P.
   • If bounce back is too high $\rightarrow$ Decrease D (or P if D is min).
   • If no bounce back $\rightarrow$ Increase D.

2. Aggressiveness (AUTOTUNE_AGGR)

This parameter controls the damping ratio target.

• 0.1 (High): Allows 10% overshoot. Results in a "locked-in" but potentially jittery tune.
• 0.05 (Low): Over-damped. Smoother, softer video, but less propwash rejection.

Codebase Investigation

1. The State Machine: AC_AutoTune_Multi::run()

The tuning sequence is implemented in libraries/AC_AutoTune/AC_AutoTune_Multi.cpp.
The sequence is:

1. RD_UP: Increase Rate D until oscillation or bounce-back is detected.
2. RD_DOWN: Decrease D until safe.
3. RP_UP: Increase Rate P to achieve the target response speed.
4. SP_UP: Increase Stabilize P (Angle P) to match the new rate loop speed.

2. Twitch Logic: twitching_test_rate()

Located in libraries/AC_AutoTune/AC_AutoTune_Multi.cpp.

• It measures meas_rate_max (Peak) and meas_rate_min (Bounce).
• Bounce Calculation:

  if (meas_rate_max - meas_rate_min > meas_rate_max * aggressiveness) {
      step = UPDATE_GAINS; // Too much bounce, back off
  }
  

3. Gain Updating: updating_rate_d_up()

Located in libraries/AC_AutoTune/AC_AutoTune_Multi.cpp.

• Iteratively steps up gains (tune_d += tune_d * tune_d_step_ratio).
• If the drone becomes unstable, it reverts to the last known good gain.

Source Code Reference

• Implementation: libraries/AC_AutoTune/AC_AutoTune_Multi.cpp

Practical Guide: AutoTune Success

1. Pre-Tune Requirements

• Filters: You must set up the Harmonic Notch Filter before AutoTune. If noise enters the gyro, AutoTune will result in very low P/D gains (under-tuned).
• CG: Center of Gravity must be perfect.

2. "AutoTune Failed"

• Cause: The drone couldn't reach the target rate (90 deg/s) even with max gains.
• Fix: Increase ATC_ACCEL_R_MAX / ATC_ACCEL_P_MAX (Physical torque authority is too low), or check for mechanical binding.

3. Post-Tune Adjustment

• Too Hot: If motors chirp after AutoTune, multiply P, I, and D by 0.8.
• Too Sloppy: If it feels loose, increase ATC_INPUT_TC (Input Time Constant) to smooth the stick response without changing the PID gains.