What's the default ATC_ANG_RLL_P value, and what should it be after AutoTune?

Default is 4.5. AutoTune is supposed to raise it — typical post-tune values land between 6 and 10 for most multirotor builds. If AutoTune produces a value below 4.5, the run failed (usually because of vibration or noise) and the gains shouldn't be trusted.

How do I know if AutoTune actually converged?

Three checks. First, ATC_ANG_RLL_P came out above 4.5 — if not, AutoTune gave up. Second, ATC_RAT_RLL_D didn't bottom out at AUTOTUNE_MIN_D. Third, the test twitch sequence completed without the autopilot reporting 'Failed to level' messages in the MSG log. If all three are clean, the tune is real.

Should I tune all three axes at once with AUTOTUNE_AXES=7?

On a new or risky airframe, tune one axis at a time. Roll first, then pitch, then yaw. A failed tune on one axis doesn't compromise the others, and you can revert per-axis with the parameter file. On a known-good airframe with a known-good tune as a starting point, all-three-at-once saves flight time.

Why does my AutoTune keep failing?

By a wide margin, the most common cause is high-frequency motor noise reaching the gyros — the harmonic notch filter isn't configured. Run a Filter Review FFT on a hover log first, configure INS_HNTCH_* against the motor peak, and AutoTune will almost always succeed on the second attempt. Vibration above 30 m/s² VibeZ is the second most common cause.

Does AutoTune work for QuadPlanes?

Yes for the VTOL axes (Q_A_RAT_RLL_*, Q_A_ANG_RLL_P). Use Q_AUTOTUNE mode. The forward-flight controller has its own AutoTune (ArduPlane), tuned separately. The two tunes are independent and shouldn't be combined into a single flight.

Can I AutoTune a heavy or asymmetric payload?

Yes, but only with the payload mounted and at flight weight. AutoTune samples the flight envelope you ran it in, so a tune on an empty airframe will be wrong for a 5 kg payload. For mission-variable payloads, tune at the heaviest configuration and verify in flight with lighter loads — the heavy tune will usually feel slightly loose with less mass.

How to Tune ArduPilot PID Controllers on a Quadcopter

TL;DR: ArduPilot Copter PID tuning lives in three parameter families per axis: ATC_RAT_RLL_P/I/D (rate loop), ATC_ANG_RLL_P (angle loop), and ATC_ACCEL_R_MAX (output limit). The fastest path to a usable tune on a new airframe is AutoTune (AUTOTUNE mode, with AUTOTUNE_AXES set per-axis); the cleanest way to confirm a tune is to plot RATE.RDes against RATE.R and check that the achieved rate follows the desired rate within roughly 10% with no sustained oscillation. The 4.5-default, ~6 tight, ~10 aggressive convention for ATC_ANG_RLL_P is your sanity check — AutoTune results under 4.5 mean the run failed and the airframe needs more attention.

What you're actually tuning

ArduPilot Copter runs a cascaded controller: angle in, rate out, motor in. Each loop has its own gains:

Angle loop — ATC_ANG_RLL_P, ATC_ANG_PIT_P, ATC_ANG_YAW_P. Converts a desired angle (from your stick or from a position controller) into a desired rate. Default 4.5.
Rate loop — ATC_RAT_RLL_P, ATC_RAT_RLL_I, ATC_RAT_RLL_D (and the pitch / yaw equivalents). Converts a desired rate into a motor mix output. The D term is the term most likely to need careful adjustment.
Output limits — ATC_ACCEL_R_MAX caps the maximum rotational acceleration the controller will command, in centidegrees per second squared.

You almost never need to touch all three loops on a new airframe. Get the rate-loop gains right with AutoTune and the angle-loop gain follows.

The AutoTune workflow

Run AutoTune on a calm day in a large open area, with the airframe in good mechanical condition (balanced props, low VIBE):

Set up the flight. AUTOTUNE_AXES = 1 tunes roll, 2 pitch, 4 yaw. = 7 tunes all three. Start with one axis at a time on a new airframe so you can revert if anything goes wrong.
Pick aggressiveness. AUTOTUNE_AGGR = 0.1 is the default (the looser end). 0.075 is medium-tight; 0.05 is aggressive. Start with 0.1.
Take off in AltHold. Climb to a safe altitude. Face the vehicle so the wind blows from the side of the axis you're tuning — pointing into the wind biases the tune.
Switch to AUTOTUNE mode. The vehicle will twitch ~20 degrees in each direction on the axis being tuned. It will spend several minutes per axis.
Land without disarming. Switch out of AUTOTUNE and back in to test the new gains. If the airframe feels right, land and disarm with the AUTOTUNE switch high — the gains save permanently. Disarm with the switch low and the autopilot reverts to the original gains.

Reading the tune in the log

The RATE message is your verdict. It records both the desired and achieved rate per axis:

RATE  TimeUS  RDes  R  ROut  PDes  P  POut  YDes  Y  YOut  ADes  A  AOut  AOutSlew

What to plot in Mission Planner:

RATE.RDes and RATE.R on the same axis — a healthy tune has the achieved rate tracking the desired rate with under ~50 ms lag and no overshoot.
RATE.ROut — the normalised motor output for the roll axis. Saturation at ±1.0 for sustained periods means the controller is asking for more than the motors can deliver.

In MAVExplorer:

MAV> graph RATE.RDes RATE.R
MAV> graph RATE.PDes RATE.P
MAV> graph RATE.ROut RATE.POut

Telling a good AutoTune from a bad one

The MethodicConfigurator project documents the practical sanity checks for an AutoTune result:

AutoTune is considered failed if ATC_ANG_RLL_P ends below 4.5. That's the default; tuning is supposed to raise it.
It's also considered failed if ATC_RAT_RLL_D ends at AUTOTUNE_MIN_D — the algorithm bottomed out the D term, usually because mechanical noise was too high.
Higher ATC_ANG_RLL_P implies a tighter response. Most well-tuned multirotors land between 6 and 10. Above 12 the response can become twitchy in manual modes; check stability in Stabilise before committing.

If AutoTune fails, the cause is usually vibration (VIBE.VibeZ above 30 m/s² corrupts the algorithm's measurements), a motor imbalance, or a frame structural problem. Fix those before retrying.

Manual tuning when AutoTune won't converge

When AutoTune fails on a difficult airframe, the manual sequence is roll first, then pitch, then yaw, then the D-yaw refinement:

Start with ATC_RAT_RLL_D = 0.001, raise it in small steps. Stop just before the motors start to whine in hover — that's the D-onset of high-frequency noise.
Raise ATC_RAT_RLL_P until small stick inputs become responsive but you don't see oscillation. Typical landing: 0.1–0.2 for small frames, 0.05–0.1 for heavy frames.
ATC_RAT_RLL_I follows P proportionally (the convention is I = P).
Raise ATC_ANG_RLL_P until the angle response feels right.
Repeat for pitch (typically the same gains as roll on a symmetric quad).

This is slower than AutoTune and only useful when AutoTune can't run. For the official procedure, follow ArduPilot's AutoTune documentation or the MethodicConfigurator tuning guide.

When the tune doesn't fix the problem

If you've run AutoTune and the airframe still oscillates, the cause is usually downstream of the PID gains:

Harmonic notch not configured. Motor noise above the gyro bandwidth aliases into the rate loop and AutoTune can't tune through it. INS_HNTCH_* needs to be set before tuning on most modern airframes.
Mechanical vibration. See our VIBE message guide; AutoTune assumes the IMU signal is trustworthy.
Wrong frame configuration. FRAME_CLASS or FRAME_TYPE set incorrectly produces mixed-up motor outputs that no PID can fix. See our takeoff-flip post.
Battery sag during tune. AutoTune samples a single flight envelope; if the pack is on its way down during the run, the gains may not transfer cleanly. Use MOT_BAT_VOLT_MIN / MOT_BAT_VOLT_MAX scaling.

When LogHat helps — and when it doesn't

Upload the tuning flight log and LogHat plots RATE.RDes versus RATE.R with overshoot and lag metrics already computed, flags any axis where the achieved rate is saturating RATE.ROut, and shows the parameter snapshot at takeoff so you can confirm the post-AutoTune gains are the ones flying. What we can't tell you is whether a heavier payload tomorrow will need a different tune — that needs you to fly another envelope and re-evaluate.