Tuning an aircraft whose weight changes

The problem

An aircraft that weighs 38 kg on take-off and 22 kg on landing is, in control terms, two different aircraft.

Its thrust-to-weight changes. Its hover throttle changes. Its rotational inertia changes. Its centre of gravity may move. And the controller — which is a fixed set of numbers, decided on the ground, before any of this happened — has to fly both of them.

Worse, the transition is continuous. A sprayer does not fly one mission at 38 kg and a second at 22 kg. It is a slightly different aircraft every minute of the flight, sliding through every intermediate state as the tank empties. There is no single condition to tune for, because the aircraft never sits in one condition for long.

This is not an exotic corner case. It is agricultural spraying, cargo delivery, cinelifting with a camera package that changes between jobs, survey aircraft that fly the same airframe with three different sensor pods. If you fly heavy for a living, you fly a changing aircraft.

What actually changes with weight

Three things, and they are not the same thing.

Thrust-to-weight and hover throttle. More mass means more of your available thrust is spent simply not falling. Your hover throttle rises, your remaining margin for climb, for gusts, and for the differential thrust the controller needs to make corrections shrinks. At the empty end you have margin to spare; at the loaded end you may have very little. This deserves its own treatment — see hover throttle and thrust margin — but hold on to one consequence: the authority the controller has to correct with is not constant across the flight either.

Rotational inertia. This is the one that gets stated lazily, so state it properly: rotational inertia is not proportional to mass. It depends on where the mass is. A payload bolted on the centreline, close to the roll and pitch axes, adds mass and hardly any rotational inertia — it makes the aircraft heavier without making it much harder to rotate. The same mass spread out along the arms, or slung below on a long mount, adds a great deal of rotational inertia, because inertia scales with the square of the distance from the axis. A heavy battery mounted low and forward changes the inertia about all three axes and moves the centre of gravity as well.

So "I added 16 kg" tells you almost nothing on its own. You need to know where the 16 kg went.

The effective gain of the control loop. This is the key insight, and it is worth deriving rather than asserting.

The controller does not command an angular acceleration. It commands a torque — it asks for a differential in thrust between motors, and that differential produces a torque about an axis. What the aircraft does with that torque is governed by one line of mechanics:

angular acceleration = torque ÷ moment of inertia

The PID gains sit on the torque side of that equation. The inertia sits on the other side, and it is a property of the airframe, not of the tune. So for a given error, the controller produces the same commanded torque regardless of what the aircraft weighs — but the aircraft's response to that torque changes with its inertia.

Load the aircraft up and its inertia goes up. Same gains, same commanded torque, less angular acceleration: the loop responds more weakly. The aircraft feels soft.

Now run it the other way. Take the tune that was right when the aircraft was heavy, and empty the tank. The inertia falls. Same gains, same commanded torque, but now more angular acceleration for every unit of error. The loop is, in effect, running at a higher gain than the number in the configurator suggests — and the aircraft that was nicely damped at 38 kg oscillates at 22 kg.

Nothing in the firmware changed. The plant changed underneath it.

The answer: tune for the light case

Here is the consensus among people who do this for a living, and it is counter-intuitive enough that it is worth spelling out the reasoning rather than just the rule.

Tune the aircraft in its lightest configuration.

The argument is an asymmetry between two failure modes.

An over-gained aircraft oscillates. Oscillation is a stability failure. It grows, it feeds itself, it shakes the airframe, it cooks motors, it can tear a payload mount or snap an arm. On a 22 kg machine with 30-inch props, an oscillation is not a handling complaint — it is a destroyed aircraft, and possibly worse.

An under-gained aircraft is sluggish. That is a performance failure. It responds slowly, it holds attitude loosely, it feels lazy. It flies badly. But it flies.

Those two outcomes are not remotely equivalent, and you do not get to avoid both. You have one tune and two aircraft, so one of the two conditions is going to be mis-tuned. Choose which.

If you tune loaded, the gains are correct at 38 kg — and as the tank empties, the inertia falls out from under the tune, the effective gain rises, and the aircraft walks toward oscillation as the flight progresses. The mission ends in the dangerous condition. That is exactly backwards.

If you tune light, the gains are correct at 22 kg — the most agile, lowest-inertia, highest-effective-gain condition. Load the aircraft and the inertia rises, the effective gain drops, and the aircraft becomes softer. It walks away from instability, not toward it, and as the mission burns off payload it converges on the tune you actually validated.

Tune for the condition that would otherwise be over-gained. Accept sluggishness at the loaded end. Sluggish is survivable; unstable is not.

What that costs you

Be honest about the price, because there is one.

A loaded aircraft flying a light tune is soft. It will be slow to pick up a commanded rate and slow to stop. It will hold attitude less crisply, and it will let itself be pushed further off target before it does much about it. In wind it will feel vague — you will command a correction and wait for it.

For a working aircraft on a survey line or a spray run, that is usually fine, because the mission does not ask for aggression. It asks for stability while a tank empties. But do not tell yourself the loaded aircraft flies as well as the empty one on the same tune. It does not, and if it feels like it does, you have probably tuned closer to the loaded case than you think — which means the empty end of the flight is closer to the oscillation boundary than you think.

Sloshing liquids are a different problem

Everything above treats the payload as a rigid mass rigidly attached. A tank of liquid is neither.

Liquid moves. When the aircraft banks, the liquid does not bank with it instantly — it flows, with a delay, and it keeps flowing after the aircraft has stopped moving. That is a mass that shifts in response to the aircraft's attitude, with a lag, which is precisely the recipe for a coupled oscillation: the controller corrects, the liquid moves, the moving liquid produces a moment, the controller corrects again. If the frequency at which the liquid sloshes lands anywhere near the frequencies your attitude loop operates at, the two can feed each other.

And the worst case is not the full tank. A half-full tank is worse than either a full one or an empty one, because a full tank has nowhere to slosh to and an empty one has nothing to slosh. Somewhere in the middle is a mass with maximum freedom to move.

This is why baffles exist. Dividing the tank into compartments limits how far the liquid can travel and raises the frequency at which it moves, which is the practical mitigation. Be honest with yourself about the limits here: slosh coupling is a genuinely hard problem, it is not something you tune your way out of with PID gains, and the fix is mechanical. If your aircraft is well behaved empty and well behaved full but misbehaves at half a tank, stop looking at the tune.

Centre of gravity changes too

A payload that is not on the centreline does not just add inertia. It moves the centre of gravity — and an aircraft whose CG is not under the centre of the motor plane can only hold station by tilting slightly, so that its thrust vector passes through the new CG. The flight controller holds that tilt with a permanent, integrated correction, and pays for it by running the motors on the heavy side harder for the entire flight.

You have seen this fault before under two different names. It is the drift you cannot calibrate away, and it is the one motor that runs hotter than the other three. On a payload aircraft it arrives from a third direction, and it has a nastier property: it changes during the flight. A tank that drains from one end, or a payload that is released asymmetrically, moves the CG as the mission runs. The standing correction the FC needs is not constant.

Balance the aircraft at both extremes. If it balances loaded and tips empty, the mounting is wrong, not the tune.

Practical approach

• Know your hover throttle at both extremes. Measure it loaded and measure it empty. Those two numbers tell you your thrust margin at the worst case and your control authority at the best case, and they are the first thing anyone competent will ask you for.
• Tune light. Empty tank, minimum payload, the configuration with the least inertia. That is where the aircraft is most willing to oscillate, and it is where your gains must be safe.
• Verify loaded. Then, deliberately and cautiously, fly the loaded aircraft on that tune and confirm it is merely soft and not unacceptably soft. Log it. If it will not hold a line in wind, the answer is more thrust margin or a better airframe, not more gain.
• Never fly a loaded aircraft on a tune you have only validated empty — or the reverse. Both halves of that sentence matter. A tune is not validated until it has been flown in the condition you intend to fly it in.
• Change one thing at a time, exactly as you would on any other aircraft — one gain, one axis, the same manoeuvre every time. The general PID reasoning does not stop applying just because the mass is moving.

Finally, the question everyone asks: can the firmware just scale the gains with weight?

In principle, yes — the concept is sound, and it is a well-established idea in control engineering. If you know the aircraft's inertia, and you know how the gains ought to scale with it, you can schedule the gains against a measured proxy for mass. Some flight-control ecosystems expose mechanisms that get you part of the way there: switchable tuning profiles you can select in flight, or adaptive schemes that adjust the tune from measured response. What is available, what it is called and how well it works varies enormously between firmware families and between versions, so go and read your firmware's current documentation rather than trusting a number you read in an article.

What is safe to say is this: Betaflight is not designed around this problem at all. It is a fixed-gain acro controller for aircraft whose mass is essentially constant across a flight, and nothing in its architecture is trying to solve payload variation. If you are flying a machine whose weight halves during the mission, you are outside the envelope Betaflight was designed for, and you should be honest about that before you go looking for a setting that will rescue you.

Tune light. Verify loaded. Fix the CG mechanically. And put baffles in the tank.