Feedback Makes Autopilot Possible

An autopilot is useful because it keeps checking whether reality matches the target.

That repeated comparison is feedback.

Without feedback, a controller could issue a command based on an initial guess and never notice that wind, mass, battery voltage, sensor error, or another disturbance changed the result.

The control loop

A basic feedback loop can be drawn as:

target state
    ↓
compare target with measured/estimated state
    ↓
error
    ↓
controller
    ↓
actuator command
    ↓
physical aircraft
    ↓
sensor measurement / state estimate
    └────────────── back to comparison

The loop runs again and again.

That is what makes the system adaptive to disturbances.

Error is a difference, not a failure

In a simple one-dimensional model:

error = target - measured state

Suppose altitude target is 10.0 m and the estimated altitude is 9.4 m:

error = 10.0 - 9.4 = +0.6 m

A positive error tells the controller that the measured state is below the target under this sign convention.

The controller then decides what actuator change should reduce the error.

Error existing is normal. A feedback controller needs error information to know what correction is required.

The real question is what happens to the error over time.

Open loop versus closed loop

Imagine a fictional altitude system.

Open-loop approach

Apply 60% thrust for 4 seconds.

The command does not check actual altitude.

If the aircraft mass or battery condition changes, the same command may produce a different result.

Closed-loop approach

Target altitude = 10 m
Measure estimated altitude
Calculate error
Adjust thrust
Measure again
Repeat

Now the controller can react to the actual response.

That is more powerful—but it introduces new failure modes because the controller depends on measurement quality, timing, and tuning.

Proportional correction

A simple controller can make correction proportional to error:

command = Kp × error

Kp is a gain: a number that determines how strongly the controller reacts.

If error is 0.5 and Kp = 2:

command = 2 × 0.5 = 1.0

If error doubles, the proportional response doubles.

This is a simplified classroom model. Real flight controllers use more complete control structures, limits, filtering, feed-forward terms, multiple nested loops, and hardware-specific logic. The point here is the relationship: larger error can produce larger correction.

Too little and too much correction

A weak response can leave the system sluggish.

An overly aggressive response can cause overshoot or oscillation.

Imagine a target altitude of 10 m.

A fictional response looks like:

0 s:  8.0 m
1 s:  9.4 m
2 s: 10.6 m
3 s:  9.7 m
4 s: 10.3 m
5 s:  9.9 m
6 s: 10.1 m

The aircraft crosses the target several times. That is overshoot followed by oscillation that appears to decay.

The important question is not “did it reach 10 m?”

It did.

The important question is how the response behaves around the target.

Time response tells the story

Useful control-response terms include:

• rise time — how quickly the response moves toward the target;
• overshoot — how far it passes beyond the target;
• settling time — how long it takes to remain near the target;
• steady-state error — the remaining offset after the response settles;
• oscillation — repeated movement around the target;
• instability — behavior where error or oscillation grows instead of remaining bounded.

You do not need to calculate advanced control metrics yet. Learn to recognize the shapes.

Feedback inherits sensor problems

Suppose the real altitude is 10 m but a biased estimate says 9.5 m.

The controller sees:

target = 10.0
estimate = 9.5
error = +0.5

It may command more thrust because, from its point of view, the aircraft is low.

The controller can be mathematically correct and physically wrong because the measurement entering the loop is wrong.

Week 6 is now inside Week 8.

Delay changes the loop

A correction based on old information can arrive too late.

Imagine this sequence:

1. altitude falls;
2. sensor or filter reports the drop late;
3. controller commands more thrust;
4. aircraft has already begun recovering;
5. late correction pushes it above target;
6. another delayed correction sends it back down.

Latency can create or worsen oscillation even when each individual calculation looks reasonable.

Feedback systems are about timing as much as arithmetic.

Worked trace: position hold in a gust

A simulated aircraft is holding position when a wind gust pushes it east.

Trace the loop:

1. position or velocity estimate changes;
2. error relative to the hold target appears;
3. controller commands a corrective attitude;
4. aircraft tilts and redirects part of its thrust west;
5. westward force reduces the eastward motion;
6. sensors report the new state;
7. controller reduces or changes correction as error shrinks.

That is the physics from Week 7 running inside the feedback loop from Week 8.

Build a control-loop trace

Choose one controlled variable:

• altitude;
• roll angle;
• yaw heading;
• horizontal position in a simulator.

Create a trace with these fields:

Then identify one sensor error and one timing error that could make the loop behave badly.

Misconception: autopilot means the drone knows what to do

Autopilot executes control logic around targets and measurements.

It does not understand the mission like a person does. It follows configured goals, models, limits, and state estimates.

If the target is wrong, the system can accurately control toward the wrong target. If the estimate is wrong, it can correct a problem that does not exist.

The idea to keep

Feedback is a repeated argument between what we want and what the system thinks is happening.

The controller’s job is to turn that difference into a correction that reduces error without creating a worse response.