I have built a quadcopter from scratch, including my own flight controller.

I have implemented a sensor fusion algorithm (Madgwick algorithm), which returns me current Yaw, Pitch and Roll angles. Then, a PID control algorithm based on these measured angles and desired angles (which are set by the control sticks on a transmitter) adjusts the speed of the motors.

For pitch and roll controls the algorithm is obvious and works just fine (i.e, pitch stick is 100% forward - desired pitch is "30 degrees", and current measured pitch is trying to catch up with that value using PID control). However for Yaw I would like the yaw stick to rotate the quad in the corresponding direction as long as I am holding the "yaw stick" in a non-zero position, with a rotating speed, proportional to the yaw stick angle. Since I have current measured Yaw angle (relative to a "reference yaw" which is the yaw that the quad had when the throttle was zero, not relative to the "absolute magnetic North-origined yaw"), I am basically just adding a "quadRotationSpeed = yawStickValue*delta_t" to the "desired yaw angle" in each iteration of the control loop, and then do the rest as a regular PID control: based on desiredYaw and measuredYaw I am calculating the motor torques.

Now the questions:

1) Is this approach wrong?

2) When I turn the yaw stick, the quad seems to be rotating with the speed, proportional to the stick angle, but when I let go off the stick, it rapidly turns back half way and then keeps turning slowly back to the "desired angle" I set with the yaw stick.

3) After this yaw maneuver the quad seems to be much less stable during the flight, but gradually it stabilizes.

My quad is about 800 grams and is quite large, about 60 cm in diagonal.


1 Answer 1


So I reimplemented the whole Yaw control thing using PID on raw gyro data, ignoring the sensor fusion current Yaw angle, which turned out to work perfectly. Since gyro returns the speed of change of the yaw angle, and I wanted to control the speed of change of the angle using control sticks, doing so directly is the most natural approach.

  • $\begingroup$ You've found the solution. So maybe you can accept the answer and close it? Though there are good topics for discussion: like, your first method and it's funny behavior was probably related to the integral term in the control loop. Using yaw rate itself was the way to go. (Maybe adjusting the gains with pitch and roll angles, if you'll do an agile quad). $\endgroup$ Commented Oct 19, 2017 at 18:47

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.