Blogs

Hi everyone,

I'm working on the BreezeUav project which is open-source autopilot for fixed-wing and multicopter. So far it's possible to fly by using the board we designed based on Atmega 2560 and we currently adapt the code to be executed on ARM Cortex soon.

Anyway today I wanted to share with you some outputs I find by running the Kalman filter with differents sets of parameters and by comparing it to the complementary filter. Since the test is made by comparing the roll angle by moving a tricopter, the motors are turned off, therefore the vibrations don't interfere with the results.

I first set high value of parameters, In kalman filter, two parameters are very important : Q and R. Q defines the relative trust to the accelerometer against the gyro, and R define the maximum jitter of the accelerometer. In the graph below, Q = 0.15 and R = 0.25

As you can see, the kalman filter has some delay and the value doesn't reach the maximum.

Therefore I decided to decrease the parameters to Q = 0.001 and R = 0.01 :

In this example, the kalman follow quite well the complementary filter. During fast maneuver, the maximum value is not reach but there is no delay anymore.

Therefore in my next flight I'm gonna try to use both complementary filter and kalman filter to have roll/pitch angles and see how stable the tricopter flies.

If you are interested I have produce similar output about vibration and weight put on the the top of the IMU to see how weight succeed to decrease the vibrations a lot.

Here is the first on board video of the breeze project :

https://www.youtube.com/watch?v=8k2iU6S42xc


Have a good flight,
Adrien