|Subject:||[Paparazzi-devel] sky uniformity and roll/pitch bias|
|Date:||Thu, 2 Oct 2008 17:35:25 +0200|
Greetings,I've been using paparazzi for some time now for some fairly precise navigation tasks (photo missions) with quite good results. I think I have my aircraft set up well and the control loops well tuned.
Under some conditions though, the ability to hold course and altitude can become poor, even to the point of loss of control. My conclusion is that the distribution of clouds and clear patches in the sky can set up a thermal bias or "tilted horizon" which gives biased angle estimates. The effect manifests itself by the plane overshooting turns on one heading and undershooting them in the opposite heading, and when following lines it may never quite reach the desired course, but fly along parallel to the line.
This is a bit different than when you have a roll bias due to misalignment between the sensor axis and the aircraft axis: here it will fly a constant offset inside a circle in one direction and outside the circle in the other direction, but there are no course dependent effects. It will also tend to turn left or right in AUTO1.
It seems that vertical contrast alone is not such a problem, as long as the sky is fairly uniform- it should only affect the resolution you can measure the angles. But the worst case seems to be a mostly clear sky with low dark (warm) clouds approaching from one direction.
I searched the list and haven't found much discussion about this phenomenon, except from some people who thought that a low sun could bias the sensors (i haven't personally found this to be much of a problem). So I was wondering if anyone else could share their experiences, solutions, control system changes, or tuning strategies to minimize the effect. I guess I would like to expand the envelope of weather conditions that the plane can navigate effectively.
My setup is an electric flying wing with the a stock FMA copilot horizontal sensor mounted on top of the fuselage and a vertical sensor mounted near the back. The horizontal sensor is oriented at 45 degrees. I run the sensors at 5v to maintain the stock gains but the output is run through a voltage divider before being read by the autopilots ADC.
|[Prev in Thread]||Current Thread||[Next in Thread]|