|
From: | Marko Thaler |
Subject: | Re: [Paparazzi-devel] Booz attitude estimation |
Date: | Mon, 7 Jun 2010 07:49:27 +0200 |
Hi Felix, Thank you for your answer and clarification. I believe we agree on everything! ;-) In the past I worked on a hexacopter project that used different fusion algorithms for attitude estimation based on the current flight mode. During hard maneuvers it used only gyro data but during hover or constant speed flight it used gyro and GPS fusion. With this approach the attitude estimation/ hovering performance was improved at the cost of filter complexity and CPU usage. The simple gyro + accelerometer fusion algorithm had a longer response time and more hover drift. I haven't looked at the Booz GPS code but I presume it uses it to counteract the hover drift problem. Anyway, when I get my Booz flying I will try out different approaches and see if I can make some useful contribution to this wonderful project. Cheers, Marko Date: June 6, 2010 11:56:04 PM GMT+02:00 To: address@hidden Subject: Re: [Paparazzi-devel] Booz attitude estimation Reply-To: address@hidden Hi Marko, the assumption is that the quad is not accelerating (which is also true if you fly at constant speed in one direction). In this case you can estimate the roll and pitch angles from the accelerometer measurements, meaning the body frame sensor data ax and ay is relevant for the attitude estimation. Of course it is correct that during manoeuvres when the quad is accelerating the estimate from the accelerometers will not be correct. But on the long run they provide useful information as cannot accelerate indefinitely in one direction and usually don't fly in tight circles all the time. This is why the simple filter "fuses" (high passed) gyro data for short term and (low passed) accelerometer data for long term estimation. Hope this helps to clear it up a little. Cheers, Felix |
[Prev in Thread] | Current Thread | [Next in Thread] |