Page 1 of 1

Process, Observation and State Covariances

PostPosted: Wed Oct 02, 2013 4:15 pm
by freekopter
How did you calculate or measure the Noise values in the configuration (_n, _v and _q)?
I would like to use the SPKF with different sensors.

Re: Process, Observation and State Covariances

PostPosted: Wed Oct 02, 2013 5:14 pm
by bn999
They are estimated from ground simulation software which takes multiple flight logs and determines the values which tend to minimize MAE when compared to externally observed data.

Although different sensors could benefit from re-tuned values, the published values have been shown to work for multiple types. I've found that the GPS measurement noise and delay characteristics to be more important to the final solution than the sensor measurement and process noise values.

Re: Process, Observation and State Covariances

PostPosted: Wed Dec 10, 2014 9:40 am
by zhlearnaq
bn999 wrote:They are estimated from ground simulation software which takes multiple flight logs and determines the values which tend to minimize MAE when compared to externally observed data.

Although different sensors could benefit from re-tuned values, the published values have been shown to work for multiple types. I've found that the GPS measurement noise and delay characteristics to be more important to the final solution than the sensor measurement and process noise values.


hello, i am learning the aq code, and i have the same question about the Noise values, do the values mentioned here are the values such as UKF_VEL_Q、UKF_VEL_V、UKF_ACC_N and so on? And do you mean those values need to be adjust or they can be got from ground simulation software?

Re: Process, Observation and State Covariances

PostPosted: Wed Dec 10, 2014 9:06 pm
by bn999
Yes - Q values are initial variance estimates, V values are process noise and N values are measurement noise. They comprise all of the tunables for the UKF.

Re: Process, Observation and State Covariances

PostPosted: Thu Dec 11, 2014 10:20 am
by zhlearnaq
Thank you very much for your reply, but i still can't find how to get those noises by the ground simulation software, i have read the IMU calibration files for many times. Could you please to point out where to get the values in detail?

Re: Process, Observation and State Covariances

PostPosted: Thu Dec 11, 2014 4:27 pm
by bn999
The values have been pre-determined though a process which has not been published.

If you are trying to apply AQ's UKF to a system with similar sensors, I believe it should continue to perform very well with the existing parameters. If you need to change the structure and add different types (vision, Doppler, etc.) of sensors or states, you will probably need to start tuning from scratch.

Re: Process, Observation and State Covariances

PostPosted: Fri Dec 12, 2014 3:52 am
by zhlearnaq
bn999,
Again, thank you for your reply, as a model aircraft enthusiast, i am deeply attracted by autoquad not only because it is open source but also it's good performance, what's more, there are so many skilled people like you who are so kind to give suggestions. I would like to learn more about it.
Now I have got another problem, i have bought a STM32f3 DISCOVERY board(with sensors LSM303 and L3GD20) and added a MS5611 to it, with the existing parameters, i can't get a good attitude. When i refer to the github, i noticed that benzeng published a version of AutoQuad for PX4/Pixhawk(the sensors used in Pixhawk are similar to mine), however the parameters he offered are still the existing parameters without any change, i think that there should be many differences between the two sets of parameters, could you offer me the set of parameters for Pixhawk if you have? my Gmail is zby.roc@gmail.com.
If you don't have the parameters right now , would you please to show me the name of the algorithm used in calculating the parameters?

Re: Process, Observation and State Covariances

PostPosted: Fri Dec 12, 2014 4:08 pm
by bn999
I was not aware of benzeng's work, nor do I have any different parameters to offer.

Re: Process, Observation and State Covariances

PostPosted: Sat Dec 13, 2014 8:01 pm
by Kisssys

Re: Process, Observation and State Covariances

PostPosted: Mon Dec 15, 2014 3:14 am
by zhlearnaq
Thank you for bn999's reply all the same, and really appreciate Kisssys for the web sites offered. I have read those codes and In this codes, benzeng added some new codes on reading data from the added sensors, but he shielded the runTaskCode which i thought is the most usefully code, i was totally confused, have you noticed this change? Did i misunderstand? Actually, I am using the same sensors, but i can't get a better attitude, there are many bugs in my attitude, I thought the bugs came from the inaccurate noise parameters. I have adjust the parameters from feelings for a long time but no positive development, i must have something wrong in my way and i don't know what to do now, could you show me some suggestions? please.