Hi,
I am working on fusing Visual Inertial Odometry (VIO) with the LPS system. This entails that I wish to be able to quantify how reliable the LPS is so that I can tune how the VIO and LPS affect the pose estimate.
For example, if the CF is outside the convex hull of the anchors, and the measurements are all over the place, I wish to give more weight to the visual odometry (to ensure a stable pose estimate). On the contrary, if the CF is inside the convex hull, I wish that the impact of the LPS is greater than the VIO.
My first idea is to calculate the trailing standard deviation of the error terms of the TDOA measurements. So, if the standard deviation is above a threshold, the weight of the LPS is decreased/totally disregarded.
Are there any other, maybe smarter ways to get a measurement of how reliable the LPS is?
Thanks!
Measure the reliability of the LPS

 Bitcraze
 Posts: 96
 Joined: Tue Jun 30, 2015 7:47 am
Re: Measure the reliability of the LPS
Hi!
I suppose another way of putting this is that you are looking for a measurement of how good the current estimated position is?
There are two possibilities that comes to mind:
1. Figure out if it is possible to get some sort measurement out of the kalman filter. I'm not an expert on EKFs and I don't know if it is possible.
2. Tracking the distance from LPS samples to the currently estimated position. If many samples are far away from the estimated position it might indicate that the estimate is bad. There is an implementation in the TDoA outlier detector (https://github.com/bitcraze/crazyflief ... erFilter.c) where we tried to use this idea to figure out if a TDoA sample is an outlier or not.
I suppose another way of putting this is that you are looking for a measurement of how good the current estimated position is?
There are two possibilities that comes to mind:
1. Figure out if it is possible to get some sort measurement out of the kalman filter. I'm not an expert on EKFs and I don't know if it is possible.
2. Tracking the distance from LPS samples to the currently estimated position. If many samples are far away from the estimated position it might indicate that the estimate is bad. There is an implementation in the TDoA outlier detector (https://github.com/bitcraze/crazyflief ... erFilter.c) where we tried to use this idea to figure out if a TDoA sample is an outlier or not.