Page 1 of 1

Loco accuracy

Posted: Thu Jan 09, 2020 12:14 pm
by chopin
Hello.
I'm checking what is the highest accuracy I can achieve with 8 anchors in TWR mode.
At the moment the results aren't promising at all, the instructions said to expect up to 10cm of inaccuracy. But my results are worse.
The anchor positions are:
0 - (0, 3.6, 0)
1 - (0, 5.4, 1.52)
2 - (1.8, 8.4, 0.6)
3 - (6, 8.4, 1.52)
4 - (6.6, 4.8, 0.6)
5 - (6.6, 2.4, 1.52)
6 - (4.8, 0, 0.6)
7 - (1.8. 0.6, 1.52)
The points have a distance of at least 2 meters from each other, and yet this is the result I have achieved.

Image
https://imgur.com/a/biOF7Pn

Those are 2 histograms are the errors. one of x and one of y.
I also seem to have an offset in the y-axis although when I stand in the same place the coordinated are almost 0,0,0 but when moving the offset appears.
I have another system that gives very accurate results up to 0.1 mm. After subtracting the Loco from my system I have achieved these histograms.
I searching for reasons why this could happen.

Moreover is there any way to increase the accuracy of the system by tweaking the settings?

Thanks in advance.

Re: Loco accuracy

Posted: Sat Jan 11, 2020 9:26 am
by jdiep
Hi,

I did a project on this topic:


Maybe this helps you out!

Re: Loco accuracy

Posted: Sun Jan 12, 2020 6:45 am
by chopin
Thanks a lot!!!
It is very helpful and well explained, I am very grateful for you sharing this.

Re: Loco accuracy

Posted: Sun Jan 12, 2020 7:50 am
by chopin
Hello!
Does anyone else get similar results? Do you have any experience in changing helpful parameters in the Kalman filter?
https://imgur.com/a/nV64Hd6
some results I've got
The red is the ground truth I got from a mocap and the blue is the Roadrunner.
the first row is the x, second is y and the third is the XY. to the right of each graph, you can see the error

Re: Loco accuracy

Posted: Mon Jan 13, 2020 2:30 am
by jdiep
I never worked with the Roadrunner but to me, the error in the localization seems pretty normal and is not the fault of a Kalman Filter or any of its parameters per se. The problem with these UWB antennas is, that the channel impulse response is dependent on the angle of arrival and departure of a signal (in case of rotation of the robot-attached tag) as well as the close and far environment (resulting in multipath signals). Consequently, this results in erroneous range measurements, which can not be easily calibrated. As long as you are not able to model the channel impulse response (which is nearly impossible) and incorporate this into the measurement equation of the Kalman Filter, these localization error occurs.

However, in my testing, I figured, that my drone flies much smoother (https://www.youtube.com/watch?v=BqnEMfDYWC4) with an uncorrected measurement model if both antenna modules are attached vertically with nothing blocking the line-of-sight (I see there is a cover for the Roadrunner module).