Kalman filter with mixed frequency

AI Thread Summary
The discussion focuses on using a Kalman filter to estimate a robot's position at 40 Hz, utilizing mixed frequency data from GPS, accelerometer, and steering wheel sensors. Two methodologies are proposed: applying the Kalman filter separately to each data series or creating a composite state vector for the filter. The second approach is favored, as it allows for handling missing data effectively and leveraging the GPS data as a benchmark. Recommendations include setting up the filter for the highest sensor frequency and adjusting the measurement matrix when GPS data is unavailable. Overall, the Kalman filter is deemed robust for this application, and additional resources for understanding its implementation are suggested.
Plouffy
Messages
1
Reaction score
0
Hi first post hoping for some advise or guidance.

Let's say I have GPS data on a robot every 10 Hz (every 0.1 second), I have accelerometer data every 40 Hz and steering wheel data every 80 Hz. I would like to use Kalman filter to produce estimates of where the robot is every 40 Hz given the previous data (e.g. interpolate the GPS data using Kalman). My question is what methodology should I use (if anyone knows a good paper on the subject I would be very grateful). My current thoughts are to either:

1. Apply a Kalman filter to each data series separately to interpolate to the highest frequency (i.e. only have lagged values of Y as its state variable) and then again with the entire datasets in our filter.

2. Have my Y (left hand side) be a vector of size [8*1] containing the GPS where the column vector is 0 everywhere but at 10Hz, and then apply Kalman filter (accelerometer vector would be treated the same but with two observations and NaN everywhere else).

Thanks a lot.
 
Option 2 sounds like the appropriate one. The KF is robust wrt missing data. Set up your design for 80 Hz and read at any frequency you desire. I assume that the GPS data will benchmark the other data, so the filter will operate well. A good reference on the KF can be found at http://www.stat.pitt.edu/stoffer/tsa3/. Now, instead of "stealing" the PDF, you can download an earlier version of the book.
 
I perform my KF at the highest rate that my sensor offers (100hz). When your GPS is not producing any data you can reflect that by modifying your H matrix accordingly. Note that IMU data shouldn't be dropped but should be averaged if you're going for a slower rate.
 
Hi all, i have some questions about the tesla turbine: is a tesla turbine more efficient than a steam engine or a stirling engine ? about the discs of the tesla turbine warping because of the high speed rotations; does running the engine on a lower speed solve that or will the discs warp anyway after time ? what is the difference in efficiency between the tesla turbine running at high speed and running it at a lower speed ( as fast as possible but low enough to not warp de discs) and: i...
Thread 'Where is my curb stop?'
My water meter is submerged under water for about 95% of the year. Today I took a photograph of the inside of my water meter box because today is one of the rare days that my water meter is not submerged in water. Here is the photograph that I took of my water meter with the cover on: Here is a photograph I took of my water meter with the cover off: I edited the photograph to draw a red circle around a knob on my water meter. Is that knob that I drew a red circle around my meter...
Back
Top