Originally Posted by

**phil0stine** Hi all, my first post, I will try to be brief but the problem is a little convoluted.

I am designing a multi-sensor kalman filter that can 1) dynamically adapt the error covariance matrices, and 2) Deal with latent data

To deal with the latency problem, I would like to include prior sensor data (say, over the last 10 time steps) in my state vector. Is this legal?

I am concerned because, lets say I have predicted state k-5, and at the current time step I receive an actual measurement of the state k-5. Then what?

I can't go to a correction step, because I let's also say I don't have a measurement of state k-2 (just a prediction). Then the state vector can't ever be corrected? Or am I totally off base?

Finally, the more obvious choice would be to use a conventional state vector, and keep track of older measurements in a buffer of sorts. For old data, I would backtrack and re-calculate each state until the current one. The problem with that is that my state covariance matrix P has changed too, and it seems extremely inefficient to store all this data for the last 10 time steps.

So basically I am stuck, and I really appreciate any help whatsoever.

Cheers

Phil