Hello everyone!
This is a quick-short question. Is there any way to parallelise the computation of the sigma-points in the UKF? This is, if there are 7 sigma-points to compute each one in a parallel thread.
Thank you so much in advance,
Antonio.
Hi!
That’s a good idea. I think we already tried this when implementing the filter and we did not see major improvements. But, that was several years ago and trying again is a good idea.
Could you open an issue in our Gitlab repo?
Thank you
Bryan
Hi,
I think that if the measurements are very close to each other in time, there won’t be a performance improvement, possibly the opposite. That could explain what you saw. You would need not-so-short propagations I guess.
Anyway isn’t parallelisation already used for the N ObservableSatellite?
Cheers,
Romain.
My memory is fuzzy, but that’s my recollection as well: we didn’t see huge performance gains running the sigma points in parallel.
At the moment each sigma point is propagated iteratively, and we reuse the propagation parameters from the same propagator builder. That would need to be approached carefully when parallelising.