3 ms·

Kalman filters are great when there is limited compute available, but I’m a bigger fan of the newer and more advanced models such as “particle filters”. The ad

Kalman filters are great when there is limited compute available, but I’m a bigger fan of the newer and more advanced models such as “particle filters”. The advantage of particle filters is that they can handle complex scenarios with nonlinear physics and non-Gaussian distributions. For example, a vehicle GPS unit could use street maps to eliminate impossible locations based on the recent history of turns. Gaussian filters can’t do that and just result in blurry blobs that cover several blocks.

Scipio_Afri 3y ago
Do you have any recommended resources for learning about particle filters? Also l, what other types of newer models or more advanced models other than the particle filter would you recommend for offline processing?
- jiggawatts 3y ago
  Unfortunately I don't have anything at hand better than what Google would provide.
- esafak 3y ago
  Look up the positioning and tracking literature; e.g., Probabilistic Robotics by Thrun et al. The method is also called "sequential Monte Carlo". I found this article useful in learning the method: Bayesian Filtering for Location Estimation (https://rse-lab.cs.washington.edu/postscripts/bayes-filter-pervasive-03.pdf https://rse-lab.cs.washington.edu/postscripts/bayes-filter-p...) It might help to first review importance sampling. https://en.wikipedia.org/wiki/Importance_sampling https://en.wikipedia.org/wiki/Importance_sampling A comprehensive survey (for 2003) is Zhe Chen's "Bayesian Filtering: From Kalman Filters to Particle Filters, and Beyond". A contemporary survey would surely incorporate machine learning. https://people.bordeaux.inria.fr/pierre.delmoral/chen_bayesian.pdf https://people.bordeaux.inria.fr/pierre.delmoral/chen_bayesi... For extensions see "An Overview of Existing Methods and Recent Advances in Sequential Monte Carlo" https://people.eecs.berkeley.edu/~jordan/sail/readings/cappe-godsill-moulines.pdf https://people.eecs.berkeley.edu/~jordan/sail/readings/cappe...

Scipio_Afri 3y ago

Do you have any recommended resources for learning about particle filters? Also l, what other types of newer models or more advanced models other than the particle filter would you recommend for offline processing?

jiggawatts 3y ago
Unfortunately I don't have anything at hand better than what Google would provide.
esafak 3y ago
Look up the positioning and tracking literature; e.g., Probabilistic Robotics by Thrun et al. The method is also called "sequential Monte Carlo". I found this article useful in learning the method: Bayesian Filtering for Location Estimation (https://rse-lab.cs.washington.edu/postscripts/bayes-filter-pervasive-03.pdf https://rse-lab.cs.washington.edu/postscripts/bayes-filter-p...) It might help to first review importance sampling. https://en.wikipedia.org/wiki/Importance_sampling https://en.wikipedia.org/wiki/Importance_sampling A comprehensive survey (for 2003) is Zhe Chen's "Bayesian Filtering: From Kalman Filters to Particle Filters, and Beyond". A contemporary survey would surely incorporate machine learning. https://people.bordeaux.inria.fr/pierre.delmoral/chen_bayesian.pdf https://people.bordeaux.inria.fr/pierre.delmoral/chen_bayesi... For extensions see "An Overview of Existing Methods and Recent Advances in Sequential Monte Carlo" https://people.eecs.berkeley.edu/~jordan/sail/readings/cappe-godsill-moulines.pdf https://people.eecs.berkeley.edu/~jordan/sail/readings/cappe...