5 ms·
Generic warning for anyone implementing the Kalman filter. Read the first few pages of https://www.stat.berkeley.edu/~brill/Stat248/kalmanfiltering.pdf https://
by bachmeier 3y ago
Generic warning for anyone implementing the Kalman filter. Read the first few pages of https://www.stat.berkeley.edu/~brill/Stat248/kalmanfiltering.pdf https://www.stat.berkeley.edu/~brill/Stat248/kalmanfiltering... for ways to handle numerical instability.
- davidw 3y agoDavid's random thoughts: for some reason I pictured Wile-e-coyote not handling numerical instability correctly and getting, as per usual, blown up.
- tnecniv 3y agoThat paper contains an interesting reference to an exact generalization of the KF to a flat initial condition. I hadn’t seen that before and solves numerical issues that come from the common practice of just picking a large initial covariance. In practice, I haven’t seen any notable numerical artifacts from doing that, but it’s quite an appealing fix!