That 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 is

That 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!