What is an a priori covariance matrix and how does it affect filter convergence?

• A. The final uncertainty after processing all data.
• B. The initial uncertainty of the state; influences how quickly and reliably the filter converges.
• C. The noise of the measurement device only.
• D. The dynamic model of the system.

Answer: (B)

An a priori covariance matrix represents how uncertain you are about the current state before incorporating the latest measurement. It sets the balance between trusting your model prediction and trusting the new data, which in turn controls how quickly and reliably the filter converges to the true state.

In a Kalman-like filter, the Kalman gain depends on this prior uncertainty. If the initial uncertainty is large, the filter gives more weight to new measurements early on, so a wrong initial guess is corrected faster. If the initial uncertainty is set too small, the filter trusts the model too much and may be slow to adapt or resist corrections when the initial state is off. As more data comes in and process noise is accounted for, the influence of the initial guess fades, but that starting P0 largely shapes the early convergence behavior and reliability.

It's not the final uncertainty after processing all data, nor is it the measurement noise alone or the dynamic model itself. Those aspects play different roles in the filter, while the a priori covariance specifically captures initial state uncertainty and its effect on convergence.