Revision history [back]

EDIT: The measurement model in a Kalman filter is meant to project the state into measurement space. Many sensors don't directly measure the quantity you want (e.g., altitude), but instead measure some other quantity (e.g., pressure) from which you can derive what you want. The measurement model is there to "convert" the current vehicle state into a measurement so that you can directly compute the innovation (error) between the two.

For the EKF in r_l, all of the measurements we receive are already in the same space as the state, i.e., we are directly measuring vehicle velocity, rather than some other quantity. So for us, the H matrix is just the identity matrix. However, we can use it for one other purpose, which is to limit which variables are affected in the state when we perform the Kalman gain computation and apply it. This is useful for one of the filter's key features: updating subsets of the state vector. Otherwise, every single measurement would have to have every variable in the state vector. Instead, we can take in different measurements from sensors that only measure some subset of the state variables.