ABSTRACT
This chapter discusses some practical aspects of Kaiman filter design and implementation. It analyzes the effects of assuming a model that does not accurately describe the physical system whose state is to be estimated. The chapter presents some techniques for designing filters of simplified structure. The Kaiman filter is designed for the deal situation when all plant dynamics and noise statistics are exactly known. The Kaiman filter has no mechanism built into its theory for dealing with inaccurate dynamics or statistics. The chapter also discusses two methods for preventing divergence: fictitious process noise injection and exponential data weighting. The dominant factor in determining the computational complexity of the Kaiman filter is the dimension of the model state vector. The chapter explains the elimination of a portion of the state vector by substitution and ignoring of a portion of the state vector.
