Recursive state estimation for multiple switching models with unknown transition probabilities

This work considers hybrid systems with continuous-valued target states and discrete-valued regime variable. The changes (switches) of the regime variable are modeled by a finite state Markov chain with unknown and random transition probabilities following Dirichlet distributions. Our work analytically derives the marginal posterior distribution of the states and regime variables, the transition probabilities being integrated out. This leads to a variety of recursive hybrid state estimation schemes which are an appealing intuitive and straightforward extension of standard algorithms. Their performance is illustrated by a maneuvering target tracking example.