The Simultaneous Localization and Mapping problem in a nonlinear parameter identifiability perspective

A feasible solution to the Simultaneous Localization and Mapping (SLAM) problem is often considered invaluable in many autonomous vehicle navigation applications. In this paper we show that SLAM initialized with a known vehicle pose can be considered as a problem of parameter identification. Using a rank test for nonlinear map state identification, we establish that all the map states in the SLAM problem are identifiable given the initial conditions of the vehicle pose with zero uncertainty. We compare and contrast the nonlinear observability of the SLAM problem with its nonlinear parameter identification characteristics. We also show using the linear single dimensional SLAM problem formulation and closed form solutions that, given the initial conditions of the vehicle state with no uncertainty, the map states are identifiable. Furthermore, we show that the map states are not identifiable if the initial conditions are not known. We also present simulation and experimental results to corroborate the theoretical results. We propose that it is possible to use machine learning techniques in SLAM in view of the new parameter identifiability perspective.