Constraint-Based Networked Inertial Navigation

Abstract

Algorithms for a personal navigation system which uses a network of IMUs are developed and evaluated. To this end, a unified framework of a state constrained Kalman filter that utilizes prior knowledge of the states among individual nodes (or agents) of an inertial navigation sensor network was developed. By exploiting the kinematic and the geometric constraints among the states, the filter enables the network to be a self-contained navigation system. The concept of a measurement emulation model (MeM) was introduced to handle the challenges in a constrained filter (e.g., stochastic constraint, state argumentation, and cross-correlation assumption). Simulation studies are performed to provide insightful analysis on the effect of the measurement emulation model. Three filter architectures for fusing the information from the network are proposed: Centralized, decentralized, and hybrid. The hybrid algorithm was shown to be more robust in handling multiple constraints which exist in a network. The performance of these algorithms was compared through a personal, indoor inertial navigation experiment. Two alternative implementations of range constraint in inertial pedestrian navigation systems are presented. The range constraint between two feet are implemented as an asymmetrical quadratic constraint (i.e., an ellipsoid) and a multiple asymmetrical linear constraints (i.e., a box) instead of a symmetrical quadratic constraint (i.e., a sphere). The results show that the measurement emulation models can significantly improve the performance of the centralized filter while they have counter productive effects on the hybrid filter. Even though the hybrid filter is a combination of the centralized and the decentralized filters, the hybrid filter behaves similarly to the decentralized filter. The results demonstrated that a properly designed constrained filter can enable a self-contained personal inertial navigation system.