Networked Inertial Navigation with Constraints Generated by Neural Networks

This paper deals with the subject of networked inertial navigation. The term networked inertial navigation refers to the PNT (Positioning, Navigation, Timing) technique whereby the measurements from a network of individual inertial measurement units (IMU) are fused to generate a navigation solution while mitigating the drift error inherent in the process of integrating accelerations and angular rates. Kinematic constraints between the IMUs in the network are exploited to help reduce the overall drift error of the PNT solution. However, most kinematic constraints in practice are difficult to derive precisely or are not known a priori. This limits the utility of the networked inertial navigation approach. In most prior work in the literature, a common and conservative way of dealing with kinematic constraints is to formulate them as inequality constraints (such as an upper bound constraint) to account for the poor knowledge of the constraints. A projection approach is then often used to solve the inequality constraint problems. However, it is shown in this paper that the projection approach comes with several shortcomings in dealing with constraints. This paper deals with one key challenge encountered in the networked inertial navigation problem: unknown or poorly known constraints. In this paper, a deep-learning-based approach is proposed to learn the kinematic constraints within the networked inertial system from training data sets. This method fills the void of the poorly known or unknown stochastic constraints and allows us to obtain those constraints with a given uncertainty from data. This approach is tested in a human pose estimation problem and experimental results show that it is able to effectively mitigate the error drifts and accurately capture the relative pose between IMU components within in the networked inertial system.