Identificação e Modelagem de Erros Estocásticos em Sensores Inerciais via Variância de Allan e Otimização

Abstract

With the growing opportunities in the most diverse areas of engineering associated with autonomous vehicles, much has been studied about the application of low-cost inertial sensors, especially of the Micro-Electro-Mechanical Systems (MEMS) type. However, the recognized predominance of stochastic errors in such sensors becomes an obstacle, since most of the documentation provided by the respective manufacturers is incomplete. In this sense, this work aims to investigate the Allan Variance (AV) technique as a tool for identifying the nature of Inertial Measurement Unit (IMU) stochastic errors. In addition, a discrete-time state-space model capable of reconstructing these errors from the stochastic characteristics identified for the inertial sensors is evaluated. Finally, an optimized version of the aforementioned technique is proposed, which is capable of refining the dynamic model of stochastic errors, as well as its correspondence with the AVs experimentally obtained for the inertial sensors. Experimental results, based on data collected from different commercially available low-cost IMUs, show that the models and techniques discussed are capable of adequately estimating the behavior of such errors and, therefore, guaranteeing the adequate tuning of eventual estimators (such as the Kalman filter, for example), when merging IMUs with auxiliary sensors for integrated multi- sensory navigation.