Simulação e análise de algoritmos para estimação de atitude utilizando sensores de inerciais

Abstract

The present work analyzes algorithms for attitude determination from simulated measu- res of accelerometer, magnetometer and gyroscope. These sensors were modeled in the

Simulink software, considering white noise and random walk additive noise, and hard-iron and soft-iron interferences for the magnetometer. Aiming at the future implementation of the

presented algorithms, the sensor models were developed in order to reproduce the beha- vior of the MPU-9250 Inertial Measurement Unit (IMU). Attitude measurement algorithms

are presented in the context of the Wahba Problem solution, and the covariance analysis of the estimation error for two measurements using the Quest Measurement Model (QMM) is detailed. In order to integrate the measurements of all IMU sensors and the development of a robust algorithm, some attitude estimation methods based on the Extended Kalman

Filter (FEK) are presented. The following topologies are discussed for the FEK measure- ment equation: non-normalized measure of the attitude sensors; normalized measurement

of attitude sensors using QMM; quaternion preprocessed by some static algorithm. It is

discussed how the normalization of the quaternion takes the covariance matrix to the sin- gularity, and the FEK with multiplicative error is presented to solve this problem. The results

of this work range from the modeling and calibration of the inertial sensors, to the results

of simulations with the FEK. The IMU MPU-9250 sensor modeling results include the appli- cation of the Allan Variance method to characterize the additive noises, and practical and

simulated results on the magnetometer calibration and alignment process. Regarding the estimation algorithms, simulation results are presented evaluating the accuracy of the bias

and attitude quaternion estimation, as well as the attitude convergence during the initializa- tion step.