B8.1 - Motion State Prediction by Unaided Inertial Micro-Machined Accelerometers

Abstract

Abstract: Estimating motion velocity state is crucial in dealing with inertial navigation systems (INS) or in situations where backing of an INS by other technologies is difficult or impossible. This study therefore investigates time sequences delivered by embedded inertial sensors in order to draw conclusions about the motion state of moving objects. Various probability tests were evaluated by a simple but typical measurement setup to assess robustness against random walk fluctuations and behavior in the constant velocity state, in order to detect transition from standstill to motion and vice versa. Our investigations end with a proposal for advanced motion state estimation algorithms, where different statistical approaches have been combined. Key words: accelerometer, indoor localization, inertial navigation system, random walk, zero-velocity update. 1 Introduction The satellite based global positioning system (GPS) has certainly revolutionized our everyday lives. However, electromagnetic waves are strongly shielded by the outer shell and walls of buildings. Even if the electromagnetic waves can penetrate into a building, reflection and diffraction effects makes localization and positioning based on electromagnetic wave propagation nearly impossible. Hence, there is still a strong demand for indoor navigation systems, e.g. inertial navigation systems (INS). Due to enormous development steps in performance, size and price during the last decade many INS today are based on microelectromechanical systems (MEMS). The output of a MEMS sensor typically represents the acceleration or the angular rate of the moving object. Consequently, if one is interested in the position of the moving object, the sensor signal has generally to be integrated twice with respect to time [1]. One of the most challenging tasks in the development of MEMS based INS must surely be to cope with different types of deterministic and stochastic error sources [2][3], e.g. random offset variations, also known as Brownian motion or random walk, infiltrating MEMS at the proof mass by random drifts of molecules [4][5]. Regardless of how random signals are modeled [6][7][8], mastering random fluctuations is still a very challenging task in designing advanced inertial positioning/tracking solutions, even if only the motion state is of interest. Motion state estimation can essentially be subdivided into the seemingly simple tasks of reliably detecting (i) standstill, (ii) transition from zero velocity to moving state, (iii) the unfailing detection of ongoing motion, and (iv) transition from moving to standstill. This article will describe various approaches with the aim of determining the motion state of an object equipped with strapdown INS. Insight into the demands of unaided inertial positioningis delivered in Section 2. In Section 3 various probability tests are discussed. Measurements to characterize their performance are presented in Section 4. Combining probability tests improves estimation results substantially. This will be demonstrated in Section 5. Concluding remarks are given in the final Section 6.