Search

Air data systems require costly calibration of their static pressure sensors to characterize errors caused by the act of flying. Altitude-based methods for measuring these so-called static position...

This research proposes a novel, autonomous, regression-based methodology for Allan variance analysis of inertial measurement unit (IMU) sensors. Current methods for Allan variance analysis have been rooted in the human-based interpretation of linear trends, referred to as the slope method. The slope method is so prolific; it is referenced among electrical and electronics engineering standards for IMU error analysis. However, the graphical nature and visual-inspection–based use of the method limit its ability to be programmed as a generalized algorithm,which hinders the autonomy desired in modern-day navigation computations. Using nonlinear regression with a ridge-regression initial guess, the proposed method is shown to produce comparable results to the gold standard slope method when using standard-length data collections and outperforms the slope method when the amount of available data is limited. This development directly enables accurate navigation solutions for all vehicles in land, air, sea, and space operations.

In network localization scenarios, the geometry between the anchor nodes and the node of interest determines the minimum variance of the relative position error, commonly known as the Cramer-Rao Lower Bound. With modern computing resources, methods for real-time trajectory optimization are possible. The primary contribution of this paper is a distributed method for real-time self-localization trajectory optimization in a wireless sensor network of random formation aircraft. To perform this task, this paper introduces Self-Aligning Swarm, a distributed heuristic algorithm which generates a real-time trajectory for optimized self-localization based on local minimization of the trace of the Cramer-Rao Lower Bound. Using a 10K-trial Monte Carlo simulation, the proposed algorithm is shown to improve 2D position MSE of the optimized aircraft, reduce MSE variance, and demonstrate the MSE distribution approaches normality.

The task of inertial sensor calibration has required the development of various techniques to take into account the sources of measurement error coming from such devices. The calibration of the stochastic errors of these sensors has been the focus of increasing amount of research in which the method of reference has been the so-called “Allan variance (AV) slope method” which, in addition to not having appropriate statistical properties, requires a subjective input which makes it prone to mistakes. To overcome this, recent research has started proposing “automatic” approaches where the parameters of the probabilistic models underlying the error signals are estimated by matching functions of the AV or wavelet variance with their model-implied counterparts. However, given the increased use of such techniques, there has been no study or clear direction for practitioners on which approach is optimal for the purpose of sensor calibration. This article, for the first time, formally defines the class of estimators based on this technique and puts forward theoretical and applied results that, comparing with estimators in this class, suggest the use of the Generalized method of Wavelet moments (GMWM) as an optimal choice. In addition to analytical proofs, experiment-driven Monte Carlo simulations demonstrated the superior performance of this estimator. Further analysis of the error signal from a gyroscope was also provided to further motivate performing such analyses, as real-world observed error signals may show significant deviation from manufacturer-provided error models.

Multi-sensor navigation is quickly becoming operational as more alternative sensors continue to mature. Unfortunately, increasing the number of sensors also increases the possibility of corrupting the navigation solution with incorrect measurement models and undetected sensor failures. The multi-sensor resiliency chal- lenge remains largely unsolved, with only some sensor combinations (e.g., visual-inertial and GPS-inertial) having dedicated resiliency research. This research aims to provide multi-sensor resiliency through an au- tonomous sensor management framework. The proposed framework places each sensor into one of four modes: monitoring, validation, calibration, and remodeling. Each mode contains particular tasks that affect how the filter processes sensor measurements. The framework is developed by generalizing and interfacing functions found in existing research (fault detection, parameter estimation, and model selection) to achieve similar goals, but in a robust and sensor-agnostic manner. The proposed framework is compared against con- ventional filtering using two simulated scenarios including multiple sequential sensor failures and incorrectly modeled sensors.

Conventional methods for assessing navigation sensor performance focus on detecting sensor failures through the use of measurement residuals in a Kalman filter or Extended Kalman Filter. Additionally, even though there has been a considerable amount of research into sensor-specific online calibration algorithms aimed at estimating sensor model parameters, none of these methods consider the possibility a sensor model has been improperly specified prior to integration into the ongoing navigation solution (i.e., unmodeled biases or inaccurate measurement noise strength). Due to the growing interest in alternative, all-source navigation, there exists a need for a general, sensor-independent statistical framework for validating the specified sensor model prior to integration into the navigation solution. This research proposes a generalized sensor validation algorithm, in which sensor models are statistically validated using a novel online likelihood ratio test, while mitigating the risk of corrupting the ongoing solution due to corrupted measurements. This validation process is part of a larger three-phased framework aimed at managing alternative sensor model validation, integrating sensor-specific online calibration routines, and providing a means for online sensor remodeling instead of simply declaring a failure and rejecting its measurements. The proposed validation algorithm is developed in the context of the larger framework and its benefits are demonstrated and compared to conventional methods using simulated navigation data.

The availability of an accurate inertial navigation solution depends largely on the proper calibration of the deterministic errors usually found in inertial navigation systems, but more importantly, on the adequate stochastic modeling of the random errors associated with each inertial sensor. As such, a considerable amount of time and energy has been invested in the understanding and modeling of error sources found in a variety of inertial sensor types and their applications. However, a comprehensive review of academic, vendor, and research literature on the subject matter reveals that although the body of knowledge has large depth and breadth, the community lacks consistency across specialty fields with regards to terminology, unit standardization, and modeling techniques. pecialty fields with regards to terminology, unit standardization, and modeling techniques. This paper aims at providing the navigation community with a common framework for the modeling, simulation, and characterization of errors associated with inertial sensor measurements. The goal is not to replace the various inertial error model descriptions that exist, but to provide a common understanding of how these all relate together. After presenting the range of error models and terms found in literature, the subtle differences in terminology and categorization are compared, correlated, and consolidated into a common model. Next, the Allan Variance method is used to carefully and progressively illustrate how each error component in the common model arises and is identified. The effectiveness of the common model is demonstrated by analyzing stationary data from a real-world inertial measurement unit, comparing the identified error parameters to the vendor specifications, and using the same parameters to produce a simulated signal that closely matches the original. Armed with this common framework, the navigation community will be able to more effectively communicate and exchange ideas on the subject matter, thereby accelerating the progression towards cheaper, more reliable navigation technology with properly modeled inertial sensors at its core.

The Air Force Institute of Technology's Advanced Navigation Technology center has invested significant research time and effort into alternative precision navigation methods in an effort to counteract the increasing dependency on Global Positioning System for precision navigation. The use of visual sensors has since emerged as a valuable and feasible precision navigation alternative which, when coupled with inertial navigation sensors, can reduce navigation estimation errors by approximately two orders of magnitude [1] when compared to inertial-only solutions. A key component of many image-aided navigation algorithms is the requirement to detect and track salient features over many frames of an image sequence. However, feature matching accuracy is drastically reduced when the image sets differ in 3-D pose due to the affine distortions induced on feature descriptors [2]. In this research, this is counteracted by digitally simulating affine distortions on input images in order to calculate more accurate feature descriptors, which provide improved matching across large changes in viewpoint. These techniques are experimentally demonstrated in an outdoor environment with a consumer-grade inertial sensor and three imaging sensors, one of which is orthogonal to the others. False matches generated by using the orthogonal camera are shown to degrade the navigation solution if change in 3-D pose is not accounted for. Using a tactical-grade inertial sensor coupled with GPS position data as the truth source, the improved image-aided navigation algorithm, which accounts for changes in 3-D pose, is shown to reduce navigation errors by 24% in position, 16% in velocity, and 35% in attitude compared to the standard two-camera image-aided navigation setup.