Reconfigurable Integration Filter Engine (RIFE) for Plug-and-Play Navigation

A. Soloviev, C. Yang

Abstract:	It is widely recognized that GNSS availability and quality is decreasing rapidly as operational arenas shift from open sky to degraded signal environments where multi-sensor augmentations of GNSSS are sought to maintain required positioning, navigation and timing (PNT) capabilities. Current multi-sensor implementations are rather ad hoc and sensors-specific. Short-term gains in implementation efficiency are soon offset by non-recurring engineering costs of initial development and long-term higher integration costs whenever changes or upgrades are required. This paper will introduce a concept of a universal plug and play navigation (UPNAV) solution. The solution reconfigures itself on-the-fly in a truly plug-and-play manner as the sensors are connected (disconnected) to the system, without the need to redesign the system architecture or its specific components. Plug and play sensor fusion is supported by the reconfigurable integration filter (RIFE). The filter mechanization is abstracted as object-oriented multi-sensor estimation. Various sensors are represented by generic classes in the RIFE library. Each class is designed for a generic type of sensor (rather than for a specific sensor) wherein sensor types are defined by the type of their measurements. When a sensor is connected to the system, the RIFE is reconfigured by identifying the sensor’s measurement type and instantiating a sensor object using a corresponding class from the RIFE class library without a need of redesign or any coding anew. When the sensor is disconnected from the system its object is removed to free memory and computation resources. RIFE utilizes a self-contained inertial navigator as its core sensor. The INS does not rely on any type of external information and as a result can operate in any environment. However, inertial navigation solution drifts over time. To mitigate INS drift, this core sensor is augmented by reference navigation data sources (such as, for example, GPS or a laser scanner). Reference data sources generally rely on external observations or signals that may or may not be available. Therefore, these sources are treated as secondary sensors that provide aiding measurements in order to reduce drift in inertial navigation outputs. The estimation of INS drift terms is performed using the concept of complementary filtering. The idea is that aiding measurement can be generally represented as a function of position, velocity and attitude. This function is computed based on INS navigation outputs and then compared to the actual measurement. A discrepancy between the INS-based prediction and the actual measurement is used by the complementary filter mechanization to estimate INS drift terms. It is difficult, if not impossible, to create an exhaustive list of all possible aiding sensors. Yet, it is possible to categorize aiding sensors into generalized types. The RIFE design is abstracted for generic aiding sensors that are grouped into classes according to the type of their measurements. The filter adapts to the available types and number of measurements by resizing error states and system matrices without any coding and compiling anew. Classes are designed as generic templates that represent each type of measurements. A template includes generic variables (sensor states such as clock bias and drift or INS error states) and generic computational routines (pre-processing and computation of filter measurement observables). Once a particular sensor is connected, the system recognizes its type and instantiates a corresponding sensor object using a generic template and specifics of the sensor. For instance, the integration filter is reconfigured to add GPS receiver clock states and include range and carrier phase measurement observables, if the GPS receiver is connected to the system. In other words, RIFE is reconfigured by instantiating or removing sensor objects using their templates from the library of classes. As a result, the filter adapts to the available types and number of measurements by resizing error states and system matrices without any coding and compiling anew. The software design is also upgradable in a sense that if a totally new type of sensors (that is not covered by any of current classes) becomes available in the future, a new template can be designed and then added to the overall software without the need to modify other components of the system. The paper will describe formulation of aiding observables for example generic sensors such as relative and absolute position measurements; velocity measurements; attitude measurements; and, ranging measurements. The use of generic aiding measurements for incorporation of specific sensors into the plug-and-play navigation architecture will be illustrated. For instance, relative position measurements are abstracted as a projection of position change vector onto axis or axes of navigation or body frames of reference. As a result, the same generic measurement observable can be applied to process temporal changes in GPS carrier phase; odometer-based position increments; and, changes in ranges to line features that are extracted from images of a scanning lidar. The paper will demonstrate plug and play navigation capabilities using simulated test environments and by using experimental data recorded from actual multi-sensor configurations.
Published in:	Proceedings of the 26th International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2013) September 16 - 20, 2013 Nashville Convention Center, Nashville, Tennessee Nashville, TN
Pages:	2075 - 2083
Cite this article:	Soloviev, A., Yang, C., "Reconfigurable Integration Filter Engine (RIFE) for Plug-and-Play Navigation," Proceedings of the 26th International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2013), Nashville, TN, September 2013, pp. 2075-2083.
Full Paper:	ION Members/Non-Members: 1 Download Credit Sign In