Previous Abstract Return to Session A6

Session A6: Adaptive KF Techniques, Data Integrity, and Error Modeling

Towards an Online Sensor Model Validation and Estimation Framework
Juan D. Jurado and John F. Raquet, Air Force Institute of Technology
Location: Big Sur

Over the past two decades, a large portion of navigation research has focused on using multiple sensors to aid navigation algorithms in the absence of GPS. These “all-source” navigation research efforts require diligent calibration of each sensor’s measurement model, and robust monitoring of its performance in order to detect model changes or sensor failures. In order to bring all-source navigation to operational readiness, this technology must be developed to allow useful sensors with known and unknown characteristics to be properly integrated, validated, and calibrated by a navigation computer in an online or plug-and-play framework.
Current all-source navigation research to date has focused on online calibration of alternative sensors with known model parameters, basic statistical methods for assessing sensor model adequacy, or adaptive methods for managing sensor failures. Additional research in the plug-and-play framework area has also focused on sensor-specific solutions for online calibration or optimal computational resource management in the presence of multiple sensors.
This research aims at improving overall all-source navigation capabilities by developing a generalized, comprehensive sensor validation and integration framework for navigation applications. The proposed framework will integrate robust statistical model estimation, as well as diagnostic and remedial techniques, into a general process for validating, calibrating, and re-modeling alternative navigation sensors, all within a real-time, or online environment. The framework is divided into three distinct phases: online sensor model validation, online sensor calibration, and online sensor model repair.
The framework-specific research will entail developing the criteria, tasks, and interactions between each of the three proposed phases. Phase I research will provide a robust online model diagnostics process for all-source navigation sensors and will be composed of a set of comprehensive model diagnostic tests in an online or real-time environment. The diagnostic tests include but are not limited to: lack of fit, constant error variance, outlying measurements, correlated error terms, and Normality of distribution. Phase II research will entail integrating sensor-specific online calibration techniques (which are prolific within current literature) into the overall framework as well as providing general non-sensor-specific techniques for model calibration. Phase III research will provide an advanced online model repair process for all-source navigation sensors and will be composed of a set of comprehensive model remedial measures, each aimed at resolving any issues identified in the previous two phases.
Given the scope of the proposed research, the subset to presented will lay the foundational requirements for the overall framework, highlight the various statistical model diagnostic and repair techniques to be used, and present preliminary results from the online model validation and repair phases. Most notably, a novel online sensor model validation and repair technique, which is driven by likelihood maximization, will be presented as a key enabler in Phases I and III of the proposed framework. The work presented will directly enable plug-and-play operational use of trusted and untrusted sensors into an all-source, alternative navigation system by ensuring sensor models provided to the system are validated and/or repaired prior to full integration into the navigation solution.



Previous Abstract Return to Session A6