Machine Learning to Enhance AF Simulator Training Systems

The application of machine learning technologies to a broad range of systems and problems has been under development for many years. The emergence of feature based adaptive algorithms as well as generalized deep networks has enabled the broad application. All of these techniques are dependent, at least initially, on labeled data. Many applications have serious confidentiality, privacy or propriety issues associated with their data sets. i.e, Limitations and bounds are placed on data. Our ML techniques have been focused on commercial applications in a wide range of markets - financial, biometric, monitoring - but all have been limited by data sources. We are proposing to use our ML techniques applied to AF VR enabled simulation systems to enhance and automate these systems as a low cost training technique to help pilots build and retain proficiency. This is enabled because the AF generates labeled simulation data every time these systems are used. This capability can be applied to ML enabled training and later can mature out of the simulation environment to become a pilot aid or core engine of an autonomy system.