GPU-Enabled Large-Scale State Estimation

Recursive filtering and batch estimation problems can be found in a myriad of applications of greatest interest for both DoD and civilian applications. Air Force test ranges routinely apply these methodologies to assess, validate, and verify performance of critical assets, for a variety of vehicles and weapon systems. A recurrent scenario is the need to precisely measure maneuvers, trajectories, and similar navigation tasks to assess the performance of such systems in a relevant deployment scenario. Precision navigation requires combining heterogeneous and complementary data sources, such as GPS, IMUs, and potentially other sensors, to improve the accuracy of the position and time sources. While the application is always similar, the performance, number of sensors, quality of sensors and their data rates varies greatly between test events and systems being tested.  Therefore, Lynntech proposes to develop general purpose computing tools that allow researchers and technician to customize trajectory estimation solutions using state of the art hardware and software resources. Lynntech will exploit the general structure of such estimation problems, in addition to advances in massively parallel computing, to develop self-structuring and self-tuning software capable to provide a fast prototyping and computing environment to obtain high quality trajectory data in a significantly reduced time frame, without sacrificing accuracy. In Phase I we have focused our attention on developing methods for solving Non-Linear Least Square problems relative to trajectory estimation using a sensor fusion approach. The objective of the Phase II effort is to a) Improve the performance and generality of the parallel algorithms, and b) build a robust user interface layer that can be used by Air Force personnel and then commercially to solve these type of problems in a much reduced time frame and with greater scalability whether the computing resources (GPUs for instance) are located locally on a server or remotely, in the cloud.