SBIR-STTR Award

For any NASA mission, a key component to the operational success of the mission, for both operations planning and science data processing, is orbit determination. Traditional methods of orbit estimation are very good at producing state estimates when the force and data noise models are well defined. However, practical implementations of orbit estimation require a large staff of operations personnel to manually adjust tuning parameters to account for irregularly occurring but repeatable variations in the force and data noise models. Such labor intensive processes are costly. AI Solutions proposes to develop a self-tuning orbit estimation system that uses a Kalman filter to estimate orbit states and a neural network to actively monitor and control the estimation process. A neural network integrated with a Kalman filter will remember and classify the complex relationships between dependent (state estimation) and independent (tuning parameters) data. Based on these complex relationships and error feedback, a neural network will tune the Kalman filter, thereby replacing the manual tuning process and significantly reducing operations costs. Once proven on the ground, this system can then be integrated with on-board maneuver control systems for an autonomous on-board navigation and control system.