Neural Space Navigator

The Neural Space Navigator (NSN) is an innovative fully autonomous navigation system for deep space or near the surface of moons, asteroids or planets, It employs dual cameras. One is on a pan and tilt platform for tracking celestial objects and for terrain tracking. The other is a star camera that provides a star reference. An optional sun distance sensor provides a sun range nbsp;reference in interplanetary space. The terrain tracker is a neural network trained using Deep Learning. Training is done offline from images taken by other spacecraft or by the spacecraft itself prior to use in the navigation system. The neural network determines points in 3 dimensions that provide a 3 dimensional vector as an input to the navigation Kalman Filter. Navigation and attitude determinationnbsp;are performed with a Schmidt Form Unscented Kalman Filter with an IMU as the navigation and attitude. base. NSN can seamlessly incorporate other position and velocitynbsp;information, such as from communications links or from GPS. The system alsonbsp;includes a neural network for obstacle detection and avoidance.nbsp;The version shown in the proposal computes a two dimensional trajectories and obstaclenbsp;by training on sub images using a classification convolutional neural network. The project will explore more sophisticated object identification and tracking.A laboratory prototype will be built and tested during the project. The navigation system will be demonstrated in an existing high fidelity lunar spacecraft simulation. The simulations will include an end-to-end simulation from LEO to the lunar surface.The hardware/software combination is small enough for a 6U CubeSat. The software can be used on any spacecraft with a camera for a wide variety of applications. The MATLAB software will be delivered at the end of Phase I for immediate use by NASA engineers. In Phase II Lockheed-Martin, which provided a letter of support, will support the development of flight hardware.