Non-Cooperative Spacecraft Pose Estimation Using Convolutional Neural Networks

With the advent of privatized space flight and the growing number of satellites in low earth orbit, the ability to remove subsequent debris and defective satellites is increasing in need. Autonomous spacecraft will likely be at the forefront for this due to the dangerous nature of approaching uncontrolled space debris as well as the logistics of removing large amounts of debris from earth orbit. This thesis provides a new convolutional network architecture that is capable of tracking three degrees of motion of the test platforms while be able to be deployed on an embedded platform. The novel convolutional model, coined SPOTNet, uses the input of a stereo camera to be able to resolve the relative x, y and attitude of the target spacecraft