Space domain awareness requires coordination and tasking of networks of ground-based sensors to observe, track, and identify space objects of interest. As the resident space object population grows rapidly, traditional person-in-the-loop systems become challenging and cumbersome to operate. Smart sensing and machine learning have the potential to revolutionize how ground-based remote sensing of space objects is conducted by automating sensor tasking and data processing to rapidly and accurately provide awareness to decision-makers on the ground.
Characterization of space objects is a remote sensing problem that involves classifying physical state (e.g. orbit), positively identifying space objects, and detecting events such as when two objects are closely spaced. Numerous sensor modalities exists, varying across field of view, resolution, and observed wavelength (e.g. visual vs infrared), for example, and the sensor modality impacts how it is utilized. Tasking a network of sensors to maintain a catalog of space objects efficiently (or even optimally) is a challenging problem. Machine learning techniques have been applied to remote sensing to improve object detection and perception; however, what we seek are novel methods, machine learning or otherwise, to autonomously task and operate a network of various remote sensors to maintain custody of known space objects and identify and adapt to new objects or events in space. Methods for achieving these results may incorporate reinforcement learning, and “optimal” decision-making may be tied to notions of information gain, but we are open to proposals. We are interested not only in how to frame and model this autonomous sensing problem, but also on any emergent behavior that develops.
Sign up now to be invited to the webinar series where we’ll discuss the context behind each problem statement and answer questions from startup and university teams.