Interplanetary exploration occurs at vast distances that severely limit communication bandwidth to spacecraft exploring other planets. It is possible to collect much more scientific data than can ever be downlinked given current communication capabilities. Therefore, we are developing a system called COSMIC (Content-based Onboard Summarization to Monitor Infrequent Change) that will opportunistically analyze data onboard a Mars orbiter to alert scientists when meaningful changes have occurred. COSMIC will allow future spacecraft to continuously collect data to search for rare, transient phenomena such as fresh impacts or seasonally changing polar landforms under a constrained downlink budget. In this paper, we describe the overall goals and architecture of COSMIC, plans to enable specific scientific studies, label acquisition to enable supervised approaches to surface landform classification, a new machine learning evaluation framework for analyzing the trade-offs between classifier accuracy and computational requirements, and lessons learned about constraints that COSMIC will face operating onboard a spacecraft. In particular, we discuss design considerations surrounding computational and storage constraints, change detection strategies, and localizing detected landforms of interest within a global coordinate frame. Finally, we describe challenges and open research questions that must be addressed prior to deploying COSMIC.
展开▼
