Clusters for Autonomous Deep Space Missions
NASA Jet Propulsion Laboratory
California Institute of Technology
A revolution in unmanned spacecraft for remote exploration will be precipitated by a new generation of onboard computing based on advanced clusters derived from commercial components. The NASA Jet Propulsion Laboratorys Remote Exploration and Experimentation (RE&E) program is developing a new family of spacecraft scalable computing systems that will permit the exploitation and tracking of rapid improvements in commercial device technology and design, provide high performance capability at low cost, and deliver high availability through software fault tolerance and graceful degradation. More than reducing costs, cluster-based spaceborne computing platforms enables a critical paradigm shift in the role and methodology of applying onboard processing to meeting mission requirements for deep space missions. Historically, slow radiation hardened microprocessors representing architectures that date back as much as twenty years ago (e.g. Mars Pathfinder used an 80C85) were employed for the simple purpose of data acquisition from onboard sensors, data compression, and downlink to Earth ground-based systems of the Deep Space Network. But future missions will require an entirely new strategy of autonomous mission operation driven by on-board high performance computing. The principal motivations are the reduction of total mission costs by as much as an order of magnitude and the implementation of mission types simply infeasible if limited to ground-based near real-time supervision and control. By engaging in spacecraft data processing, only the science product need be returned to Earth with potential bandwidth savings of between 4 and 7 orders of magnitude, thus greatly reducing the size, weight, power, and cost of spacecraft transmission, power, volatiles, and propulsion. Asteroid mapping and ephemeredes determination requiring the acquisition and processing of tens of thousands of images need only return the data product of less than a hundred Mega bytes through the mission duration. Spacecraft employing advanced AI based planning, control, and analysis will be able to operate in contexts beyond Earth control such as a fly-by of Pluto and its moon, the traversing the surface of Titan hidden by its dense atmosphere, or swimming the seas of Europa below kilometers thick ice exploring for unimagined forms of life. This talk will describe the challenges and requirements of this new class of spaceborne computer systems and discuss the RE&E strategy for meeting their needs through software implemented fault tolerance on commercial processor clusters.