A Constraint-Based Planner Applied to Data Processing Domains

Abstract

Earth-science data processing at NASA is the problem of transforming low-level observations of the Earth system, such as data from Earth-observing satellites, into high-level observations or predictions, such as crop failure or high fire risk. Given the large number of socially and economically important variables that can be derived from the data, the complexity of the data processing needed to derive them and the many terabytes of data that must be processed each day, there are great challenges and opportunities in processing the data in a timely manner, and a need for more effective automation. Our approach to providing this automation is to cast it as a constraint-based planning problem: we represent data-processing operations as planner actions and desired data products as planner goals, and use a planner to generate data-flow programs that produce the requested data. The planning problem is translated into a constraint satisfaction problem (CSP) and solved by constraint propagation and search algorithms.