Capturing Collaborative Decision-Making Rationale during Science Mission Operations

[Abstract] The development of increasingly sophisticated autonomous vehicles capable of being deployed from great distances and for extended periods of time brings new challenges, including tracing collaboratively-made decisions regarding these assets and relating collected data to such decisions and their rationale. For scientific data gathering, this rationale commonly is in the form of research questions, hypotheses, and observations. By “closing the loop” between motivations and rationale for gathering data and the data themselves, their intended uses are made more clear, and a precise, evaluable trace of mission actions is generated. Tracking asset deployment rationale facilitates long-term mission planning and, particularly for complex data gathering missions, can prevent repetitive or extraneous deployments and reduce the likelihood of overlooking deployment opportunities. It also can provide a valuable reference for planning subsequent missions. We have had several opportunities to develop ontological models for describing data gathering rationale and linking them to collected data stored in the ScienceOrganizer semantic information repository. The models include concepts for describing assets, activities, and various types of scientific rationale, and also, importantly, the logical connections to data products (e.g., instrument data). We have demonstrated use of these models: for automated scientific operations for the NASA Astrobiology Institute; as part of NASA’s Mobile Agents Architecture; to demonstrate round-trip data tracking for the Mars Exploration Rovers (MER) missions; and for the Adaptive Ocean Sampling Networks Monterey Bay 2006 data collection campaign (MB06). The decision-making processes in each of these operational settings have a number of similarities: the generation of research questions and/or specific hypotheses that may be tested by observations; negotiation among multiple parties on proposals for observations through voting or other processes; execution of observations yielding instrument data; analyses of data that can refute or support hypotheses. This suggests the kinds of decision and data-tracking models we have developed could be applied to a wide variety of collaborative-control data gathering missions. However, a key problem is rationale acquisition; manual entry of rationale knowledge is quite labor intensive. Further efforts to improve rationale capture and develop specialized applications to evaluate and display rationale information are needed.