The proposed KFlow cognitive architecture supports the continuous acquisition, storage, and application of diverse knowledge to provide on-board autonomy and crew assistance during deep space missions.
KFlow comprises a knowledge repository and an intelligent agent. The repository stores diverse knowledge objects that support mission operations, such as system models, planning models, flight rules/notes, and procedures. These knowledge objects are enriched with annotations to support concept-based retrieval and automated reasoning. They are linked to related objects from which they are derived and to supporting datasets, documents, diagrams, and engineering studies (i.e., their knowledge provenance). Using the TEAM toolkit will streamline KFlow’s development.
The KFlow agent acquires knowledge (i.e., learns) by deriving models of system or crew behavior from data, using the models to assess situations and generate procedures, and posing questions to human experts, such as flight controllers, to elicit rationale for other knowledge and hypotheses. The agent applies knowledge to respond to crew requests for assistance; estimates crew mental state and intervenes when appropriate; and responds to system failures, hazardous conditions, or new priorities.
KFlow is innovative because it provides a framework for managing the multi-step transformation and flow of source data, implicit knowledge, and unstructured data into more actionable knowledge that is used to provide automation and crew assistance. KFlow’s approach is pragmatic because it flexibly tailors its assistance to the depth of knowledge it possesses.
During the project, we will identify KFlow functional requirements, develop a concept of operations & high-level system design, develop test scenarios to clarify requirements and design challenges, create a prototype that illustrates our approach and demonstrates feasibility, and develop a preliminary high-level design & project plan to be implemented during Phase II.
The primary NASA application will be the provision of autonomy and crew support during long-duration missions, especially when continuous and comprehensive Ground support is not feasible or economical. KFlow could also be used on Earth to capture and share expertise between back room and front room flight controllers or between more experienced or specialized controllers and less specialized controllers who handle multiple positions during periods of low activity.
On Earth, the resulting technology could provide a foundation for autonomous operations and user assistance for knowledge-intensive tasks. Examples include engineering design, planning, military command and control, facility maintenance, smart buildings, laboratory automation, autonomous manufacturing and logistics, smart grids, and other critical infrastructure.