Goal-Driven Autonomy and Robust Architectures for Long-Duration Missions
Abstract
We propose to articulate a computational foundation for robust long-duration agent autonomy and to provide a prototype implementation that exhibits goal-driven autonomy on an actual physical platform. We will integrate a robotic agent architecture with the MIDCA metacognitive architecture by formalizing an abstract model of the robot control structure thereby enabling MIDCA to reason over representations that instantiate this model. Robustness is ensured by adaptation to uncertain and changing environments while performing long-duration activities. The key idea is to implement a metacognitive mechanism that records traces of agent reasoning as it dispatches between functional modules in service of its prior goals. If these traces are structured in a representation that is understandable by the meta-level of the MIDCA architecture, then MIDCA can reason about the agent’s decision making as well as the agent’s behavior. Control is then passed back through the individual modules in terms of new goal structures and subsequently to agent effectors. By doing so, not only will this proposed project integrate two different inferential mechanisms, but it will coalesce research ideas across theory and applications, provide a mechanism to interpret and understand potential failures and their root causes in hardware and software, and, we expect, generate novel representations and inferences in metacognitive systems, thus providing a more robust robotic controller for longduration autonomous operations.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Aug 12, 2016
- Source ID
- N000141512080
Entities
People
- Michael T. Cox
Organizations
- Office of Naval Research
- United States Navy
- Wright State University