Dual Knowledge Representations for Adaptive & Efficient Autonomous Agents
Abstract
As large-scale autonomous agents take on more complex missions with greater responsibilities, a significant challenge is ensuring th,at their behavior is consistent with the multiplicity of interacting constraints that are inherent to military missions. These inclu,de established doctrine, tactics, and rules of engagement (ROE), commanders intent, as well as ethics, safety, and norms. Because o,f the multiplicity of ways constraints can apply to different situations and can interact with other constraints and mission paramet,ers, existing approaches are inadequate for developing agents that can interpret and apply such constraints without interfering with, real-time task execution. We propose to develop novel approaches for real-time intelligent agents that modulate their performance u,sing abstract behavioral constraints during autonomous operation.The core research challenges of the proposed project are developing,, implementing, and evaluating 2) formalisms for encoding and interpreting abstract behavior constraints in realistic Navy mission e,nvironments and 2) metacognitive strategies for efficiently monitoring and operationalizing constraints in novel situations. The res,earch is critical for development of trustworthy and flexible autonomous agents; ones that not only autonomously execute complex tas,k behaviors, but that can also adjust and specialize their behavior to the vast range of situations and constraints common in today,s complex missions.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Apr 01, 2022
- Source ID
- N000142212358
Entities
People
- John E. Laird
Organizations
- Office of Naval Research
- United States Navy