100 likes | 114 Views
Discover the world of Knowledge-Intensive Agents (KIAs) in defense modeling and simulation projects. This summary covers features, projects, and components related to KIAs, highlighting their importance in modern technological applications. Explore the interactive capabilities, knowledge representation structures, and innovative uses of KIAs in various projects and domains.
E N D
Thinking… …inside the box Knowledge-Intensive Agents in Defense Modeling and Simulation Randolph M. Jones
What is a Knowledge-Intensive Agent? • A software system that • Is interactive with an external environment • Incorporates a fairly large body of long-term knowledge • Introduces unique concerns about the organization, implementation, and run-time use of knowledge • Creates an maintains significant internal representations of its situational understanding • This summary also covers agents with smaller knowledge bases, but that are designed in the same spirit as K-I agents
Features of K-I Agents • Knowledge representation must support relational pattern-based matching and retrieval • Long-term knowledge must be retrieved associatively and efficiently • Run-time data representations must support multi-valued relations (and pattern matching) • Some form of truth maintenance system should support efficiency and consistency in situational representations • For maintainability and extensibility (and perhaps efficiency), knowledge representation language must support alternative high-level organizations (e.g., goals, beliefs, problem spaces) • Logic flow of decision making must be re-entrant and use least commitment
K-I Agents in Soar Technology M&S Projects • JFCOM • Joint Urban Operations Human-In-The-Loop Experiment • SOF-Soar agents masquerade as civilians, recon enemy forces • Move along roads using JSAF path planning libraries • Pass contact information to SLAMEM sensor modeling system • Accept user tasking from JSAF GUI • Display planned and used routes on JSAF GUI • Enduring Freedom Reconstruction • SOF-Soar agents call in air strikes to TacAir-Soar agents • TacAir-Soar agents add behaviors for strafing, laser-guided CAS, bomb patterns, B-52 missions • SOF-Soar integration with DI-Guy to provide high resolution visual representation of individual combatants • Fleet Battle Experiment/Millennium Challenge ’02 • TacAir-Soar agents integrate with TBMCS to receive air tasking orders, launch from carriers, fly naval air missions
K-I Agents in Soar Technology M&S Projects • Automated Wingman • Provide Army helicopter teammates for human pilots in experimental scenarios • SOF Air Ground Interface Simulation (SAGIS) • Provide Close-Air Support and Indirect Fire behaviors to support training of Terminal Air Controllers • Advanced Global Intelligence and Leadership Environment (AGILE) • Simulate national or organizational decision making • General dynamics scout robot (and simulation) • Soar integration to GDRS control architecture • Team coordination, route planning/following
K-I Agents in Other Projects • User interface agents • Cooperative Interface Agents for Networked Command, Control, and Communications (CIANC3) • Assist command, control, and communication during mission execution • Battlespace Information and Negotiation through Adaptive Heuristics (BINAH) • Provide data fusion and information display for time critical targeting • Knowledge Enablers for the Unit of Action (KEUA) • Determine requirements, acquire, fuse, and present information to support command decision making • VISTA Explanation Agents • Architecture-neutral facility for communicating agent behavior explanations to end users through the VISualization Toolkit for Agents
“Knowledge/Agent Components” • Soar Technology Goal System (STGS) • Declarative means-ends-analysis style representations of goal and operator trees • Onto2Soar • Structured declarative knowledge with compiler to procedural Soar productions (CIANC) • Communications infrastructure • Layered transport- and content-neutral processing of agent communications (VIRTE, TacAir-Soar, Automated Wingman, SAGIS) • SoarSpeak • Speech recognition/generation integrated with Soar
“Knowledge/Agent Components” • High Level Symbolic Representation language • High level object- , agent-, and symbol-oriented behavior language with compiler to Soar productions • TCL-based parser providing HLSR prototype implementation • Qualitative spatio-temporal models/representation • Structured representation and reasoning over events in real time and projected time (Augmented Warrior, BINAH, HLSR) • Qualitative spatial representations and reasoning (BINAH) • End-user specification language for data fusion, display, and high-level knowledge design with compiler to Soar productions (AGILE, BINAH) • Deontic reasoning infrastructure • Structured representations of authority, obligation, permission, responsibility for teamwork (CIANC)
Nuggets • Soar directly supports much of the essential low-level functionality for K-I agents • We are using Soar to build a wide variety of K-I agents • We are developing a number of supporting components and technologies for developing K-I agents (within Soar and otherwise)
Lumps • We still need improved high-level organizations and tools for managing large agent knowledge-bases • Customer enthusiasm for K-I agents waxes and wanes • Most of the supporting components and technologies are not yet “ready for prime time”