Thinking… …inside the box Knowledge-Intensive Agents in Defense Modeling and Simulation Randolph M. Jones © 2004 Soar Technology, Inc. Slide
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 © 2004 Soar Technology, Inc. 16 March 2018 Slide 2
Features of K-I Agents ¡ Knowledge representation must support relational patternbased 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 © 2004 Soar Technology, Inc. 16 March 2018 Slide 3
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 Tac. Air-Soar agents · Tac. Air-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 · Tac. Air-Soar agents integrate with TBMCS to receive air tasking orders, launch from carriers, fly naval air missions © 2004 Soar Technology, Inc. 16 March 2018 Slide 4
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 © 2004 Soar Technology, Inc. 16 March 2018 Slide 5
K-I Agents in Other Projects ¡ User interface agents Cooperative Interface Agents for Networked Command, Control, and Communications (CIANC 3) · 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 © 2004 Soar Technology, Inc. 16 March 2018 Slide 6
“Knowledge/Agent Components” ¡ Soar Technology Goal System (STGS) Declarative means-ends-analysis style representations of goal and operator trees ¡ Onto 2 Soar Structured declarative knowledge with compiler to procedural Soar productions (CIANC) ¡ Communications infrastructure Layered transport- and content-neutral processing of agent communications (VIRTE, Tac. Air-Soar, Automated Wingman, SAGIS) ¡ Soar. Speak Speech recognition/generation integrated with Soar © 2004 Soar Technology, Inc. 16 March 2018 Slide 7
“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) © 2004 Soar Technology, Inc. 16 March 2018 Slide 8
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) © 2004 Soar Technology, Inc. 16 March 2018 Slide 9
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” © 2004 Soar Technology, Inc. 16 March 2018 Slide 10
Скачать презентацию Thinking inside the box Knowledge-Intensive Agents in Defense
ba67055ee1cec9d7bf458a096141aeb5.ppt