Download presentation
Presentation is loading. Please wait.
1
9th CGF & BR Conference 16 - 18 May 2000 Copyright 1998 Institute for Simulation & Training Modeling Cooperative, Reactive Behaviors on the Battlefield with Intelligent Agents Dr. Thomas R. Ioerger Dr. Richard A. Volz Dr. John Yen University XXI Group Department of Computer Science Texas A&M University College Station, TX
2
9th CGF & BR Conference 16 - 18 May 2000 Copyright 1998 Institute for Simulation & Training University XXI Program Funded through STRICOM Leverage innovative research and technology development in academia to support the Army’s Digitization effort and Force 21 Joint collaboration between Texas A&M and UT-Austin Multiple on-going tasks
3
9th CGF & BR Conference 16 - 18 May 2000 Copyright 1998 Institute for Simulation & Training Task 3, FY’99 Objective: develop intelligent agent technology to support more effective digital battlestaff training for TOC officers Want to simulate more intelligent autonomous behaviour at aggregate levels (company, battalion, brigade) in simulations
4
9th CGF & BR Conference 16 - 18 May 2000 Copyright 1998 Institute for Simulation & Training State of the Art in Training Most training provided by networked battlefield simulations (e.g. JANUS, MODSAF ) Run “scripted” scenarios to provide decision- making opportunities Limited responsiveness and interactivity Often requires “pucksters” to play role of subordinates or enemy
5
9th CGF & BR Conference 16 - 18 May 2000 Copyright 1998 Institute for Simulation & Training Technical Challenges Want commanded units to nominally follow orders, procedures, doctrine, etc. Units must be adaptive to react to unexpected changes in conditions (e.g. opportunities, threats, etc.) Want to capture cooperative interactions (information sharing, coordination) Hence must model TOC teamwork process
6
9th CGF & BR Conference 16 - 18 May 2000 Copyright 1998 Institute for Simulation & Training Overview of Approach 1. Develop a knowledge representation language for tasks, methods, and procedures 2. Develop an agent algorithm (e.g. interpreter) for hierarchical plan generation and dynamic plan monitoring & execution 3.Create multi-threaded implementation in Java 4.Create hooks into OneSAF Testbed (OTB) and graphical interfaces for reports, etc.
7
9th CGF & BR Conference 16 - 18 May 2000 Copyright 1998 Institute for Simulation & Training Task Representation Language (TRL) Provides descriptors for: goals, tasks, methods, and operators Tasks: “what to do” Methods: “how to do it” Operators: lowest-level actions that can be directly executed in simulation environment, e.g. move unit, send message, fire on enemy Each descriptor is a schema with arguments and variables
8
9th CGF & BR Conference 16 - 18 May 2000 Copyright 1998 Institute for Simulation & Training Tasks –can associate alternative methods, with priorities or preference conditions –can have termination conditions Methods –can have pre-conditions that must be satisfied –process net procedural language for specifying how to do things while loops, if conditionals, sequential, parallel constructs can invoke sub-tasks or operators semantics based on Event Logics Conditions are treated as queries to Jess
9
9th CGF & BR Conference 16 - 18 May 2000 Copyright 1998 Institute for Simulation & Training Examples of TRL Knowledge (:Task attack-enemy (?company-id ?enemy-id) (:Term-cond (enemy-destroyed ?enemy-id)) (:Method (call-for-indirect-fire ?company-id ?enemy-id) (:Pref-cond (have-priority-of-fire ?company-id))) (:Method (attack-enemy-with-mortars ?company-id ?enemy-id) (:Pref-cond (small ?enemy-id) (have-mortar-assets ?company) (< (dist (position ?company-id) (position ?enemy-id)) 7-km)) (:Method (attack-enemy-by-direct-fire ?company ?enemy) (:Pref-cond FALSE))) (:Method M1 (?a) (:pre-cond (not (moving ?a))) (:term-cond (moving ?a)) (:Process (:Parallel (:Sequence T1 T2 T3) (:Branch (:Cond (weather raining)) T4 T5) ) ) )
10
9th CGF & BR Conference 16 - 18 May 2000 Copyright 1998 Institute for Simulation & Training Agent Algorithm Top-level task decomposes into tree of methods and sub-tasks Select methods based on current situation: evaluate method preferences, operator pre- conds, etc. Keep expanding till hit operators at leaves May be multiple “first steps” due to parallelism Pick one (randomly) as initial action
11
9th CGF & BR Conference 16 - 18 May 2000 Copyright 1998 Institute for Simulation & Training Illustration of Task Decomp. Tree
12
9th CGF & BR Conference 16 - 18 May 2000 Copyright 1998 Institute for Simulation & Training After first time step is complete, tell thread for selected action to “step” forward; remaining actions are still pending in the tree Important: after each time step, re-assess validity of all significant decision points in tree Allows reactivity by detecting termination conditions, failed pre-conditions, etc. If conditions have changed, back-track and dynamically re-construct new branch of tree
13
9th CGF & BR Conference 16 - 18 May 2000 Copyright 1998 Institute for Simulation & Training Overview of System Architecture
14
9th CGF & BR Conference 16 - 18 May 2000 Copyright 1998 Institute for Simulation & Training Knowledge Acquisition Simulate Bn TOCs for training brigade staff Movement-to-contact (MTC) scenario Task knowledge in TRL Domain knowledge (inference rules) in Jess Interviews with military experts Training manuals, e.g. FM 101-5 Prior C2 & BFA cognitive task analyses
15
9th CGF & BR Conference 16 - 18 May 2000 Copyright 1998 Institute for Simulation & Training Interfaces Hooks into OTB (read DIS PDUs) Puckster interface (to eventually be replaced) Brigade interface (for S2, S3, FSO, etc.): –receive reports from Bn’s: SALTs, SITREPs, RFIs, RFSs –send reports to Bn’s: RFIs, INTSUMs, OPORDs, FRAGOs Implemented in Java using Swing and RMI
16
9th CGF & BR Conference 16 - 18 May 2000 Copyright 1998 Institute for Simulation & Training Example in MTC Scenario 1) Info from Bde recon team comes into Bn TOC about enemy in the area 2) Bn agent sends scout plattoon in nearby OP to monitor enemy - from on-going Monitor_Enemy task 3) Bn agent orders nearby company to move to and engage enemy - from on-going MTC task 4) Based on assessed level of threat, Bn agent requests artillery (if has priority of fire) or close air support (if available) - from on-going Protect_the_Forces task
17
9th CGF & BR Conference 16 - 18 May 2000 Copyright 1998 Institute for Simulation & Training Conclusions Agent technology can be useful in generating intelligent, autonomous, adaptive, and cooperative behaviors of aggregates in sim. Developed a knowledge representation language (TRL) for capturing task/procedural knowledge Developed an interpreter algorithm for providing reactive behavior (plan monitoring and execution)
18
9th CGF & BR Conference 16 - 18 May 2000 Copyright 1998 Institute for Simulation & Training Conclusions Can capture teamwork processes in TRL (e.g. information exchange between Bn and Bde), which is essential to TOC staff training Future work –expand knowledge acquisition –link into ABCS boxes –greater interaction with OTB –add temporal reasoning to TRL
Similar presentations
© 2025 SlidePlayer.com Inc.
All rights reserved.