Presentation Transcript
Soar-RWA:Planning, Teamwork, and Intelligent Behavior for Synthetic Rotary Wing Aircraft : Soar-RWA: Planning, Teamwork, and Intelligent Behavior for Synthetic Rotary Wing Aircraft Randall W. Hill, Jr., Johnny Chen, Jonathan Gratch, Paul Rosenbloom, Milind Tambe University of Southern California Information Sciences Institute
The Synthetic Battlefield : The Synthetic Battlefield Purpose
Training
Mission planning and rehearsal
Tactics and doctrine development
Weapon-system concept evaluation
Project Goals : Project Goals Build Virtual Humans and other Intelligent Agents
Apply these agents in a variety of different domains
Goals for Synthetic Force Agents : Goals for Synthetic Force Agents Act autonomously for days at a time
Reduce load on human operators
Behave in human-like manner
Produce high quality simulation
Perform commander’s functions
Reduce the number of human operators
Outline : Outline High level capabilities
AI Technology
Soar
Perception
Teamwork
Planning
STOW-97 ACTD results
Current research
Pilot Agents : Pilot Agents Fly helicopters in synthetic terrain
Requires perceptual and motor skills
Execute missions
Reactive planning
Act as a team
Coordinate activities
Recover from failures
Command Agent : Command Agent Plan missions
Monitor execution of plans
Communicate with superiors & subordinates
Mission Capabilities : Mission Capabilities Army Deep Attack
AH-64C Apache Rotary Wing Aircraft (RWA)
Company-level planning and execution
Attack Helicopter Battalion deployed
Marine Transport
CH-46E Sea Knight & CH-53E Super Stallion
Marine Escort
AH-1W Super Cobra
Army Deep Attack Mission : Army Deep Attack Mission Engagement Area Assembly
Area Battle Position Holding
Area Phase Line Phase Line LD ~5 km ~70 km
Marine Transport/Escort Mission : Marine Transport/Escort Mission Landing
Zone
Architecture : Architecture
ModSAF Simulator Pilot Agent Distributed Interactive Simulation Network Entity State Pilot Agent Pilot Agent Pilot Agent Company Commander Agent Battalion Operation Order Company Operation Order Situation Report Sensing Rotary Wing Aircraft Rotary Wing Aircraft Rotary Wing Aircraft Rotary Wing Aircraft
AI Technology : AI Technology High level capabilities
AI Technology
Soar
Perception
Teamwork
Planning
STOW-97 ACTD results
Current research
Soar : Soar Rosenbloom, Laird, & Newell
Integrated intelligent system
Unified theory of cognition
Represent tasks and goals as operators
Rules and subgoals:
Suggest and compare operators
Execute and terminate operators
Operator hierarchy for pilot agent
Pilot Agent:Partial Operator Hierarchy : Pilot Agent: Partial Operator Hierarchy Execute-Mission Fly-Flight-Plan Engage Prepare-to-
return-to-base Fly-control-route Select-
point Select-
route High-
level Low-
level Contour NOE Mask Unmask Employ-
weapons Initialize-
hover Return-
to-
control-
point
Pilot Agent Architecture : Pilot Agent Architecture
ModSAF Simulator Pilot Agent (Soar)
Rotary Wing Aircraft
Distributed Interactive Simulation Network Entity State Motor Commands Perception Soar-ModSAF Interface
Perception (Randy Hill) :
ModSAF Perception (Randy Hill) Challenges:
Rich visual field overloads visual processing
Need to understand groups of entities Pilot Agent (Soar) Rotary Wing Aircraft tank tank tank tank air defense artillery tank tank air defense artillery artillery artillery artillery artillery Other Battlefield Entities: Sense
Perception of Groups : Perception of Groups Filter out unneeded details
Agent can dynamically change filters
Perceive groups of entities
Cluster by proximity
bottom-up, automatic
Cluster by attribute
voluntary, used for tracking
Group attributes
used for tracking, filtering, understanding
Teamwork (Milind Tambe) : Teamwork (Milind Tambe) STEAM: General model of teamwork
Utilizes joint intentions framework
(Cohen & Levesque, 91)
Monitor team progress
Individual may fail or succeed in own role
Reorganize when critical role failures
Reassign critical roles based on joint intentions
Decision theoretic communication selectivity
Planning (Jon Gratch) : Planning (Jon Gratch) IPEM-like planner in Soar
Ambros-Ingerson & Steel, 1987
Uses hierarchical, partially-ordered tasks
Tasks have duration, asynchronous effects
Interleaves planning and execution
Unanticipated/anticipated exogenous events
Alter plans in response to world changes
Hierarchical Task Network : Hierarchical Task Network Attack Move (HA) Move (BP) Engage (E) Move (LD) Passage of
Lines Move (HA) Destroyed (E) @(HA) @(HA) @(BP) @(LD)
STOW-97 ACTD : STOW-97 ACTD Army Attack Missions
Three company-level missions
Five Apaches per company + Company Commander
Last-minute planning
Scrapped pre-planned missions—OPFOR locations
Employed Exercise Editor and OpOrder Generator
Mission Results
Encountered large bodies of OPFOR
Destroyed 82 OPFOR, lost 3 Apaches
STOW-97 ACTD : STOW-97 ACTD Marine Transport/Escort Missions
Five company-level missions
8 CH-46E’s, 4 CH-53E’s, and 4 AH-1W’s
Last-minute Planning
Employed Marine RWA Exercise Editor
Mission Results
Completed a battalion-level lift & 2 company lifts
Development Effort : Development Effort Army Pilots: 1577 rules; 93 operators
Marine Pilots: 1341 rules; 60 operators
Significant sharing with Army pilots
Commanders: 826 rules; 116 operators;
Soar-ModSAF Interface
Sensing & perception, weapons, vehicle dynamics, communication
Current Work : Current Work Group Perception
Focus of attention
Recognize low-level aspects of group behavior
Group Understanding
Model others by “thinking like them”
Understand dissimilar others
Current Work : Current Work Group Planning
Scale-up to Battalion-level
Develop collaborative planning techniques
Learning and Emotions
Learn from experience
Model effects of fatigue and stress
Incorporate emotions in Soar architecture
Benefits : Benefits Provides command structure to forces
Autonomous behavior by pilots
High quality simulation
Teamwork
Project History : Project History Phase One
Develop intelligent forces (entity-level)
Focused on Fixed Wing Aircraft domain
U. of Michigan, USC-ISI, CMU
Phase Two
Develop command forces
USC-ISI started Rotary Wing Aircraft domain
Synthetic Theater of War ‘97
Catch the
buzz on authorSTREAM
Copyright © 2002-2008 authorSTREAM. All rights reserved.