Integrating a Distributed Agent Based Simulation into an

Integrating a Distributed Agent. Based Simulation into an HLA Federation Gary Kratkiewicz Amelia Fedyk Daniel Cerys

Purpose of this Session • Understand how to: – Use distributed multi-agent systems, with a specific emphasis on the Cougaar architecture – Integrate a distributed multi-agent simulation into an HLA federation – Design a full-scale Cougaar society for integration with an HLA federation

Outline • • • Motivation and Goals Distributed Agent Computing and Cougaar Integration Approach Lessons Learned Design considerations for Cougaar societies Conclusion

Motivation • Military combat simulations lack sophisticated logistics components • Logistics support is critical to combat operations • Cougaar-based logistics simulations provide capabilities not available in current military simulation systems • Roadblock: lack of interoperability

Purpose • Demonstrate the ability of Cougaar societies to interact with other simulations via HLA • This would allow the following: – Credible proposals of simulation systems based on Cougaar-HLA linkages – The addition of realistic simulation of logistics to combat simulations

Technical Goals (1) • Demonstrate that a Cougaar-based society of agents can act as an HLA federate • Show that Cougaar time mechanisms can integrate with HLA time synchronization • Confirm that a Cougaar society can work with a variety of RTIs: – HLA 1. 3 and HLA 1516 – Fully asynchronous and partially asynchronous – Java and non-Java based

Technical Goals (2) • Develop techniques for creating or modifying Cougaar societies to be HLA federates • Identify – Further areas of research – Additional functionality required in a full-scale Cougaar HLA federate

Outline • • • Motivation and Goals Distributed Agent Computing and Cougaar Integration Approach Lessons Learned Design considerations for Cougaar societies Conclusion

Agents • Agents are independent software entities that react to events and initiate actions by themselves • Different agent-based architectures yield agents with varying capability levels • Agents can have or define roles, tasks, beliefs, desires, or intentions • Agents can be static or mobile • Some agents or agent systems are considered intelligent

Distributed Agent Computing • A distributed system is a collection of separate processes or information systems that can act together as a single system • Distributed agent computing involves agentbased systems that operate in a distributed manner • Purpose: – Implement complex behavior – Model or simulate complex systems

Cougaar • Cognitive Agent Architecture • Java-based architecture • Rich design can model diverse objects and interactions • Agents are Cougaar components with a defined functionality and local memory store (Blackboard) • The behavior of the agent emerges from the composite of plugins

Cougaar • Blackboard(BB) implements a publish/ subscribe API • Plugins view objects on the BB by creating subscriptions • Logic Providers are lightweight agent components responsible for messaging and BB modifications • A collection of agents make up a cougar society

Cougaar Plugin, Agent, and Node Structure of the Cougaar Architecture (K. Kleinmann et al, 2003)

Ultra. Log • 4 year project sponsored by DARPA • Layer built on top of Cougaar Architecture • Developed to model military logistics within a distributed multi-agent system • Added Security, Robustness and Adaptivity to the cougar infrastructure. • Test society models interaction between large set of military organizations.

Ultra. Log • Society runs in strictly planning mode(predictive solution) and execution mode(simulated solution) • Solution includes detailed transportation and supply chain plan • Society capable of garbage collection of completed/stale BB objects

FCS Supportability • Future Combat Systems (FCS) Supportability extends Ultralog Functionality • Agents designed to reduce the logistics footprint • Additional business logic added to plugins to model highly mobile, self-sufficient units. • Plugins enhanced to accept 3 rd party simulation data

Outline • • • Motivation and Goals Distributed Agent Computing and Cougaar Integration Approach Lessons Learned Design considerations for Cougaar societies Conclusion

HLA Overview Simulations Data Collectors Interfaces to Live Players to Interfaces Live Players Interface Runtime Infrastructure (RTI)

Federation and Society Design • Created small Ultra. Log logistics society – Pared down to model only bulk fuel consumption – 2 fuel-consuming organizations (agents) – 18 other organizations in supply chain and chain of command • Two society federates based on above: – High-fidelity logistics society for general simulation – Demand generation society to model fuel demand

Prototype Society

Non-Cougaar Federate • Performed verification with Java-based, non -agent, non-Cougaar federate • Demand generation society to model fuel demand

RTI Selection • Tested prototypes with following HLA RTIs: – Pitch 1516 LE – Pitch 1. 3 LE – DMSO NG 1. 3 • Demonstrated the feasibility of using Cougaar with: – HLA 1. 3 and IEEE 1516 RTIs – Fully asynchronous (Pitch) and partially asynchronous (DMSO) RTIs – Java-based (Pitch) and non-Java (DMSO) RTIs

Simulation Models • Cougaar time: – Continuously advances with real (wall clock) time – Can be advanced ahead to some future time • Mapping to simulation model: – Advance time with real time or ahead in equal steps time-stepped simulation model – Advance time in unequal steps event-driven model • Most Cougaar simulations run for hours and simulate operations over months – Equal steps scaled real-time – Unequal steps non-real-time

Time Management • Cougaar time – Initialized to physical time – Advances at same rate as wall clock – Can be stepped to future point in time – Society should be in quiescence before stepping • Cougaar and HLA time synchronization – None: not useful for simulation, OK for planning – Conservative: appropriate – Optimistic: not appropriate; Cougaar cannot roll back time • OK since different federates can use different synchronization methods

Iterative Development Approach 1. Build a federation without HLA time management • • Two federates Allowed us to get running quickly 2. Build a federation with HLA time management • • Two federates Added RTI time synchronization 3. Build a federation with a multi-node Cougaar society • • Three federates total Two federates map to one society

Phase 2: HLA Time Management High - Fidelity Logistics Society Cougaar Node Demand Generation Society Cougaar Node Advance NCA Time 1 - 35 - ARBN 1 - 6 - INFBN Ambassador Service Federate 1 Federate 2 Time Management HLA RTI Supply Task Interaction

Phase 3: Multi-Node Society High - Fidelity Logistics Cougaar Node Advance Society Cougaar Node Demand Generation Society Cougaar Node Advance NCA Time NCA 1 - 35 - ARBN 1 - 6 - INFBN Ambassador Service Federate 1 1 - 35 - ARBN 1 - 6 - INFBN Ambassador Service Federate 2 Time Management HLA RTI Time Supply Task Interaction Service Federate 3

Results of Integration Experiments • Demonstrated interoperability between Cougaar and HLA RTIs in various combinations • Demonstrated various operations: – Subscribing a Cougaar federate and plugins to specific HLA interactions – Interfacing a Cougaar society to the RTI via a single ambassador class – Interfacing a Cougaar society to the RTI via a Cougaar service – Synchronizing society time via the HLA mechanism using conservative synchronization – Detecting tasks in one society, transferring the task info via HLA interactions, reconstituting the tasks in a second society • Learned techniques and identified design considerations

Outline • • • Motivation and Goals Distributed Agent Computing and Cougaar Integration Approach Lessons Learned Design considerations for Cougaar societies Conclusion

Interfacing Between Society and RTI • Singleton ambassador class – Separate ambassador class per RTI – Performs following actions: • • Instantiates the RTI’s ambassador class Joins the federation execution (creating it first if necessary) Connects to interactions and objects Handles the interactions and object attributes from the federation • Distributes them to the appropriate agent in the society • Later moved ambassador class to Cougaar service • Separate class instantiated per Cougaar node

Time Synchronization: Phase 1 • Used a servlet-based, user-directed time advancement mechanism – Allowed us to operate with a variety of simulation models – Used for all phases • Initially implemented without HLA time management – For development purposes only – Advanced time in high-fidelity society – Time advance passed to demand generation society via interactions • Small society – always ready

Time Synchronization: Phase 2 • Conservative time synchronization – Standard HLA mechanism – Modified version of the user-directed Cougaar time advancement mechanism. • Cougaar mechanism modified – Obtains permission from the RTI before advancing time in the society – Required ambassador class, adding plugins to handle time, and modifying the time advance servlet

Time Management Details 1. User manually advances time via servlet 1. Must manually determine quiescence 2. Servlet places time change request object on agent’s blackboard 3. Time plugin subscribes to object on blackboard 1. Reads it when it appears and calls ambassador service 4. Service checks for pending request; if not, sends request to RTI 5. RTI grants advance when appropriate 6. Service calls callback method in plugin 1. Changes society time 2. Stores new federation time 7. RTI delivers appropriate messages to society

Mapping Agents, Actions, and Objects • Cougaar societies contains many objects – Map these objects to HLA actions and objects carefully to ensure acceptable performance – Data can be transferred in HLA via interactions (events) and/or object attributes • Mapping is heavily dependent on the society design – Cougaar agents or asset objects map to HLA objects – Cougaar events map to HLA interactions – Cougaar objects such as Ultra. Log tasks could be mapped either way • Depends on how long they can live and whether their contents change

Outline • • • Motivation and Goals Distributed Agent Computing and Cougaar Integration Approach Lessons Learned Design considerations for Cougaar societies Conclusion

Federate Granularity • Federate granularity: the level of mapping of federates to portions of a Cougaar society – Consider size and organization of the society, and design of other (non-Cougaar) federates • In our demonstration prototypes, each Cougaar node mapped to a federate. • In a large complex society, a logical group of agents (covering multiple nodes) might constitute a single federate – Transportation organization, combat organization, etc. • Single agents could map to individual federates – Make sure this makes sense

Society Design • Selection of agents when federation contains multiple societies • Functional split – Same agents, different functions • Organizational split – Different agents in logical groupings

Time Synchronization • Need automated time advance mechanism – Trigger time advance requests automatically • Based on schedule or events – Check for quiescence before making request • Society time is always advancing – Time step needs to be large enough that society time advances an insignificant amount between steps

Adaptive Society Design • Prototype HLA ambassador class hardcodes HLA interactions, objects, attributes – Need to modify class when plugins change • Better way: hardcode nothing in the ambassador class – Each plugin would • Register with the ambassador class • Tell it which interactions and objects it was interested in – The ambassador class would then • Dynamically subscribe to the requested interactions • Read the FOM file to get the attributes for each interaction • Dynamically get the attribute handles • Leverage Cougaar object discovery to match up objects in federates with those in the society

Operation Independent of Federation • For more flexibility and survivability, design your society to operate – Whether or not the appropriate federates are present – If not available, use default plugins • Or, use the Cougaar Message Transport Service (MTS) and create an HLA-specific link protocol – Appropriate where a society • May or may not be a member of a federation • Has a number of other communication channels at its disposal – MTS would select communication method based on availability and prioritization of channels – Allows interoperation with simulations that exist outside of both the Cougaar society and the HLA federation

Outline • • • Motivation and Goals Distributed Agent Computing and Cougaar Integration Approach Lessons Learned Design considerations for Cougaar societies Conclusion

Conclusion (1) • Successfully Integrated a Cougaar-based agent society as a federate in V 1. 3 and 1516 HLA federations • Integration leverages the benefits of both architectures – Distributed agent-based architecture – Standard simulation interoperability architecture • Established how to – Interface the society and the HLA RTI – Synchronize society time with HLA RTI time – Map agents, objects, and actions in the society to HLA objects and interactions

Conclusion (2) • Examined integrating a full-size Cougaar society with an HLA federation • Lessons learned can be applied to largescale efforts – Such as integrating large Cougaar-based logistics simulations with combat simulations – Such integrated simulations would provide greater effectiveness than combat-only simulations

Integrating a Distributed Agent. Based Simulation into an HLA Federation Gary Kratkiewicz Amelia Fedyk Daniel Cerys