Agent-based Simulation Platform Evaluation in the Context of Human Behavior Modeling

1
Agent-based Simulation Platform Evaluation in the
Context of Human Behavior Modeling

• Michal Laclavík, Štefan Dlugolinský, Martin
  Šeleng, Marcel Kvassay,
• Bernhard Schneider, Holger Bracker,
• Michal Wrzeszcz, Jacek Kitowski, Ladislav Hluchý

2
IKT Group - Institute of Informatics SAS
URL http//ikt.ui.sav.sk

• Dept. of Parallel and Distributed Computing
• Research and Development Areas
• Large-scale HPCN and Grid applications
• Intelligent and Knowledge oriented Technologies
• Experience from European projects
• 6 project in FP6 EGEE II, K-Wf Grid, DEGREE
  (coordinator), EGEE, int.eu.grid, MEDIGRID
• 4 projects in FP7 Commius, Admire, EGEE III,
  Secricom
• 1 EDA project EUSAS
• Several National Projects (SPVV, VEGA, APVT)
• IKT Group Focus
• Multi-Agent Systems
• Information Processing
• Semantic Web
• Knowledge oriented Technologies
• Parallel and Distributed Information Processing
• Solutions
• AgentOWL semantic web and FIPA agents
• Ontea Pattern-based Semantic Annotation
• ACoMA KM tool in Email

Director leader of PDC Dr. Dipl. Ing.
Ladislav Hluchý
3

• EDA RT JIP FP project (for RT Joint Investment
  Programme on Force Protection)
• European Urban Simulation
• for Asymmetric Scenarios

4
EUSAS Objectives
results
qualitative evaluation
Serious Game
updates
interactions
Training
Modelling
Rules Of Engagement
updates
results
models
Data Farming
updates
interactions
quantitative evaluation
Analysis
results
5
ABS Evaluation Approach

• Survey Literature
• List of available simulation platforms is in
  Deliverable Appendix
• Existing ABS evaluations were considered
• Evaluation Criteria/Features
• 12 features selected list on the next slide
• Principle Evaluation through implementation
• Exemplary Human Behavior scenario defined
• Civilians getting angry (throwing stones) or
  afraid (running to safety).
• Soldiers arresting Civilians if hit by stone
• Implemented in MASON, NetLogo and VBS2

6
Survey Literature

• Stupid Agent Model
• 16 features
• S.F. Railsback, S.L. Lytinen and S.K. Jackson
  Agent Based Simulation Platforms Review and
  Development Recommendations Simulation 89 (2005)
  
• NetLogo, MASON, Repast, Swarm and Java Swarm
• Later also others, like EcoLab
• Human Behavior Modeling
• We have chose 12 features
• Generic, but evaluated on Human Behavior Scenario

7
12 Features selected as evaluation criteria

1. Loading and Representing the Environment and the
   Scenario
2. Creating and Representing Agents
3. Behavior Implementation
4. Movement Implementation
5. Visualization
6. Parameterization
7. Model check-pointing
8. Analytical Tools
9. Logging
10. Performance
11. Standards
12. Development Environment

8
Exemplary Human Behavior Scenario

• Soldiers
• Catching civilians if hit twice by stone
• If civilian is caught (arrested), civilian will
  disappear
• Soldiers are robotic (no emotions)
• Civilians
• Driven by 2 emotions (fear and anger)
• When angry, trying to find stone, going to
  soldier and throws the stone
• When afraid, flying to safety area (yellow)

9
NETLOGO 4.1

• Advantages
• NetLogo can be invoked and controlled by another
  program running on the JVM by Controlling
  facility API (e.g. app which automates series of
  model runs, embed NetLogo models in a larger app)
• Simulation state (i.e. world) can be saved to a
  CSV file and later loaded
• Java API for creating custom extensions to
  NetLogo (commands, reporters)
• Models can be run without visualization
• Ability to load vector GIS data (points, lines,
  and polygons - ESRI shapefiles), and raster GIS
  data (grids) into NetLogo by GIS extension
• Easy to draw graphs, create simulation parameter
  controllers (sliders, buttons, etc.)
• Many useful tools like BehaviorSpace, System
  Dynamics Modeler, HubNet, Logging

10
NETLOGO 4.1

• Disadvantages
• Slower than Mason, some parts of user code are
  interpreted at runtime
• The Controlling facility API is considered
  experimental. It is likely to continue to change
  and grow.
• Support for creating 3D worlds is still in an
  experimental state. Only 2D world is fully
  supported.

11
MASON

• Advantages
• Fast, overhead of simulation environment is
  minimal
• Java
• Models are completely independent from
  visualization
• Models may be checkpointed and recovered
• Agents are not forced to have a physical
  location, which is good if we want to create
  agents representing groups (meso and macro
  levels)
• Physical environment any number of 2D or 3D
  layers
• Multiple Displays of simulation
• Time series Graphs, variable inspectors
• GIS data can be loaded
• Disadvantages
• NetLogo has better support for movement and
  analysis of distances, objects etc. in physical
  environment this impacts development speed, but
  gives flexibility

12
VBS2

• Integration Challenges
• VBS is thread- and event-based. Our candidate
  ABS systems (MASON and NetLogo) are step-based.
  Integration is not straightforward but feasible.
• Changing the action in the middle of its
  execution may cause a jerkinganimation. For
  example while throwing a stone the agent
  decides torun to the safety area
• Movement in VBS may be executed a bit differently
  from what was planned and simulated in ABS we
  need to use waypoints to minimize the discrepancy.

13
Evaluated Features

• Loading and Representing the Environment and the
  Scenario
• MASON
• 2D, 3D, layered IntGrid2D, Continuos2D
• GIS support tested
• NetLogo
• 2D two-dimensional grid of patches, 3D
  experimental
• easy import from bitmap
• GIS support tested
• Creating and Representing Agents
• Soldier, Civilian, Stone
• MASON
• Represented by Java class (Steppable interface),
  step(SimState state) method
• access to environment SimState state
• NetLogo
• Turtles (dynamic), patches, links and the observer

14
Evaluated Features

• Behavior Implementation
• NetLogo
• turtle variable and the RUN command
• MASON
• step(SimState state)
• Movement Implementation
• NetLogo
• Direction and step
• MASON
• Go to X, Y
• Flocking, steering implemented in demo
• Visualization
• MASON strong separation of Model and
  Visualization
• NetLogo possibility to switch off visualization,
  speed does not change.

15
Evaluated Features

• Parameterization
• supported
• Model check-pointing
• Supported in both
• MASON platform independent
• Analytical Tools
• MASON improvement over the years
• Property inspectors
• Video, snapshot, streaming, charts
• NetLogo
• Property inspectors
• snapshot, streaming, charts
• Logging
• MASON log4j can be used
• NetLogo using log4j integration

16
Evaluated Features Performance
NetLogo
Mason
17
Evaluated Features

• Standards
• Agent Standards FIPA not relevant for
  simulation agents
• DIS and HLA standards
• Relevant but we did not test
• VBS2 will be integrated for training
• we plan to use the plug-in functionality in VBS2
  and CORBA technology
• need to create a FOM - Federation Object Model
• Java based HLA
• poRTIco
• Java port of CERTI
• Development Environment
• Both step based, easy debug, better then thread
  based MAS
• NetLogo
• NetLogo IDE, debugging using variable inspectors
• MASON
• Any Java IDE, standard Java debbuging
• We have used Eclipse

18
Conclusion

• Both are almost equal in many features
• NetLogo better in physical movement support and
  some analytical tools.
• MASON much faster, supports strong separation of
  visualization and behavior models, better support
  for 3D environment, Java based - easier to
  integrate with other systems.