1. DrillSim
July 2005

2. DrillSim: a Testbed
DrillSim simulates a crisis response activity (e.g, evacuation).
DrillSim allows testing IT solutions in the context of the simulated response activity.
Translates IT metrics to response metrics.
E.g., Bandwidth usage, time, accuracy to casualties, response time, risks taken.

3. DrillSim: a multi-agent system
One agent simulates one person.
Multiple roles:
Evacuee
Floor warden
Building coordinator
Zoon captain
Etc.
Individual profiles (physical and cognitive abilities)
Relationship between agents -> social networks.
Agents involved in information flow
Collect information via their own sensors (e.g., eyes), or the devices they carry (e.g., PDA)
Analyze information (i.e., assimilate information, make decisions)
Share and disseminate information via their own communication devices (e.g., speech) or the devices they carry (e.g., PDA)

4. DrillSim: a plug-and-play system
Plug models that drive simulation such as
human behavior,
spatio-temporal models (e.g., geography, class schedules)
crisis models,
response activity,
response plans.
Plug IT solutions.
Response activity projected as information flow
Information is
Collected
Analyzed
Shared
Disseminated
IT solutions can be plugged at any of these points

5. DrillSim: more than a simulator
DrillSim can simulate a drill
DrillSim can also be synchronized with an on-going drill:
Simulated drill augments an on-going drill adding
Simulated people
Simulated communications infrastructure
Simulated sensing infrastructure
 Lowers cost of a real drill, adds flexibility
On-going drill augments simulated drill
Real people
Real communications infrastructure
Real sensing infrastructure
 Adds realism, allows for calibration of the simulated entities

6. DrillSim: more than a simulator
DrillSim can simulate a drill
DrillSim can also be synchronized with an on-going drill:
Simulated drill augments an on-going drill adding
Simulated people
Simulated communications infrastructure
Simulated sensing infrastructure
 Lowers cost of a real drill, adds flexibility
On-going drill augments simulated drill
Real people
Real communications infrastructure
Real sensing infrastructure
 Adds realism, allows for calibration of the simulated entities

7. DrillSim Sensing Infrastructure Model DrillSim World Simulated World
Real world
Agent Model
Environment Model
Geography
Model
Observed World
Crisis Model
Calibrator
Initial
Scenario
Generator
Response
Activity
Module
Resources
Communication
Media Model
Scenario Model
People

8. DrillSim Models Models Models to represent spatial temporal data
Geographical space, dynamic changes in the space, traffic, people, etc.
A response plan
Plan of evacuation.
A crisis model
Represent disasters and their effects.
Human Models
Models to represent agents (e.g., information assimilation, decision making, sensing capabilities).

9. Scenario Generator
Scenario is the location of people in the geographical space and their activities at the moment.
The initial scenario generates a scenario based on
Statistics like classes in universities
Pedestrian traffic
Information about meetings, games, etc
Location of response personnel
The scenario is generated by observing the physical world.

10. Response Activity Module
The main module that drives the simulator
Major components
Agents
Log of events
Plays out the activities based on the information in the observed world

11. The 4 worlds Real world: an on-going drill
Simulated world: computer simulated drill
DrillSim world: merging of real world and simulated world.
Observed world: DrillSim world as observed by the agents
DrillSim World
Simulated World
Real world
Observed World

12. Calibrator Calibrator is used to update the
Models used to represent agent behaviour
e.g: floor warden spent 5 min in each room
Decision making of the agent
Helps in Virtual reality, augmented reality integration
Agent behavior can be updated by its human counterpart’s behavior

13. DrillSim Sensing Infrastructure Model DrillSim World Simulated World
Real world
Agent Model
Environment Model
Geography
Model
Observed World
Crisis Model
Calibrator
Initial
Scenario
Generator
Response
Activity
Module
Resources
Communication
Media Model
Scenario Model
People

14. Research Problems Dynamic Data Management Navigation VR/AR Integration
Reasoning/Behavior Modeling
Languages
Data representation
Representing geography (at different scales)
Representing dynamic objects and phenomena
Architecture

15. Dynamic Data Management
There are continuous queries from different sources
The data is highly dynamic
Optimizing querying and updating in such scenario
Dynamic queries over mobile objects
Iosif
Querying imprecise data in moving object environments
Dmitry
Push Pull
Chen Li

16. Navigation
Navigation in the context of highly dynamic environments, and multiple agents
Should be optimal (compute/resource intensive)
Take into consideration real time changes
Representation for such planning algorithms
Modeling human navigation
Path planning for a group of agents
Optimizing location of agents, common goals, etc
flow control
Evacuation planning

17. VR/AR Integration Goal: Interaction between real and virtual worlds
Real to virtual
Immersing people into the simulator
Interface for real people to communicate with simulator
Tracking via sensors
Integrating real sensors
Integrating real communication devices and media
Calibration
Data from drills
Live actors
Virtual to real

18. Data representation Representing geography (at different scales)
Representing dynamic objects and phenomena

19. Reasoning Modeling human behavior Decision making
Event based or taken every time unit?
Humans as information processors?
Social networks
Human communication
Representation of knowledge and awareness

20. Languages Languages Communication between agents and real people
Languages for specifying evacuation procedures and activities.
Integrating IT solutions.

21. Where we are: v0.2 Response activity: evacuation 4th floor Cal(IT)2
Grid representation
A* algorithm + obstacle avoidance
Stochastic decision making based on ANN
Evacuee and floor-warden roles
2D/3D interface
Human can control an agent
Human can communicate with agents
Simplified crisis model (spreading fire)
Response metric: number of people in building vs time

22. Demo

23. Where we are going Campus-wide evacuation Plug-and-play architecture
Scalable architecture
Integration with real world and real activities
Realistic models for crisis, agent behavior, and initial scenario generator.