DARPA Simulations and Games

1
DARPA - Simulations and Games

• Discussions with Bob Parker
• 31JAN05

2
Assumptions

• DARPA fully committed to this action
• Funding in the offing is O 100M
• ISI goal is to positively impact creation of
  structure of nascent program
• DARPA leadership is
• H/W oriented
• Intrigued by games
• Under fire due to publicity disasters
• Committed to certain approaches

3
Goals for DoD

• Use computer simulations and games to assist in
  the defense of the realm
• Create, enable and optimize
• Realistic environmental play space
• Effective models
• Existing entities (vehicles, sensors, )
• Proposed new capabilities and weapons
• Useful and illuminating user interface
• Reliable play-ahead capability
• Insightful analytical tools and viz-aids

4
Goals for DARPA

• Effectively break out from constraints imposed
  by operational exigencies
• Provide research where engineering has failed to
  keep pace with needs
• Implement new technology in sims such as limited
  self-awareness and learning
• Produce highly visible warfighter support
  effort
• Regain/retain support from Congress

5
Traditional Uses of Simulation

• Training individual/crew training, mission
  rehearsal, etc. so prevalent in minds of many as
  to preliminarily exclude the rest
• Analysis - considering courses of action on the
  strategic and tactical level
• Evaluation - comparing the prospective worth of
  various configurations of weapons, equipment and
  systems

6
Accepted Sim Terminology

• Live - use of humans to act out various roles in
  scenarios developed by the exercise designers,
  e.g tanks at Ft Irwin
• Virtual - use of operators controlling computer
  simulated vehicles, e.g. pilots in flight
  simulator
• Constructive - use of computer generated and
  controlled entities and units to simulate combat,
  e.g. JUO clutter

7
Where we really are today
Pointing to a terrain map of the town
constructed from cut-up cardboard meal trays
laid out on the floor, a maneuver company
commander showed how his Bradley fighting
vehicles and M113 troop carriers would remain
south of town, ready for Hickeys order to move
in. Then they would establish an inner security
ring while infantry loaded munitions from the
mosque into trucks. An intelligence officer
marked on the map where roadside bombs, snipers
and small arms fire might be expected.
8
Where we could be today

• JUO ran in October using a number of federates
  made from existing simulations
• JUO fielded SAF entities with behaviors
• 300,000 simple clutter civilians
• 3,000 friendly blue forces
• 300 enemy red forces
• On global terrain (DTED1), with hi-res city
• Ran on 128 nodes of Linux cluster
• Subject Matter Expert (SME) evaluation

9
Can Increase numbers by 10X
2,000,000
Future experiments require orders of magnitude
larger more complex battlespaces
MUST do SCALE and FIDELITY!
Number of JSAF Entities
1,000,000
SPP Proof of Principle DARPA / Caltech
320,000
107,000
50,000
3,600
12,000
SAF Express (1997)
UE 98-1 (1997)
J9901 (1999)
JSAF/SPP Goal (2006)
AO-00 (2000)
JSAF/SPP UrbanResolve (2004)
JSAF/SPP Tests (2004)
10
JUO Environment Images
11
JUO Simulation Images
12
JUO Network

• Deployed, spring 04
• 2 Linux Clusters
• 256 Nodes
• 512 Intel Processors
• 4GB RAM-40GB Disk per node
• 24x7support by HPCMP
• DREN Connectivity
• Users in VA and CA
• Could have users anywhere in world

13
Where we need to be tomorrow

• Standing Simulation/Gaming Environment
• Self-aware(limited), learning entities/units
• Higher fidelity simulations (see next chart)
• Distributed and scalable access
• Reach-back to high-end computers
• Powerful analytical tools and capabilities
• Real-time input of situational data
• Management and visualization of massive data
  streams and archives

14
Improvement Needed in ALL Three Dimensions
e.g. lt1 Meter Res.Global Terrain
Resolution
Quantity of Entities
e.g. gt10 Million SAF Entities
Sophistication of Behaviors
e.g. Human Emotions
15
Automate Reduce Set-up Time
e.g. Currently Can Take Months
Terrain DB Creation
New Entity Models
e.g. Requires Experienced Programmer
e.g. Most Lay-down and Direction Generated
Manually
Scenario Generation
16
Major Impediments to Success

• Existing simulations, programs and communities do
  NOT communicate due to H/W, S/W and stove-piping
• Current infatuation with cluster computing has
  enervated needed high-end computational research
• Present analytical techniques are primitive,
  crude, and sub-optimal to the point of being
  derelict

17
We hold these truths to be self evident,

• ISI is only group scaling at these levels
• High-end computing reach-back is a condition
  precedent to success
• Often, improving what you have beats succumbing
  to the allure of doing it new
• While there are things to be learned from
  civilian games, they may be more seductive than
  productive

18
How to spend 100M

• DARPA needs to attack where others are either
  retreating or just holding the line
• Develop and deliver three capabilities
• Universal simulation engine (Rosetta stone)
• Novel high-end computing capabilities
• Leading edge analytical/visualization suite
• Industry/Academic Research is optimal

19
Who MUST be Involved

• ISI
• USC (campus, IMSC, )
• Caltech
• HPCMP
• Naval Postgraduate School
• IDA