Project Albert M&S Review 13 November 2002 Gary Horne Sarah Johnson 1 Project Albert Summary • Chartered by Congress with the vision to address questions of military decision-makers not supported by traditional methods • Leverages High Performance Computing in innovative ways • Stresses interdisciplinary, joint, and coalition collaboration • Supports rapid investigation of a wide range of alternatives through distillation models and data farming methods • Applications include efforts in the areas of surf zone obstacles, Command and Control, and enhanced blast weapons 2 Distillations • • • • Distillation Models – Agent-based model that is intuitive, transparent, transportable, and farmable. Agents – autonomous, reactive, motivated, adaptive, mobile, proactive, and they communicate. A bottom up distillation of the essence of a question Quick implementation - less than a few hours 3 What Are Distillations? • A distillation is a bottom-up abstraction of the essence of a situation. • Characteristics of a distillation? Can handle nonlinearities, binary events, sensitivity to initial conditions, emergence Adaptable to massively parallel machines Can create data needed for analysis Shows promise to capture intangibles and model coevolving landscapes Good for modeling complex adaptive systems And enhances our intuition 4 Which Distillations Does Project Albert Use? – ISAAC – Mana – Socrates – Pythagoras 5 ISAAC 6 ISAAC Interfaces SELECT RUN OPTION [1] Run ISAAC engine with new input [2] Playback old run [3] Quit ?1 SPECIFY FORM OF INPUT [1] Prompt from screen [2] Read from datafile ?2 File name ?ISAAC.DAT SPECIFY FORM OF OUTPUT [1] Terminal [2] File [3] Both ?3 NUMBER OF ITERATIONS TO STORE IN OUTPUT FILE ?20 7 ISAAC Features • Physical Characteristics – Movement range, sensor range, fire range, communications range • Personality Traits – Internal value system which weights movement toward or away from same/opposing goal, alive/injured friendly/enemy agents as applied to all other info used by agent to choose a move – Used to rank set of possible moves an agent can choose to take during current time step • Battlefield – Two-dimensional lattice of discrete sites • Agent States – Alive, injured, dead 8 ISAAC Features • Movement – Next move based on best matching the agent’s personality • Group Sociological Factors – Constraints such as Advance, Cluster, Combat included • Advance: toward enemy goal if #of friendly agents exceeds defined threshold • Cluster: stop moving toward surrounding friendly agents if # exceeds defined threshold • Combat: move toward enemy if difference between friendly/enemy agents exceeds defined threshold • Scenario Generation – Text file – On-screen manipulation • Data Farming – Web Interface to MHPCC 9 Mana 10 Mana Interfaces 11 Mana Features • Originated by the Defence Technology Agency (DTA), New Zealand – • Event-Driven Personalities – – • Squad-based memory of enemy contacts Terrain Map – • Speed, sensor range, fire range, stealth, max targets to engage - defined for each personality state Situational Awareness – • Triggering events such as squad shot at, squat taken shot, reached way point which cause a change in personality weight vector for a specified length of time before reverting back to the default personality or until another event triggers another personality type Personality traits defined for each personality state • Internal value system which weights movement toward or away from waypoints, alive/injured friendly/enemy agents as applied to all other info used by agent to choose a move Physical – • Parameters/personality traits/rules defined similarly to the ISAAC distillation, if not the same for some variables Roads that agents can follow, impassable terrain objects Way Points – Establish set of points to follow (interim goals), not just an ultimate goal 12 Mana Features • Group Sociological Factors – – • Battlespace – – • Situational awareness map - agents communicate enemy positions to other squads outside the local area • HQ squad sends info in their collective sensor range to the SA map where other squads of the same allegiance can view the information Edited/created using paint program (create object/obstacle blocks) without specifying coordinates. Agents recognize blocks as visible while program executes • Can also import maps into paint - can draw roads for agents to follow Scenario Generation – – – • Unit Cohesion, Aggression • Advance: toward enemy goal if # of friendly agents exceeds defined threshold • Cluster: stop moving toward surrounding friendly agents if # exceeds defined threshold • Combat: move toward enemy if difference between friendly/enemy agents exceeds defined threshold Defined for each personality state Scenario text description opens upon execution of the model - provides background information of scenario for new users of that file Text file with corresponding map file On-screen manipulation using previously created scenario Data Farming – Gilgamesh Cluster 13 Socrates 14 Socrates Interfaces 15 Socrates Features • • Improvements under development Methodology – • Command hierarchy – • – sensors, weapons (lethal/non-lethal), movement, communications, sentiments, inculcation, accommodation personality defined by list of decisions (based on level of command) Battlefield – • Commander, Leader, Grunt Agents – • Values-driven behaviors Obstacles (impede movement only) Data Farming Support – Inputs/Scenario XML-based 16 Pythagoras 17 Pythagoras Interfaces 18 Pythagoras Features • • Improvements under development Methodology – • Agents – • Soft Decision Rules sensors, weapons (lethal/non-lethal; direct/indirect), movement desires, triggers “Sidedness” – multiple sides possible based upon RGB color scale – sides change dynamically throughout scenario execution – affiliations categorized by unit, friendly, neutral, enemy • Battlefield – – • Base terrain properties of mobility, concealment, protection, height Terrain features with same property categories can overlay base Data Farming Support – Inputs/Scenario XML-based 19 Distillation Model Comparison Decision Logic Agent States Command & Control Behavior Rules • • • • • • ISAAC Mana Socrates Pythagoras Physics-based, hard-coded decision methodology Physics-based, hard-coded decision methodology Values-Driven DecisionMaking Methodology Physics-based, “soft” decision logic Alive, Injured, Dead Event-Triggered Personalities Alive, Dead Event-Triggered Personalities Global Commander Local Commander Agent Personality Weights Physical characteristics Trust, Cohesion, Aggression • Squad • Personality Weighted by Triggering Event • Physical characteristics by Agent Definition • By Squad Triggering Event • Cohesion, Aggression, Allegiance • By Squad Squad Definition • Multiple agent • Multiple Triggered personalities • Multiple physical parameters Personalities • Multiple Physical Parameters • • • • Commander Leader Agent Hierarchical positions determine decisions/tactics • Physical characteristics • Trust, Allegiances • Threshold-defined Leadership • Agent • Personality Weighted by Triggering Event • Physical Characteristics by • • Individual agent defined • Instantiate any number of • • given agents • Leadership hierarchy • defined • • Terrain Physical & Tactical Parameters Development Status Rectangular obstacles impede movement and shot capability, but not LOS • sensor, fire, communications ranges, speed Complete – No new development planned Bitmap with pre-defined terrain-types impede movement, but not LOS • sensor, fire, situational awareness ranges, speed Movement obstacles impede movement, but not LOS nor shot capability • sensor, weapons, • Triggering Event Cohesion, Aggression, Sidedness Individual agent defined Instantiate any number of given agents Group personalities based on triggers Can be defined by “Sidedness” Multiple physical parameters Terrain Editor with user definable polygons provide protection, concealment, trafficability sensor, fire, speed communication channels/ranges, speed Versions released; Versions released; Beta released; Under development Under development Under development Distillation Models Status ISAAC Mana • Development completed • First distillation developed for Marine Corps • Batch processing/Data Farming mode is ready • Newer versions under development by and in use; Web interface available • Limited set of questions/scenarios to explore • Continued use is includes scenario setup, execution, and translation into other models/validation against other models Defence Technology Agency, New Zealand to expand features, parameters, MOEs, batch processing • Batch processing/Data Farming mode is available on Gilgamesh Cluster; in development for MHPCC; Requires email submission Socrates Pythagoras • Developed by L-3 Communications Analytics • Developed by TRW under contract with under contract with Project Albert • Data farming sessions available at MHPCC; Web interface available • Prioritization and development of additional features, visualization tools for result analysis, user interfaces for scenario creation and initial scenario testing is in progress Project Albert • Version 1 delivered • Data farming sessions available at MHPCC; Web interface available • Prioritization and development of additional features in progress