Autodesk Game Technologies Group SeriGamEx – November 12, 2008 ‐ Civitavecchia

Image courtesy of Bugbear Entertainment Agenda

ƒ Introduction to Autodesk ƒ Game technologies in simulation ƒ The production chain ƒ AI and behaviors ƒ AI data model Agenda

ƒ Introduction to Autodesk Why am I before you? ƒ Game technologies in simulation ƒ The production chain ƒ AI and behaviors ƒ AI data model Autodesk M&E: “Create, Animate, Integrate”

3ds Max

Maya

MotionBuilder

Mudbox

HumanIK

Kynapse

Images Courtesy of Public Titles using Kynapse (Artificial Intelligence) Kynapse leads both Game & Serious Game markets

Some Kynapse references

Game Serious Game Agenda

ƒ Introduction to Autodesk ƒ Game technologies in simulation Game technologies: what for? ƒ The production chain ƒ AI and behaviors ƒ AI data model History: a shift of technological power over 40 years

ƒ The seventies: military or industrial simulations inspired video games ƒ Flight trainers ƒ War games ƒ The nineties: increase in CPU and graphic cards power, and reduction of computer cost ƒ Booming video game market: turns into a mass market ƒ Technology and creativity race between studios ƒ 2000…: technology “maturity” ƒ Consoles: more action, faster CPU, faster visualisation ƒ (animation, physics, AI): packaged technology ƒ Internet: Massively Multiplayer On‐line Games Middleware segments for AAA games

Specialized Infrastructure Commodities Modules

Animation Rendering Sound

AI Interface Techologies Physics Network

ƒ Ageia (physics) ƒ Epic Games ƒ Audiokinetic (sound) ƒ (physics & Anim.) ƒ Emergent ƒ Fmod (sound) ƒ Kynogon (AI) ƒ Trinigy ƒ SpeedTree (landscape)

Players ƒ Natural Motion (Anim.) ƒ Crytek ƒ FaceFX ƒ Autodesk (Anim.) ƒ Instinct ƒ GameSpy ƒ Simutronics ƒ VoiceIn ƒ BigWorld ƒ Bink ƒ Etc. Game technologies are for immersive «first person» simulators ƒ Driving simulators for armoured vehicles ƒ Instruction, training, operational preparedness of crew ƒ Tank leader or Squad leader training Environmentofchoice: ƒ Convoy training cave or visualization dome ƒ Flight simulator: helicopters ƒ Red forces simulation ƒ Squad members simulation ƒ Flight simulator: plane ƒ Civilian white noise, and military ground traffic simulation for target acquisition and selection ƒ Use of force and judgment training ƒ Shooting simulator for infantry ƒ Instruction to procedure

ƒ Marksmanship JTAC AFRL MetaVR ƒ Use of force training But these technologies can be used to improve other simulations

ƒ Open constructive simulations The key triggers: ƒ Operator assistance • Visual realism is a must for ƒ Spatial reasoning simulation acceptance ƒ Immersive static simulators • Operator’s capabilities, ƒ Check point productivity or efficiency can ƒ Peace keeping operations be improved ƒ Weapon system simulation (SBA) ƒ UAV, ROVs ƒ Pay load study ƒ Operator training ƒ Sensor stimulation ƒ Doctrine preparation and validation simulation ƒ New doctrine conception, entity level ƒ Section head training ƒ Judgment training for urban guerilla operations ƒ Operational training Rheinmetall ƒ Flight deck of plane carrier Agenda

ƒ Introduction to Autodesk ƒ Game technologies in simulation What components ƒ The production chain of simulation are concerned? ƒ AI and behaviors ƒ AI data model Simulation and video game production are similar in many respects… with some differences

HLA, DIS, TENA distribution Procedural Data EO/IR Weather AI Data Data Imagery model model 3D Terrain Model Elevation data Behavior Rendering Others: Engine Engine sensors Behaviors Physics Commu- RT 3D Vector data engine nications Scene Level/Game Runtime 2D plans Scenario/ 3D Building Engine Model Simulation CGF 3D Models Editor

3D Models 3D Models (Entities Operator Player Vehicles Animations Weapons) Movement models Physics Data AAR Simulation logging DCCbuilder Runtime Kynapse provides 3 components of Real Time 3D simulation

HLA, DIS, TENA distribution Procedural Data EO/IR Weather AI Data Data Imagery model model 3D Terrain Model Elevation data Behavior Rendering Others: Engine Engine sensors Behaviors Physics Commu- RT 3D Vector data engine nications Scene Level/Game Runtime 2D plans Scenario/ 3D Building Engine Model Simulation CGF 3D Models Editor

3D Models 3D Models (Entities Operator Player Vehicles Animations Weapons) Movement models Physics Data AAR Simulation logging DCCbuilder Runtime Agenda

ƒ Introduction to Autodesk ƒ Game technologies in simulation ƒ The production chain One in three steps in the ƒ AI and behaviors real time simulation loop ƒ AI data model AI brings an answer to 2 questions: why and how

Human Factors: learning, adaptability, evolution, emotions High Level Why? Decision modules: constrained optimization, multi-objective optimization infinite state machines, bayesian networks, Petri networks, rule based systems

Scenario

Coordinated movements, complex movements Basic movements Low Level How? Spatial reasoning Path planning, path following, dynamic avoidance

Based on a specific data model AI computes “next frame” behaviours

ƒ Real time loop ƒ Non real time

Input: Services • all entities positions and • Path finding behavior parameters • Spatial analysis • all behavior relevant simulation events (ie: Data changes in world) • Graphs Compute: • Meshes behaviors for all AI controlled • Others entities (runs all entities brains) Behaviours Output: High level decision model • All AI driven entities wanted speed and orientation • Animation related entities parameters Simulation is much more demanding on AI than regular games

Game Simulation ƒ Designed for fun ƒ Designed for realism ƒ Most behaviors are short (30s) ƒ Behaviors may last hours ƒ Scripted behaviors to serve game ƒ Fully autonomous individual play behaviors ƒ Nothing unexpected that could ƒ The unexpected should occur like in harm game play real life ƒ Dozens of entities at 60 herz ƒ Hundreds of entities at 10 herz ƒ Destroy everything for fun ƒ Destroy for a purpose ƒ No need for determinism ƒ Determinism and AAR ƒ Scripts written by programmers ƒ Exercises designed by operators ƒ Dozens of scenes, with dozens of ƒ Hundreds of scenarios with unique behaviors thousands of behaviors Proper character centric solution also requires high performance physics

ƒ High performance physics for human entities ƒ >10 hertz clamping of hundreds of entities ƒ Full body contact for dense crowds ƒ “the real performance bottleneck” ƒ Destructibility ƒ 2 states (no physics) ƒ Fracture ƒ Full destructibility Demand on animation depends on simulation purpose and constrains

ƒ High fidelity animation ƒ Side steps, backward walk ƒ Proper foot clamping ƒ Proper blending ƒ Facial expressions ƒ Full body inverse kinematics ƒ High performance ƒ Level of detail ƒ Retargetting

(HumanIK) Agenda

ƒ Introduction to Autodesk ƒ Game technologies in simulation ƒ The production chain ƒ AI and behaviors A rich robust and light ƒ AI data model data model is a necessity A basic data model, with AI mesh and Path Data

1 – boundaries 2 – connectivity network graph Computed through Computed through detailed exploration exploration of of physical world accessible world (collision world)

Topological model, AI Light, robust, Mesh and Path Data complete AI data are optimised for maximum efficiency at runtime: CPU consumption and memory footprint

B ƒ Path planning uses the Path Data ƒ Light representation of the world ƒ A* algorithm ƒ Euclidian heuristics ƒ Paths under constrain, non‐euclidian A heuristics

ƒ The followed trajectory is computed based on the graph and the mesh ƒ Final destinations on graph ƒ Next point to reach on graph ƒ « Light » accessibility test on mesh ƒ «Forward»test to look for more appropriate target point Spatial reasoning explores the graph

ƒ Reasoning start on a given point of the graph. ƒ Algorithm selects neighbourgh points on the graph. ƒ Each points is checked for validity according to model’s criteria (hide, flee, other…) ƒ Algorithm propagates on the graph until one or several valid candidates are found. ƒ Selected points are proposed to the decision layer. The model incorporates elements affecting movement capabilities or perception

«Topological objects » are objects Window that may affect path planning, path following or spatial reasoning

ƒ Doors ƒ Windows ƒ Ladders ƒ Elevators ƒ Stairs ƒ Escalators ƒ Special corridors (narrow) ƒ Manholes ƒ Woods border Door ƒ … At runtime, impact of such objects on path finding and spatial reasoning depends on their state or status Recommendations for a richer model

ƒ Identifying doors, windows, ladders, elevators, stairwells… ƒ Conditional path edges can be pre‐processed during generation for future use at runtime ƒ Adding pertinent semantic information on ground nature ƒ Automatic data model generation can then intelligently filter data according to entities properties, and decorate the model for future use at runtime ƒ Requires subject matter competencies, and a good understanding of the objectives and use of the simulation ƒ Examples: soil type, room function… ƒ Identifying and segregating fundamentally static or potentially dynamic elements of the simulation ƒ Automatic generation will not consider dynamic objects in the fixed data model, as they will be handled at runtime only ƒ Example: tables, chairs, doors, desks or sofas in a building ƒ Pre‐process elements whose topological properties require advanced subject matter competency: ƒ Custom pre‐processing of such elements can then be included in automatic generation ƒ Example: crest line Towards a richer object model in DCC

« Identifying early the « topological objects » facilitates automatic pre‐processing during model generation »

ƒ All objects in the simulation can be created with all properties: ƒ Visual ƒ Physical ƒ Animation ƒ AI ƒ These properties should be kept intact throughout the production process ƒ “two ways production pipeline” Seamlesss production pipeline, from modeler to simulation engine

HLA, DIS, TENA distribution Procedural Data EO/IR Weather AI Data Data Imagery model model 3D Terrain Model Elevation data Behavior Rendering Others: Engine Engine sensors Behaviors Physics Commu- RT 3D Vector data engine nications Scene Level/Game Runtime 2D plans Scenario/ 3D Building Engine Model Simulation CGF 3D Models Editor

3D Models 3D Models (Entities Operator Player Vehicles Animations Weapons) Movement models Physics Data AAR Simulation logging DCCbuilder Runtime Thank you for your attention