IMGD 1001 Programming Practices Artificial Intelligence by Mark

IMGD 1001: Programming Practices; Artificial Intelligence by Mark Claypool (claypool@cs. wpi. edu) Robert W. Lindeman (gogo@wpi. edu) Claypool and Lindeman, WPI, CS and IMGD

Outline o Common Practices o Artificial Intelligence Claypool and Lindeman, WPI, CS and IMGD 2

Common Practices: Version Control o Database containing files and past history of them o Central location for all code o Allows team to work on related files without overwriting each other’s work o History preserved to track down errors o Branching and merging for platform specific parts Claypool and Lindeman, WPI, CS and IMGD 3 Based on Chapter 3. 1, Introduction to Game Development

Common Practices: Quality (1 of 3) o Code reviews – walk through code by other programmer(s) n Formal or informal n "Two pairs of eyes are better than one. " n Value is that the programmer is aware that others will read o Asserts n Force program to crash to help debugging o Ex: Check condition is true at top of code, say pointer not NULL before continuing n Removed during release Claypool and Lindeman, WPI, CS and IMGD 4 Based on Chapter 3. 1, Introduction to Game Development

Common Practices: Quality (2 of 3) o Unit tests n Low level test of part of game o See if physics computations correct n Tough to wait until very end and see if there's a bug n Often automated, computer runs through combinations n Verify before assembling o Acceptance tests n Verify high-level functionality working correctly o See if levels load correctly o Note, above are programming tests (i. e. , code, technical) n Still turned over to testers that track bugs, do gameplay testing Claypool and Lindeman, WPI, CS and IMGD 5 Based on Chapter 3. 1, Introduction to Game Development

Common Practices: Quality (3 of 3) o Bug database n Document & track bugs n Can be from programmers, publishers, customers n Classify by severity and priority n Keeps bugs from falling through cracks n Helps see how game is progressing Claypool and Lindeman, WPI, CS and IMGD 6 Based on Chapter 3. 1, Introduction to Game Development

Common Practices: Pair (or "Peer") Programming o Two programmers at one workstation o One codes and tests, other thinks n Switch after fixed time o Results n Higher-quality code o More bugs found as they happen n More enjoyable, higher morale n Team cohesion n Collective ownership Claypool and Lindeman, WPI, CS and IMGD 7 http: //en. wikipedia. org/wiki/Pair_programming

Group Exercise o Consider game where hero is in a pyramid full of mummies. Mummy – wanders around maze. When hero gets close, can “sense” and moves quicker. When it can see hero, rushes to attack. If wounded, flees. o What “states” can you see? What are the transitions? Can you suggest Game Maker appropriate code? Claypool and Lindeman, WPI, CS and IMGD 8

Outline o Common Practices (done) o Artificial Intelligence (next) Claypool and Lindeman, WPI, CS and IMGD 9

Introduction to AI o Opponents that are challenging, or allies that are helpful n Unit that is credited with acting on own o Human-level intelligence too hard n But under narrow circumstances can do pretty well n Ex: chess and Deep Blue o Artificial Intelligence n Around in CS for some time Claypool and Lindeman, WPI, CS and IMGD 10 Based on Chapter 5. 3, Introduction to Game Development

AI for CS different than AI for Games o Must be smart, but purposely flawed n Lose in a fun, challenging way o No unintended weaknesses n No "golden path" to defeat n Must not look dumb o Must perform in real time (CPU) o Configurable by designers n Not hard coded by programmer o "Amount" and type of AI for game can vary n RTS needs global strategy, FPS needs modeling of individual units at "footstep" level n RTS most demanding: 3 full-time AI programmers n Puzzle, street fighting: 1 part-time AI programmer Claypool and Lindeman, WPI, CS and IMGD 11 Based on Chapter 5. 3, Introduction to Game Development

AI for Games: Mini Outline o Introduction (done) o Agents (next) o Finite State Machines Claypool and Lindeman, WPI, CS and IMGD 12

Game Agents (1 of 3) o Most AI focuses around game agent n Think of agent as NPC, enemy, ally or neutral o Loops through: sense-think-act cycle n Acting is event specific, so talk about sense+think Claypool and Lindeman, WPI, CS and IMGD 13 Based on Chapter 5. 3, Introduction to Game Development

Game Agents (2 of 3) o Sensing n Gather current world state: barriers, opponents, objects n Need limitations: avoid "cheat" of looking at game data n Typically, same constraints as player (vision, hearing range) o Often done simply by distance direction (not computed as per actual vision) n Model communication (data to other agents) and reaction times (can build in delay) Claypool and Lindeman, WPI, CS and IMGD 14

Game Agents (3 of 3) o Thinking n Evaluate information and make a decision n As simple or elaborate as required n Two ways: o Pre-coded expert knowledge, typically handcrafted if-then rules + randomness to make unpredictable o Search algorithm for best (optimal) solution Claypool and Lindeman, WPI, CS and IMGD 15 Based on Chapter 5. 3, Introduction to Game Development

Game Agents: Thinking (1 of 3) o Expert Knowledge n Finite state machines, decision trees, … (FSM most popular, details next) n Appealing since simple, natural, embodies common sense o Ex: if you see enemy weaker than you, attack. If you see enemy stronger, then flee! n Often quite adequate for many AI tasks n Trouble is, often does not scale o Complex situations have many factors o Add more rules o Becomes brittle Claypool and Lindeman, WPI, CS and IMGD 16 Based on Chapter 5. 3, Introduction to Game Development

Game Agents: Thinking (2 of 3) o Search n Look ahead and see what move to do next n Ex: piece on game board, pathfinding (ch 5. 4) o Machine learning n Evaluate past actions, use for future n Techniques show promise, but typically too slow n Need to learn and remember Claypool and Lindeman, WPI, CS and IMGD 17 Based on Chapter 5. 3, Introduction to Game Development

Game Agents: Thinking (3 of 3) o Making agents stupid n Many cases, easy to make agents dominate o Ex: bot always gets head-shot n Dumb down by giving "human" conditions, longer reaction times, make unnecessarily vulnerable o Agent cheating n Ideally, don't have unfair advantage (such as more attributes or more knowledge) n But sometimes might, to make a challenge o Remember, that's the goal, AI lose in challenging way n Best to let player know how agent is doing Claypool and Lindeman, WPI, CS and IMGD 18 Based on Chapter 5. 3, Introduction to Game Development

AI for Games: Mini Outline o Introduction (done) o Agents (done) (next) o Finite State Machines Claypool and Lindeman, WPI, CS and IMGD 19

Finite State Machines (1 of 2) o Abstract model of computation o Formally: n n Set of states A starting state An input vocabulary A transition function that maps inputs and the current state to a next state Claypool and Lindeman, WPI, CS and IMGD 20 Based on Chapter 5. 3, Introduction to Game Development

Finite State Machines (2 of 2) o Most common game AI software pattern n Natural correspondence between states and n n n behaviors Easy to understand Easy to diagram Easy to program Easy to debug Completely general to any problem o Problems n Explosion of states n Often created with ad-hoc structure Claypool and Lindeman, WPI, CS and IMGD 21 Based on Chapter 5. 3, Introduction to Game Development

Finite-State Machines: Approaches o Three approaches n Hardcoded (switch statement) n Scripted n Hybrid Approach Claypool and Lindeman, WPI, CS and IMGD 22 Based on Chapter 5. 3, Introduction to Game Development

Finite-State Machine: Hardcoded FSM void Run. Logic( int * state ) { switch( state ) { case 0: //Wander(); if( See. Enemy() ) break; { *state = 1; } case 1: //Attack(); if( Low. On. Health() ) { *state = 2; } if( No. Enemy() ) { *state = 0; } break; case 2: //Flee(); if( No. Enemy() ) break; { *state = 0; } } } Claypool and Lindeman, WPI, CS and IMGD 23 Based on Chapter 5. 3, Introduction to Game Development

Finite-State Machine: Problems with Switch FSM 1. Code is ad hoc n Language doesn't enforce structure 2. Transitions result from polling n Inefficient – event-driven sometimes better 3. Can't determine 1 st time state is entered 4. Can't be edited or specified by game designers or players Claypool and Lindeman, WPI, CS and IMGD 24 Based on Chapter 5. 3, Introduction to Game Development

Finite-State Machine: Scripted with alternative language Agent. FSM { State( STATE_Wander ) On. Update Execute( Wander ) if( See. Enemy ) Set. State( On. Event( Attacked. By. Enemy ) Set. State( Attack ) State( STATE_Attack ) On. Enter Execute( Prepare. Weapon ) On. Update Execute( Attack ) if( Low. On. Health ) Set. State( if( No. Enemy ) Set. State( On. Exit Execute( Store. Weapon ) State( STATE_Flee ) On. Update Execute( Flee ) if( No. Enemy ) Set. State( } Claypool and Lindeman, WPI, CS and IMGD STATE_Attack ) STATE_Flee ) STATE_Wander ) 25 Based on Chapter 5. 3, Introduction to Game Development

Finite-State Machine: Scripting Advantages 1. Structure enforced 2. Events can be triggered, as well as polling 3. On. Enter and On. Exit concept exists 4. Can be authored by game designers n Easier learning curve than straight C/C++ Claypool and Lindeman, WPI, CS and IMGD 26

Finite-State Machine: Scripting Disadvantages o Not trivial to implement o Several months of development n Custom compiler o With good compile-time error feedback n Bytecode interpreter o With good debugging hooks and support o Scripting languages often disliked by users n Can never approach polish and robustness of commercial compilers/debuggers n Though, some are getting close! Claypool and Lindeman, WPI, CS and IMGD 27 Based on Chapter 5. 3, Introduction to Game Development

Finite-State Machine: Hybrid Approach o Use a class and C-style macros to approximate a scripting language o Allows FSM to be written completely in C++ leveraging existing compiler/debugger o Capture important features/extensions n n n n On. Enter, On. Exit Timers Handle events Consistent regulated structure Ability to log history Modular, flexible, stack-based Multiple FSMs, Concurrent FSMs o Can't be edited by designers or players o Kent says: "Hybrid approaches are evil!" Claypool and Lindeman, WPI, CS and IMGD 28 Based on Chapter 5. 3, Introduction to Game Development