Download as PDF, PPTX
Uploaded byNick Pruehs
Game Programming 09 - AI
Game Programming AI This document discusses behavior trees, an approach to game AI architecture. Behavior trees split AI decision logic from actions and organize them into a directed tree structure. The root node executes logic which reports back as success, running, or failure. This passes control to child nodes. Key nodes include sequences which run children in order until failure, and selectors which run the first successful child. Behavior trees are modular, reusable, and can be data-driven to design AI visually without code. They have been successfully used in many games due to their flexibility and performance.
More Related Content
Similar to Game Programming 09 - AI
![Coded Agents – with UiPath SDK + LangGraph [Virtual Hands-on Workshop]](https://cdn.slidesharecdn.com/ss_thumbnails/codedagentsdeck-251215155422-5497c599-thumbnail.jpg?width=640&height=640&fit=bounds)
Game Programming 09 - AI
- 1. Game Programming AI Nick Prühs,Christian Oeing
- 2. Objectives • To understandthe importance of AI in games • To learn about different approaches to game AI architectures • To take a closer look at behavior trees in particular 2 / 90
- 3. Why AI? • Enemies •Allies • Challanges • Vivid game world • Toy 3 / 22
- 4. Early Example: PacMan “This is the heart of the game. I wanted each ghostly enemy to have a specific character and its own particular movements, so they weren’t all just chasing after Pac Man in single file, which would have been tiresome and flat.” - Toru Iwatani, Pac-Man creator 4 / 22
- 5. Early Example: PacMan 5 / 22 http://gameinternals.com/post/2072558330/understanding-pac-man-ghost-behavior
- 6. Left 4 Dead 6/ 22 http://www.valvesoftware.com/publications/2009/ai_systems_of_l4d_mike_booth.pdf
- 7. Naive Approach: Hard-CodedRules • If-Then-Else • Hard-coded • Fast to implement • Access to whole data model • No fixed structure 7 / 22
- 8. Finite State Machine •Splitting AI into different states • Arbitrary number of states • Well-defined state transitions 8 / 22
- 9. Hierarchical FSM 9 /22 • Hierarchical Structure • Allows transitions at same level, only • Drastically reduces the total number of transitions
- 10. Planner • Goal-oriented • SpecificAI not defined by developer 10 / 22
- 11. Utility-based • Action-based, similarto planner • No goal, thus no action chains 11 / 22
- 12. Neural Networks • ImprovingAI through training data • No way of modifying the AI after training • Further reinforces the disadvantages of action-based architectures • Rarely applied in games • Exceptions: ▪ Creatures ▪ Supreme Commander 2 12 / 22
- 13. Behaviour Trees: History •Became popular in 2005 with Halo 2 • Successfully implemented in ▪ Halo 2 + 3 ▪ Spore ▪ Swords & Soldiers ▪ GTA: Chinatown Wars ▪ The Bourne Conspiracy ▪ SWAT 4 ▪ Bioshock ▪ Dark Sector ▪ … and more 13 / 22
- 14. Behaviour Trees: What‘snew? • Similar to Finite State Machines • Split decision logic from actual actions 14 / 22
- 15. Behaviour Tree Structure •Directed tree ▪ Single root node ▪ Inner nodes = decision logic ▪ Leafs = Actions 15 / 12
- 16. Behaviour Tree Ticks •Tick starts at root node • Current node executes its logic • Node state is reported back to parent node: ▪ Success ▪ Running ▪ Failure 16 / 12
- 17. Behaviour Tree TickExample 17 / 12
- 18. Prioritized Structure • Childrenare executed left-to-right 18 / 12
- 19. Modular Structure • Addingnew nodes possible everywhere 19 / 12
- 20. Parametrized Structure • Nodedefines parameters and their types • Parameters definie actual execution • Increases reusability • Example: Move to position 20 / 12
- 21. Data Exchange • Achievedthorugh a blackboard • Dynamically defines parameters of other nodes • Nodes remain independent of each other • Simple implementation, e.g. Dictionary or Hashtable 21 / 12
- 22. Node: Selector • Arbitrarynumber of children • Only one child node active at the same time • Selects first child that doesn‘t return Failure 22 / 12
- 23. Node: Sequence • Arbitrarynumber of children • Only one child node active at the same time • Selects children one after another until one returns Failure 23 / 12
- 24. Combining Sequences andSelectors 24 / 12
- 25. Node: Decorator • Singlechild 25 / 12
- 26. Blackboard • AI memory •Exchanges data between nodes 26 / 12
- 27. Blackboard • Keys needto be unique • May cause conflicts 27 / 12
- 28. Picklock • Easiest wayto enter building: Open door! 28 / 12
- 29. Picklock • Not alwaysthat easy 29 / 12
- 30. Picklock • Generic wayof entering a building through any kind of entrance 30 / 12
- 31. Picklock • There aremany ways to get in 31 / 12
- 32. Crafting • Using ingredientsto create new items • Ingredients need to be available • What to do, if missing any ingredients? • Ingredients might be crafted as well 32 / 12
- 33.
- 34. More Examples • Combat:Fighting, healing, running, taking cover • Survival: Running, guarding, fighting, eating • Strategy: Training workers, building structures, training fighters • Dialogues: Conditional answers, linked answers 34 / 12
- 35. Data Driven • Structureallows building a generic visual editor • Usable by game designers as well • Can be used in multiple projects 35 / 12
- 36. Available Tools • Unity3D ▪Behave 2 (Angry Ant) ▪ Behavior Designer (Opsive) • Unreal Engine ▪ Built-in • Cry Engine ▪ Built-in 36 / 12
- 38. Custom Tools • Definedata structure ▪ e.g. XML, easy to read and edit 38 / 12 <BehaviorTree Name="CraftItem"> <Sequence> <Action Type="HasRecipe" List="Ingredients" /> <Decorator Type="Iteration" List="Ingredients" Iterator="Ingredient" > <Sequence> <Selector> <Action Type="TakeItemFromInventory" ItemRef="Ingredient" StoreIn="IngredientItem" /> <Action Type="SearchItem" ItemRef="Ingredient" Distance="50" StoreIn="IngredientItem" /> <BehaviorTreeRef Name="CraftItem" ItemRef="Ingredient" StoreIn="IngredientItem" /> <Action Type="SearchItem" ItemRef="Ingredient" Distance="500" StoreIn="IngredientItem" /> </Selector> <Action Type="PutItemInList" ItemRef="IngredientItem" List="IngredientItems" /> </Sequence> </Decorator> <Action Type="MixIngredients" List="IngredientItems" /> </Sequence> </BehaviorTree>
- 39. Custom Tools • Genericapproach • How to make game-specific actions available to generic editor? ▪ Use editor to add actions ▪ Export from game and import in editor ▪ Generic actions ▪ Import from DLL using reflection 39 / 12
- 40. Behaviour Tree Summary •Generic structure leads to many applications ▪ Reuse of decision logic and game-specific actions ▪ Common editor for multiple projects ▪ Learn once, apply again and again • Strictly follows rules ▪ No dynamic AI, such as when using Planner 40 / 12
- 41. Conclusion • Many differentarchitectures ▪ (Hierarchical) Finite State Machines ▪ Planner ▪ Utility-based ▪ Neural Network ▪ Behavior Trees • Choice depends on actual game • Each architecture is better than no architecture! 41 / 22
- 42. References • Heiko Klinge.Warum KI in Spielen stagniert. http://www.tecchannel.de/webtechnik/entwicklung/1744817/warum_kuenstliche_intelligenz_ki_in_ spielen_stagniert/ • Chad Birch. Pac Man AI. http://gameinternals.com/post/2072558330/understanding-pac-man- ghost-behavior • Don Hodges. Pac Man AI Bug. http://donhodges.com/pacman_pinky_explanation.htm • Alex Champandard. Top AI Games. http://aigamedev.com/open/highlights/top-ai-games/ • Mike Robbins. Neural Network in Supreme Commander 2. http://twvideo01.ubm- us.net/o1/vault/gdc2012/slides/Summit_AI/Robbins_Michael_Off%20the%20Beaten.pdf • Mike Booth. The AI systems of Left 4 Dead. http://www.valvesoftware.com/publications/2009/ai_systems_of_l4d_mike_booth.pdf • Damian Isla. Handling Complexity in the Halo 2 AI. http://www.gamasutra.com/view/feature/130663/gdc_2005_proceeding_handling_.php • Alex Champandard. Behavior Trees: Three Ways of Cultivating AI. http://www.gdcvault.com/play/1012744/Behavior-Trees-Three-Ways-of
- 43. References • Chris Simpson.Behavior Trees for AI. http://www.gamasutra.com/blogs/ChrisSimpson/20140717/221339/Behavior_trees_for_AI_How_t hey_work.php • Cry Engine. Coordinating Agents with Behavior Trees. http://docs.cryengine.com/display/SDKDOC4/Coordinating+Agents+with+Behavior+Trees • Chris Hecker. My Liner Notes for Spore. http://chrishecker.com/My_Liner_Notes_for_Spore#Behavior_Tree_AI • Joost "Oogst" van Dongen. AI in Swords & Soldiers. http://joostdevblog.blogspot.de/2010/12/ai-in- swords-soldiers-part-1.html • Unreal Engine. Behavior Trees. https://docs.unrealengine.com/latest/INT/Engine/AI/BehaviorTrees/index.html • Cry Engine. Modular Behavior Tree. http://docs.cryengine.com/display/SDKDOC4/Modular+Behavior+Tree
- 44.
- 45. 5 Minute ReviewSession • Name a few applications of AI in general! • How do finite state machines work? • How do hierarchical state machines improve on that? • What other AI architectures do you know? • What’s the basic structure of a behavior tree? • How do sequences and selectors work? • What’s a decorator node? • How do nodes exchange data?