Source metadata
- Type: University module lecture slides
- Module: CRE341 — Game Character Control and Behaviour
- Lecturer: Prof Darryl Charles
- Institution: Ulster University
- Academic year: 2025–26
- Lectures ingested: 8 of 10 weeks (Wk 1.2, 2.1, 3.r1, 4.2, 5.1, 6.1, 7, 8+9); Wk 10 (Game Issues and Ethics) not yet uploaded to Drive
Key takeaways
- Game narrative design is framed as system design, not story writing — narrative emerges from mechanics, rules, and player actions as well as authored content
- Environmental storytelling communicates through implication — players infer events from spatial evidence (layout, damage, materials, light, sound) without exposition; legibility and ludonarrative alignment are the key design tests
- Character design is best framed as goal-driven systems in motion (Sheldon-style): desire creates momentum; mechanics are character expression; conflict is the engine of play
- NPC AI architecture selection is a design decision: FSM for simplicity, BTs for reactivity, GOAP for emergent tactics, Utility AI for smooth adaptation, social simulation for relationship-driven worlds, LLMs for conversational depth (with safety guardrails)
- Ludonarrative harmony/dissonance is a design problem (not a player problem): when optimal play contradicts narrative framing, that is a design flaw to fix
- Game AI agent architectures form a clear hierarchy: Behaviour Trees for reactive behaviours, GOAP for deliberative planning, both driven by Action Selection → Steering → Locomotion layers
- OOP + State Pattern is the standard professional approach to AI state machines: each behaviour isolated in its own class, state machine delegates to the current state via polymorphism
- PCG is primarily a replayability and scale tool; the key production decision is topology (BSP vs random walk vs growing algorithm) which determines the feel of generated spaces
- Genetic algorithms in games are best applied via concrete problem encodings: binary (Knapsack), permutation (TSP), sequence (Lunar Lander); elitism is essential for preventing regression
- Reinforcement learning in games uses Q-tables (or neural networks) to learn state-action-reward mappings; Q-Learning and SARSA are the foundational temporal difference algorithms
- Generative AI now covers every phase of the game dev pipeline (design → art → animation → code → testing → analytics); tools include Microsoft Muse, Adobe Firefly, Scenario AI, SoundRaw, GitHub Copilot
- Computational creativity is an open question; early computer art pioneers (Vera Molnar, 1960s) predate modern generative AI by decades
- AI, ML, and deep learning are nested concepts; most game AI is Weak AI (narrow task-specific); AGI and Superintelligence remain research horizons
Notable claims
- “Game narrative design is the design of meaningful player experience through systems, choices, and structured delivery, rather than the authoring of a fixed story.” (Prof Charles, Wk 1.2)
- “Ludonarrative dissonance: optimal play contradicts the story’s framing or values. This is a design problem, not a player misunderstanding.” (Wk 1.2)
- “Players forgive simplicity more readily than inconsistency.” (Wk 1.2, on narrative coherence)
- Artificial stupidity is a design goal, not a failure: “Real people are stupid… sometimes! There can be much humour in a well-designed stupid-bot!” (Wk 5.1, citing Falstein)
- Elite (1984) holds the Guinness World Record for first use of procedural generation in a video game (Wk 6.1)
- Entropy: “learning that an unlikely event has occurred is more informative than learning that a likely event has occurred” (Wk 8+9, citing Goodfellow et al.)
- AI definition: “AI is concerned with intelligent behaviour in artefacts. Intelligent behaviour involves perception, reasoning, learning, communicating and acting in complex environments.” (Wk 7, citing Nils Nilsson)
- Human intelligence definition: “a mental quality that consists of the abilities to learn from experience, adapt to new situations, understand and handle abstract concepts, and use knowledge to manipulate one’s environment.” (Wk 7, citing Britannica)
Relevance
Directly informs or supplements:
- narrative-design — player story/plot/story distinction; ludonarrative; narrative topology; emotion through interaction; delivery channels
- ai-state-machine-pattern — new page (OOP + State Pattern for AI)
- game-ai-agent-design — new page (AI definitions, agents, BTs, GOAP)
- overview-cre341-agent-ai-route — recommended teaching route through FSM, BTs, utility AI, GOAP, HTN, and player modelling
- procedural-generation — terrain algorithm comparison table; dungeon generation algorithms; PCG history
- genetic-algorithms — elitism; Knapsack/TSP/Lunar Lander worked examples; encoding strategies
- machine-learning-games — new page (ML taxonomy, probability, Bayes, entropy, RL, neural networks); updated with AI taxonomy, CNN/GAN architecture, history
- generative-ai-game-dev — new page (AI in game dev pipeline, generative AI tools, computational creativity, intelligent interactive storytelling)
- environmental-storytelling — new page (Wk 2.1: environmental narrative, audio-visual design, legibility, case studies)
- character-design — new page (Wk 4.2: character-as-system, NPC AI architectures, utility AI, social simulation, player modelling, LLM NPCs)
Open questions raised
- How does deliberative planning (GOAP) scale to complex multi-NPC scenarios without performance issues?
- At what point does procedural generation require ML/PCGML methods rather than algorithmic approaches?
- What is the minimum Q-table complexity for practically useful RL behaviour in a game?
- How do modern LLM-driven dialogue systems (e.g. Yarn Saliency) change the ludonarrative relationship?