Roguelike Radio Episode 144: AI
Feb 6, 2018
Brian Bucklew, indie dev behind Caves of Cud who builds practical AI for mutable worlds, and Tommy Thompson, lecturer and creator of the AI and Games channel with ML and NPC research. They explore AI in procedurally generated games. They cover testing and modular behaviors, risky emergent interactions and fixes, procedural sims and scaling simulated worlds, and blending ML with narrative and agent systems.
AI Snips
Chapters
Transcript
Episode notes
Procedural Worlds Require Reactive AI
- Procedural worlds force AI to be reactive to highly variable environments rather than preprocessed contexts.
- Brian Bucklew explains mutable levels in Caves of Qud like lava floors, flying creatures, and sentient walls that break static nav approaches.
Modularize Behaviors To Keep Testing Tractable
- Modularize behaviors so you can test components independently and avoid combinatorial QA explosion.
- Brian Bucklew says encapsulated behaviors let new environmental objects work predictably without testing every combined interaction.
Cat Keeps Burning Itself On Torches
- A cat in Caves of Qud repeatedly sat on torches and self-immolated after torches were made hot.
- Community videos flagged the issue so the team added navigation weight to torches to make the event rarer, not impossible.