Roguelike Radio Episode 142: Roguelike Development Conference 2017
Dec 15, 2017
Tom Ford (aka Akka Flend), roguelike developer known for procedural dungeon generation and narrative-driven levels. He discusses using GANs and AI for level generation, narrative-first dungeon planning with topological ordering, and practical tools and algorithms for hobbyist developers. Short takes on ASCII aesthetics, adaptive tutorials, and pacing systems round out the conversation.
AI Snips
Chapters
Transcript
Episode notes
Experiment With ML For Procedural Content
- Try using machine learning to generate game content when heuristics become cumbersome for specific level features.
- Tom Ford trained a GAN overnight on simple dungeon samples and produced plausible layouts, though connectivity issues remained.
Dictate Short Code To Engage New Programmers
- When teaching kids to code, keep initial typing short and dictate code lines to maintain engagement and pace.
- Horst's Python roguelike lessons used very small code snippets and dictation so children quickly saw characters move on screen.
Kids Test Limits By Cheating And Mashing
- Kids eagerly experiment in roguelike lessons by pushing limits like infinite hit points or moving off-screen.
- Tom Ford recalled kids mashing keys to send the player off-screen and asking for a million hit points to test system limits.