I’m working on a thesis about "story-driven NPCs in a reinforcement-learning world", and I’m building a small multi-agent RL environment as a prototype. However, I’m unsure how to push the idea further, especially regarding the “story-driven NPCs” aspect.

I’d appreciate suggestions or pointers on these questions:

  1. World design: What kind of environment best showcases story-driven NPC behavior learned through RL? How can I demonstrate that the NPCs’ behaviors are genuinely story-driven instead of just reactive?

  2. Comparison to traditional models: In what ways can RL-based story-driven NPCs differ from classical approaches like FSMs or Behavior Trees? How can I highlight these differences experimentally (e.g., metrics, tasks, ablation tests)?

  3. Existing research: Are there known methods, papers, or frameworks that combine narrative structures with RL agents or MARL environments?

Any guidance, examples, or references on these questions would be greatly appreciated.