Persona-driven Dominant/Submissive Map (PDSM) Generation for Tutorials

Key insights

6 takeaways

01
The Constrained MAP-Elites evolutionary algorithm enables the procedural generation of game tutorials that adapt to specific player behaviors.See it on page 1
02
AI agents representing three distinct personas—the Runner, the Monster Killer, and the Treasure Collector—are used to simulate and categorize level designs in MiniDungeons 2.See it on page 2
03
Levels are classified as dominant, submissive, or balanced based on how environmental variables like enemy density and item placement influence agent health and objective completion.See it on page 6
04
This framework allows developers to automate the creation of adaptive tutorials that nudge players toward optimal strategies or correct inefficient play patterns.See it on page 1
05
The system successfully populates an elite matrix of level designs, providing a scalable alternative to static, one-size-fits-all tutorial environments.See it on page 1
06
By mapping environmental features to agent success, designers can generate content tailored to a wide spectrum of individual strategic tendencies and skill levels.See it on page 2

Summary

Procedural content generation for video game tutorials is evolving through the implementation of persona-driven design, which utilizes AI agents to simulate diverse playstyles during the level creation process. By employing the Constrained MAP-Elites evolutionary algorithm, developers can generate environments that actively encourage or discourage specific player behaviors. This approach shifts the industry standard away from static, one-size-fits-all tutorials toward adaptive systems capable of providing personalized guidance based on individual strategic tendencies.

The methodology focuses on three distinct AI personas—the Runner, the Monster Killer, and the Treasure Collector—within the context of the game MiniDungeons 2. By evaluating agent health outcomes and objective completion, the system categorizes levels as dominant, submissive, or balanced. Data indicates that these generated environments successfully manipulate variables such as enemy density and item placement to nudge agents toward intended behaviors. While the Runner persona often serves as a behavioral baseline, the system demonstrates a robust capacity to populate an elite matrix of level designs that offer varying degrees of strategic headroom.

This framework establishes a viable pathway for automating the development of adaptive tutorials that respond to player performance. By mapping how specific environmental features influence agent success, designers can create content that reinforces optimal strategies or corrects inefficient play patterns. This research contributes to the broader field of experience-driven design, providing a scalable, algorithmic solution for creating responsive game environments that cater to a wide spectrum of player motivations and skill levels.