Minifundium: Exploring NPC Artificial Intelligence Behavior Models Through Farm Simulation Games

Minifundium is a farm simulation game project released by llerandi on GitHub (2026-06-12). Players take on the role of a rabbit and interact with chicken NPCs equipped with replaceable AI brains. The core goal is to provide a testing platform for AI research through gamification, exploring different NPC artificial intelligence behavior models. Project link: https://github.com/llerandi/minifundium

NPC behavior modeling in AI research faces challenges: traditional preset scripts and finite state machines lead to mechanical behaviors. With the development of machine learning, researchers need more natural and intelligent NPC behavior patterns. Minifundium combines game design with AI research, building a farm simulation environment as an ideal testing platform.

The core of the game integrates AI research into the experience: players interact with chicken NPCs that have replaceable "brains". Each "brain" is an independent module following a unified interface, supporting quick switching between different AI models (e.g., rule-based systems, reinforcement learning, etc.). The modular design lowers the threshold for experiments and supports A/B testing and educational scenarios.

The project can support research and comparison of multiple models:

• Rule-based systems (preset conditional actions)
• Finite state machines (state transitions)
• Behavior trees (tree-like decision logic)
• Reinforcement learning (learning optimal strategies through interaction)
• Goal-based planning systems (autonomous action planning)
• Hybrid architectures (integrating multiple technologies)

Academically, it provides a testing platform; for the game industry, it can enhance NPC intelligence and player immersion, extending to scenarios like open-world RPGs and strategy games; it also provides an entry point for AI ethics and interpretability research, helping the public understand the diversity and limitations of AI.

It is speculated that Unity/Unreal engines are used, with AI modules hot-swappable via plugins or scripts; the "brain" interface needs to standardize perceptual inputs (player position, environmental state) and behavioral outputs (movement, interaction, etc.); data collection records interaction data for behavior analysis and model optimization.

Minifundium represents a paradigm of gamified AI research, lowering the threshold for research and intuitively presenting complex concepts. For developers/researchers interested in game AI, NPC modeling, or reinforcement learning, it provides a lightweight experimental platform and may offer a glimpse of the prototype of future coexistence between intelligent agents and humans.