Modeling the Defense of Weak Prey Agents Against a Strong Predator Agent

Abstract

We constructed and studied a model of interaction between a community of relatively weak prey  agents and a strong predator agent in a two‑dimensional grid world (a lattice environment typical of grid automata and agent-based models). The predator can attack, kill, and consume prey agents. Each prey agent is controlled by a neural network and adopts one of two behavioral strategies: (1) normal activity, or (2) defense against the predator. In the normal activity strategy, prey agents lie dormant, feed, breed, and move through the grid. In the defense strategy, they attempt to escape, threaten, or attack the predator. The neural network outputs control each agent’s ac tions. The predator follows a simpler, rule-based protocol: it can lie dormant, evade threatening prey, or attack them. Its behavior is governed by basic logic. We analyzed the model using computer simulations. We found that, with realistic parameters, the prey agents collectively overcome the predator: prey resource levels increase steadily, while the predator’s resources decline to zero, leading to its extinction. We also discovered that successful defense requires a sufficiently abundant food supply; when prey food is scarce, the predator successfully suppresses the prey population. We used computer simulation to analyze the model. When the prey agents’ food supply is low, the predator agent suppresses the prey agents.