Evaluating Human-like Behaviors of Video-Game Agents Autonomously Acquired with Biological Constraints

Abstract

Designing the behavioral patterns of video game agents (Non-player character: NPC) is a crucial aspect in developing video games. While various systems that have aimed at automatically acquiring behavioral patterns have been proposed and some have successfully obtained stronger patterns than human players, those patterns have looked mechanical. When human players play video games together with NPCs as their opponents/supporters, NPCs’ behavioral patterns have not only to be strong but also to be human-like. We propose the autonomous acquisition of NPCs’ behaviors, which emulate the behaviors of human players. Instead of implementing straightforward heuristics, the behaviors are acquired using techniques of reinforcement learning with Q-Learning and pathfinding through an A* algorithm, where biological constraints are imposed. Human-like behaviors that imply human cognitive processes were obtained by imposing sensory error, perceptual and motion delay, physical fatigue, and balancing between repetition and novelty as the biological constraints in computational simulations using “Infinite Mario Bros.”. We evaluated human-like behavioral patterns through subjective assessments, and discuss the possibility of implementing the proposed system.