AutoBus - Computer Vision-Driven Autonomous Agent for Decision-Making in Complex Game Environments

Abstract

This thesis presents AutoBus, an advanced automation framework designed to operate within the complex environment of Limbus Company. It addresses the practical challenge of excessive repetitive labor, a hallmark of modern "gacha" gaming models, where players must perform hours of manual, low-variance tasks to secure progress. Unlike static macros, AutoBus applies a Computer Vision (CV)-centric approach, using template matching and simulated input to automate gameplay and adapt to dynamic game-state changes. The project features a modular architecture that separates core engine logic, such as screen detection and process control, from task-specific execution modules. Additionally, the framework includes a suite of developer utilities for asset cropping and click validation, along with a Streamlit interface for user configuration and live monitoring. In conclusion, the project successfully automates gameplay without human intervention and is capable of stable, long-term operation, effectively reclaiming hours of user labor.