Investigating the effects of learning speeds on Xpilot agent evolution

摘要

In this paper we present a comparison of the effects of varying play speeds on a genetic algorithm in the space combat game Xpilot. Xpilot-AI, an Xpilot add-on designed for testing learning systems, is used to evolve the controller for an Xpilot combat agent at varying frames per second to determine an optimal speed for learning. The controller is a rule-based system modified to work with a genetic algorithm that learns numeric parameters for the agent's rule base. The goal of this research is to increase the quality and speed of standard learning algorithms in Xpilot as well as determine a suitable speed for employing Punctuated Anytime Learning (PAL) in the Xpilot-AI environment. PAL is the learning component of an overall system of autonomous agent control with real-time learning.