Parallel Computation in Game AI

游戏AI中的并行计算

• 批准号: 1972112
• 金额: --
• 依托单位: Heriot-Watt University
• 依托单位国家: 英国
• 项目类别: Studentship
• 财政年份: 2016
• 资助国家: 英国
• 起止时间: 2016 至 无数据
• 项目状态: 已结题
• 来源: https://gtr.ukri.org/projects?ref=studentship-1972112
• 关键词: Parallel; Computation; Game; AI

The goal of this project is to develop parallelised versions of AI techniques and to explore the potential of these techniques within interactive games in particular in the existing Jomini Engine, a historical Massively Multiplayer Online Role Playing Games (MMORPG's). We will implement parallel versions of AI search algorithms, initially based on established sequential algorithms and later exploring novel parallel algorithms particularly suited for this application domain. We will analyse and identify performance differences between each version and reasons for the variation in performance. The main metric of success will be the (absolute) speedup achieved on large-scale parallel machines, such as clusters or GPUs, and the suitability of the results in the context of a concrete game engine, such as guaranteed response time from an AI-controlled agent. In the next phase of the project we will explore different parallel programming technologies, as MPI, OpenMP and OpenCL, and different sequential host languages, such as C/C++ & C#;, to finalise the infrastructure for the main part of this research in the area of parallel programming for (interactive) games.

该项目的目标是开发人工智能技术的并行版本，并探索这些技术在互动游戏中的潜力，特别是在现有的Jomini引擎中，这是一款历史悠久的大型多人在线角色扮演游戏（MMORPG）。我们将实现人工智能搜索算法的并行版本，最初基于已建立的顺序算法，然后探索特别适合该应用领域的新型并行算法。我们将分析和确定每个版本之间的性能差异以及性能变化的原因。成功的主要衡量标准将是在大规模并行机（如集群或GPU）上实现的（绝对）加速，以及结果在具体游戏引擎上下文中的适用性，例如来自AI控制代理的保证响应时间。在该项目的下一阶段，我们将探索不同的并行编程技术，如MPI，OpenMP和OpenCL，以及不同的顺序主机语言，如C/C++和C#;，以最终确定本研究的主要部分的基础设施在（交互式）游戏并行编程领域。