Hok_env

本文介绍了国王荣誉竞技场，这是一种基于国王荣誉的强化学习 (RL) 环境，这是目前世界上最受欢迎的游戏之一。与以前大多数工作中研究的其他环境相比，我们的环境对竞争性强化学习提出了新的泛化挑战。这是一个多主体问题，一个主体与对手竞争; 它需要泛化能力，因为它有不同的控制目标和不同的对手可以竞争。我们描述了国王荣誉领域的观察，行动和奖励规范，并提供了一个基于Python的开源界面，用于与游戏引擎进行通信。为了纪念Kings Arena，我们为20位目标英雄提供了各种任务，并提供了具有可行计算资源的基于RL的方法的初始基准结果。最后，我们展示了国王荣誉竞技场带来的普遍性挑战以及对挑战的可能补救措施。

This paper introduces Honor of Kings Arena, a reinforcement learning (RL) environment based on Honor of Kings, which is one of the most popular video games worldwide today. Compared with other environments studied in most prior works, our environment presents novel generalization challenges for competitive reinforcement learning. This is a multi-agent problem where one agent competes against opponents; generalization capability is required, as there exist diverse control objectives and a variety of opponents to contend with. We describe the observation, action, and reward specifications for the Honor of Kings domain, and provide a Python-based open-source interface for communicating with the game engine. In commemoration of Kings Arena, we have developed various tasks for 20 target heroes and released initial benchmark results of RL-based methods under feasible computational resources. Finally, we demonstrate the generalization challenges brought by Honor of Kings Arena and possible remedial measures for these challenges.

搜集汇总

背景与挑战

背景概述

Hok_env是基于王者荣耀开发的强化学习多智能体竞技环境，提供20位英雄任务和基准测试，由上海交大、腾讯AI实验室等机构联合开发，包含Python接口和完整技术文档。