μRTS

该数据集是一个简约的实时战略游戏，它保留了让实时战略游戏在人工智能视角下具有挑战性的核心特征，包括同时性和持续性动作、高分支因子以及实时决策制定。此外，动作空间随着地图尺寸的组合式增长，这增加了深度强化学习代理可能采取的无效动作数量。该数据集的动作空间可以根据地图尺寸进行扩展，任务包括资源采集。This dataset is a minimalist real-time strategy (RTS) game that retains the core challenging features of RTS games from an AI perspective, including simultaneous and continuous actions, high branching factor, and real-time decision-making. Additionally, the action space grows combinatorially with map size, which increases the number of invalid actions that deep reinforcement learning agents may take. The action space of this dataset can be scaled based on map size, and the tasks include resource gathering.

搜集汇总

背景与挑战

背景概述

μRTS是一个简约的实时战略游戏数据集，保留了实时战略游戏的核心挑战特征，如同时性和持续性动作、高分支因子及实时决策。其动作空间随地图尺寸组合式增长，增加了深度强化学习代理的无效动作数量，主要任务包括资源采集。