blind-review-data/StarCraftMotion_sample
收藏资源简介:
StarCraftMotion是一个大规模的多智能体模拟基准数据集,基于《星际争霸》游戏回放构建,专注于对抗性和部分可观测场景下的智能体行为模拟。每个数据样本是一个固定长度的场景窗口(145帧,16 FPS,约9秒),包含所有单位状态、动态地图层和每玩家经济时间序列。数据集特别强调交互密集的窗口,用于测试部分可观测条件下的多智能体预测能力。数据集包含469,187个场景,分为训练、验证和测试集,覆盖多种地图和种族对战组合。数据以parquet格式存储,包含丰富的单位和地图信息,适用于多智能体模拟、战争迷雾处理和交互场景预测等研究。
StarCraftMotion is a large-scale benchmark for agent simulation under adversarial and partial observability scenarios, built from StarCraft replays. Each example is a fixed-length scenario window (145 frames at 16 FPS, ~9 seconds) containing all unit states, dynamic map layers, and per-player economy time series. The dataset is adversarially subsampled to overweight interaction-heavy windows, making it a stress test for multi-agent prediction under partial observability. It includes 469,187 scenarios split into training, validation, and test sets, covering various maps and race matchups. The data is stored in parquet format with detailed unit and map information, suitable for research in multi-agent simulation, fog-of-war handling, and interaction-heavy scene prediction.




