ViZDoom Navigation Dataset
收藏arXiv2025-09-30 收录
下载链接：
https://sites.google.com/view/carml
下载链接
链接失效反馈官方服务：
资源简介：
该数据集涉及在模拟环境（ViZDoom）中进行的 第一人称视觉导航，场景是一个充满五个不同物体的房间，允许探索和互动。此外，该数据集支持通过无监督元训练研究任务的发现，并评估将技能转移到测试任务上的效果。数据集中的物体从50个物体集合中抽取五个不同的物体，并且它们的位置是随机放置的。该任务是在受限环境中对物体进行视觉导航和互动。
This dataset focuses on first-person visual navigation conducted in the simulated environment ViZDoom, set in a room populated with five distinct objects that support exploration and interaction. Additionally, this dataset enables research on task discovery via unsupervised meta-training, as well as the evaluation of skill transfer performance on test tasks. The objects included in the dataset are five distinct samples selected from a pool of 50 objects, with their positions randomly placed. The core task is visual navigation and interaction with objects within a constrained environment.
提供机构：
Not explicitly mentioned.
