NetHack Expert Trajectories

该数据集包含了用于在NetHack环境中训练BLINDER方法的专家轨迹集合，它由状态与专家动作的配对组成。此外，该数据集还包括了五个专门设计的测试任务，旨在评估BLINDER方法的性能，特别关注多步规划。规模上，数据集涵盖了5个训练任务，共包含25条专家轨迹，任务内容为NetHack环境下的顺序决策制定。This dataset contains a collection of expert trajectories for training the BLINDER method in the NetHack environment, where each trajectory consists of paired states and expert actions. Additionally, the dataset includes five specially designed test tasks intended to evaluate the performance of the BLINDER method, with a particular focus on multi-step planning. In terms of scale, the dataset covers 5 training tasks with a total of 25 expert trajectories, all centered on sequential decision-making within the NetHack environment.