Puddle World
收藏arXiv2025-09-30 收录
下载链接：
https://github.com/anonicml2019/icml_2019_state_abstraction
下载链接
链接失效反馈官方服务：
资源简介：
该数据集呈现了一个连续的网格世界，其中状态由两个坐标（x, y）表示。在这个世界中，智能体必须避开会造成惩罚的障碍物，以达到目标位置。此外，该数据集包含了从具有特定目标位置的泥潭实例中抽取的（状态，最优策略，奖励，下一个状态）四元组样本。其规模属于中等，包含4000个样本元组，任务类型为强化学习。
This dataset presents a continuous grid world, where the state is denoted by a two-dimensional coordinate pair (x, y). In this environment, the AI Agent must avoid obstacles that impose penalties to reach the target position. Additionally, this dataset contains (state, optimal policy, reward, next state) quadruple samples extracted from instances of this grid world with specific target positions. It has a medium scale, containing a total of 4000 sample tuples, and the task belongs to the field of reinforcement learning.
提供机构：
Anonymized code repository
