GridWorld MDP
收藏arXiv2025-09-30 收录
下载链接：
https://github.com/ddv-lab/robust-cf-inference-in-MDPs
下载链接
链接失效反馈官方服务：
资源简介：
该数据集是一个4x4的网格环境，其中智能体必须从左上角出发，避开中心位置的危险终止状态，最终到达右下角的目标状态。智能体可以向上、下、左或右移动，但存在一定概率的意外移动。此外，该环境包含两个版本，分别具有不同的意外移动概率（分别为0.9和0.4）。在4x4的网格规模下，任务目标是最大化奖励同时避免惩罚。
This dataset is a 4x4 grid-world environment where an AI Agent must start from the top-left corner, avoid the hazardous terminal state located at the center, and ultimately reach the goal state at the bottom-right corner. The agent can move up, down, left, or right, but there is a certain probability of unintended movement during execution. Additionally, this environment has two variants with different unintended movement probabilities: 0.9 and 0.4, respectively. For this 4x4 grid setup, the task objective is to maximize rewards while avoiding penalties.
