Point-Maze Dataset
收藏arXiv2025-09-30 收录
下载链接：
https://github.com/sherilan/cs285-project/tree/master
下载链接
链接失效反馈官方服务：
资源简介：
该数据集包含了2000个样本，这些样本是从一个代理在二维点状迷宫问题中导航时收集的，其奖励结构基于高斯核。该数据集被用于训练一个模型，目的是使其能够以高准确度区分出学习策略状态和随机策略状态。在此规模上，即2000个样本的任务是，在强化学习中发现技能。
This dataset consists of 2000 samples collected from an agent navigating a 2D grid-point maze problem, where the reward structure is based on Gaussian kernels. It is employed to train a model with the objective of enabling the model to distinguish between learned policy states and random policy states with high accuracy. At this scale (2000 samples), the task focuses on discovering skills in reinforcement learning.
