Puddle-Jump Gridworld

该数据集描述了一个具有离散状态和动作空间的环境，其中智能体在一个网格中导航以到达目标，同时避开障碍物——水坑。在此环境中，智能体每执行一个动作会获得-0.05的奖励，而到达目标则能获得+1000的奖励。值得注意的是，该环境存在一些无法区分的状态，这些状态会影响最优策略的制定。该数据集的规模为10x10的网格，任务则是具有稀疏奖励的导航任务。

This dataset describes an environment with discrete state and action spaces, which is built on a 10×10 grid. In this environment, an AI Agent navigates to reach a target while avoiding obstacles (puddles). The agent receives a reward of -0.05 for each action taken, and a reward of +1000 upon reaching the target successfully. Notably, this environment contains some indistinguishable states that complicate the formulation of the optimal policy. The overall task is a navigation task with sparse rewards.