Stochastic Combination Lock Environment

该数据集包含H个层级，每个层级有3种状态和4种动作。在这其中，有两种状态会带来高回报，而另一种是死亡状态。动作有可能以0.1的概率被翻转，同时，为了状态编码，还会添加随机的伯努利噪声。此外，数据集还包括两种任务变体，用于测试算法对局部最优解的鲁棒性。规模方面，多个层级包含了高回报和低回报状态。该任务旨在研究强化学习中的探索与利用问题。

This dataset consists of H levels, each containing 3 states and 4 actions. Among these, two states yield high rewards, while the remaining one is a terminal (death) state. Actions have a 0.1 probability of being flipped, and random Bernoulli noise is added for state encoding. Furthermore, the dataset includes two task variants designed to test algorithm robustness against local optima. Multiple levels in the dataset encompass both high-reward and low-reward states. This task aims to investigate the exploration-exploitation tradeoff in reinforcement learning.