ProcGen Leaper

该数据集是一个为深度强化学习中的零样本泛化而设计的程序生成游戏环境，主要关注跳跃任务。与迷宫任务类似，该环境支持计算Q函数的地面真实值及神经网络近似中的不确定性，相关结果已在补充材料中呈现。任务是在跳跃者环境中进行在线规划与学习。

This dataset is a procedurally generated game environment designed for zero-shot generalization in deep reinforcement learning, with a primary focus on jumping tasks. Similar to maze tasks, this environment supports the calculation of the ground truth for the Q-function as well as uncertainty in neural network approximations, and relevant results have been presented in the supplementary materials. The task entails online planning and learning within the jumper environment.