Procgen Environment Suite

该数据集包含了一系列通过程序生成的环境，旨在研究强化学习任务中的样本效率和泛化能力。这些环境被设计成鼓励智能体在学习过程中掌握在不同设定和难度下都能通用的策略。数据集规模涵盖了包括CoinRun、Maze I、Maze II、Keys and Chests等多个环境，任务主要集中在类似街机视频游戏的强化学习研究。

This dataset comprises a series of procedurally generated environments intended to investigate sample efficiency and generalization in reinforcement learning tasks. These environments are designed to encourage agents to master strategies that generalize across diverse settings and difficulty levels during the learning process. The dataset includes multiple environments such as CoinRun, Maze I, Maze II, Keys and Chests, with its tasks primarily focused on reinforcement learning research inspired by arcade video games.

搜集汇总

背景与挑战

背景概述

该数据集是Procgen基准测试的一个分支，专门为研究深度强化学习中的目标泛化错误而设计。它包含多个修改后的环境变体，如coinrun_aisc和heist_aisc_many_chests，通过随机放置目标、调整钥匙与宝箱比例等方式，用于评估智能体在环境变化下的鲁棒性。