10-Maze taskset
收藏arXiv2025-09-30 收录
下载链接:
http://github.com/sisl/MPHRL
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10个随机生成的迷宫,专为MuJoCo Ant环境设计,旨在评估终身学习性能。观察空间包括目标、观察空间、高斯噪声模型以及模型原语。在MPHRL算法的表现中,展示了强大的子策略迁移能力。该数据集的任务规模为10项任务,主要针对终身学习与任务分解。
This dataset contains 10 randomly generated mazes, specifically designed for the MuJoCo Ant environment to evaluate lifelong learning performance. The observation space includes the target, observation space, Gaussian noise model, and model primitives. The MPHRL algorithm demonstrates strong sub-policy transfer capabilities when tested on this dataset. This dataset consists of 10 tasks, focusing primarily on lifelong learning and task decomposition.
提供机构:
MuJoCo



