Meta-World Benchmark
收藏arXiv2025-09-30 收录
下载链接：
http://meta-world.github.io
下载链接
链接失效反馈官方服务：
资源简介：
该数据集为元强化学习提供了一个基准，其中包括了一个模拟的Sawyer机器人与日常物品互动的各种任务场景。该数据集包含了520个自然语言描述，与1300个场景配套使用，用于训练、验证和测试强化学习策略。这些场景中包含了不同数量的对象，任务是通过自然语言指导机器人进行操作探索。
This dataset serves as a benchmark for meta-reinforcement learning, encompassing various task scenarios where a simulated Sawyer robot interacts with daily objects. It contains 520 natural language descriptions paired with 1300 scenarios, which are utilized for training, validating and testing reinforcement learning policies. These scenarios feature varying numbers of objects, with the core task being to guide the robot to conduct operational exploration through natural language instructions.
提供机构：
Meta-World benchmark
