Meta-World Benchmark Simulation Tasks

该数据集是从meta-world基准中选取的十个操作任务组成，每个任务都具有随机的初始配置，并收集了专家演示。此外，每条轨迹包含了来自顶部视角相机的RGB图像，动作空间由七个离散动作组成。规模上，每个任务有100条轨迹，每条轨迹包含23至124帧不等。该数据集的任务是机器人持续模仿学习。

This dataset consists of 10 manipulation tasks selected from the Meta-World benchmark. Each task features randomized initial configurations, with expert demonstrations collected. Additionally, each trajectory includes RGB images captured by a top-down camera, and the action space comprises seven discrete actions. In terms of scale, each task has 100 trajectories, with each trajectory ranging from 23 to 124 frames. The core task of this dataset is robotic continual imitation learning.