RMBench

该数据集为机器人操作提供了一个基准，重点关注高维连续动作和状态空间，并采用观察像素输入的强化学习算法。该数据集涉及基于原始的RGB彩色像素训练智能体，并包含了多种强化学习算法，这些算法被评估其性能和稳定性。在规模上，该数据集采用了多个随机种子（每个任务5个不同的种子），任务内容为机器人操作任务。

This dataset serves as a benchmark for robotic manipulation tasks, focusing on high-dimensional continuous action and state spaces, and is tailored for reinforcement learning algorithms that use pixel inputs as observations. It involves training AI Agents with raw RGB color pixel data, and incorporates multiple reinforcement learning algorithms whose performance and stability are evaluated. In terms of scale, the dataset employs multiple random seeds, with 5 distinct seeds allocated to each robotic manipulation task.