GCRL Benchmark Manipulation Tasks

该数据集包含了针对机器人操作任务的标准基准环境，特别关注使用密集奖励的目标条件强化学习（GCRL）。实证评估确认，在多个环境中，采用密集奖励可以提升样本复杂性和策略质量。该数据集规模涵盖12个标准基准环境，其任务重点在于连续控制任务中的目标条件强化学习（Gcrl）。

This dataset comprises standard benchmark environments for robotic manipulation tasks, with a particular focus on dense-reward-based goal-conditioned reinforcement learning (GCRL). Empirical evaluations have confirmed that adopting dense rewards improves sample complexity and policy quality across multiple environments. The dataset covers 12 standard benchmark environments, with its tasks focusing on goal-conditioned reinforcement learning (GCRL) in continuous control tasks.