novastar111/pusht_v0
收藏Hugging Face2026-04-24 更新2026-04-26 收录
下载链接:
https://hf-mirror.com/datasets/novastar111/pusht_v0
下载链接
链接失效反馈官方服务:
资源简介:
PushT PPO 1Hz Success Trajectories是一个用于强化学习和机器人视觉语言任务的数据集。数据集包含训练和测试两个分割,分别有200,000和200条记录。每条记录是一个成功的PushT episode,最多包含30个环境动作。记录中包含任务名称、环境标识、环境设置、历史步骤、统计数据、初始状态等信息。图像以JPEG base64字符串嵌入在JSONL记录中。数据集是通过PPO策略生成的,仅包含成功的episodes。
PushT PPO 1Hz Success Trajectories is a dataset for reinforcement learning and robotics vision-language tasks. It contains train and test splits with 200,000 and 200 records respectively. Each record is a successful PushT episode capped at 30 environment actions. Important fields include task name, environment identifier, environment settings, history of steps, statistics, initial state, etc. Images are JPEG base64 strings embedded in the JSONL records. The dataset was generated with a PPO policy and includes only successful episodes.
提供机构:
novastar111



