| 文件 | 作用 |
modelbest_robo_dataset/data_types.py | Pydantic 类型定义 (Episode, Ref, Content, Message, TimeseriesName) |
modelbest_robo_dataset/writer.py | 通用写入器: RawEpisode → SSTable/JSONL + Parquet + MP4 + stats |
modelbest_robo_dataset/reader.py | 读取验证器: SSTable/JSONL + Parquet + MP4 → 训练样本 |
modelbest_robo_dataset/validator.py | ActionValidator: state/action 一致性校验 |
modelbest_robo_dataset/sources/base.py | EpisodeSource 接口 + RawEpisode 中间表示 |
modelbest_robo_dataset/sources/lerobot.py | LeRobot v3.0 → RawEpisode |
modelbest_robo_dataset/sources/rh20t.py | RH20T → RawEpisode (含 force+audio) |
modelbest_robo_dataset/sources/fuse.py | TFRecord/RLDS → RawEpisode (含 IMU+audio) |
modelbest_robo_dataset/sources/robomind.py | RoboMIND HDF5 → RawEpisode (puppet/franka/tiangong 三变体) |
modelbest_robo_dataset/embodied_dataset.py | 训练侧 PyTorch Dataset 封装,按 timestamp 组装视频与时序样本 |
modelbest_robo_dataset/scripts/convert.py | 统一转换入口 (--source lerobot/rh20t/fuse/robomind) |
modelbest_robo_dataset/scripts/expand_skeleton.py | episode 骨架展开为单帧 / context 骨架 |
modelbest_robo_dataset/scripts/generate_viz.py | 生成交互式 HTML 可视化页面与 datasets.json |