chore: normalize tsfile dataset metadata

README.md

@@ -0,0 +1,65 @@
+---
+license: apache-2.0
+task_categories:
+- robotics
+tags:
+- robotis
+- LeRobot
+- omx_f
+- timeseries
+- tsfile
+modality:
+- timeseries
+configs:
+- config_name: default
+  data_files:
+  - split: train
+    path: data/quvi_train.tsfile
+size_categories:
+- 10K<n<100K
+---
+
+# QUVI
+
+Converted from [`seongjun-k/QUVI`](https://huggingface.co/datasets/seongjun-k/QUVI), a LeRobot v2.1 robot dataset created with [Physical AI Tools](https://github.com/ROBOTIS-GIT/physical_ai_tools) and [LeRobot](https://github.com/huggingface/lerobot).
+
+## Source Dataset
+
+- Original dataset: [`seongjun-k/QUVI`](https://huggingface.co/datasets/seongjun-k/QUVI)
+- License: `apache-2.0`
+- Task category: robotics
+- Robot type: `omx_f`
+- Task: `pick` (`task_index=0`)
+- Source scale: 95 episodes, 14,263 frames, 95 videos, 30 fps
+- LeRobot codebase version: `v2.1`
+- Modalities: Time-series
+
+## Converted Files
+
+- TsFile table: `quvi_train`
+- TsFile path: `data/quvi_train.tsfile`
+- Rows: 14,263
+- Episode coverage: episode indexes `0..94`
+- Metadata: source `meta/` is mirrored in this repository, with `meta/info.json` updated to describe the converted TsFile artifact.
+
+## TsFile Schema
+
+- `Time`: integer milliseconds, computed as `round(timestamp * 1000)`.
+- TAG columns: `episode_index`, `task_index`.
+- FIELD columns: `frame_index`, `sample_index`, `observation_state_0..5`, `action_0..5`.
+- Source `index` is renamed to `sample_index`.
+- Source `observation.state[6]` is flattened to `observation_state_0..5`.
+- Source `action[6]` is flattened to `action_0..5`.
+
+## Conversion Notes
+
+The source `timestamp` column is not kept as a separate field because it is represented by `Time / 1000` seconds after conversion. The source video feature `observation.images.camera1` is not converted to TsFile and videos are not uploaded here; use the original dataset video tree at [`seongjun-k/QUVI/videos`](https://huggingface.co/datasets/seongjun-k/QUVI/tree/main/videos).
+
+The original `meta/info.json` reports `splits.train` as `0:94`, but `meta/episodes.jsonl` contains episode indexes `0..94` and their lengths sum to `total_frames=14263`. This conversion includes all 95 numeric episode parquet files and records the converted train split as `0:95`.
+
+## Minimal Read Example
+
+```python
+# Read data/quvi_train.tsfile with the Apache TsFile Java or Python SDK.
+# Table name: quvi_train
+```