Datasets:
Download README.ja.md from cy0307/ropedia-xperience-10m-task-suite-artifacts: direct link, hf CLI and curl.
- Browser
- Download file 8.8 kB
-
https://huggingface.co/datasets/cy0307/ropedia-xperience-10m-task-suite-artifacts/resolve/main/README.ja.md
- Command line
-
hf download hf://datasets/cy0307/ropedia-xperience-10m-task-suite-artifacts/README.ja.md
-
curl -L -o README.ja.md https://huggingface.co/datasets/cy0307/ropedia-xperience-10m-task-suite-artifacts/resolve/main/README.ja.md
Ropedia Xperience-10M Task Suite
Xperience-10M の公開タスク・評価レイヤー: サンプルデータ、20 個の embodied-AI タスク、ベースライン、Qwen3-Omni と Cosmos3 診断、基盤モデル訓練方向。
English · 中文 · Español · Français · Deutsch · 日本語 · 한국어 · Português
プロジェクト概要
このリポジトリは、公開 Xperience-10M サンプル episode を、検証可能な embodied AI タスク実験面に変換します。まずダッシュボードとプロジェクト状態を見て、その後 20 タスク、結果行列、Hugging Face ミラーを確認してください。
更新日: 2026-06-23。
範囲: 完全に再現可能なタスク suite は 1 つの公開サンプル episode に基づきます。128-episode の結果は public-safe な指標、レポート、予測要約、モデルカードのみを公開します。元の MP4/HDF5/RRD、完全な Qwen 重み、gated データは再配布しません。
2 つの証拠ライン
| ライン | データ単位 | 手法と結果 | 用途 |
|---|---|---|---|
| 1 sample episode | 5,821 frames、1,161 aligned 20-frame windows、8,546 dimensions。 | Minimal + Neural MLP が 20 tasks を覆盖; 40/40 scored records; すべて direct scores。 | Raw sample files、task definitions、reproducible baselines、task validity を確認。 |
| 128 selected episodes | 96/16/16 split、34,269 exported windows、public-safe features が official gated episode paths に対応。 | Metadata simple/NN、raw-feature simple/NN、Qwen3-Omni v6、Cosmos3-Super、Cosmos3-Nano; 140/140 scored records; 134 direct + 6 compact proxy。 | 同一 split の metadata/raw baselines、Qwen3-Omni diagnostics、Cosmos3 diagnostics を比較; proxy targets は明示。 |
式: 1-episode methods 2 個 x 20 tasks = 40、128-episode methods 7 個 x 20 tasks = 140、公開 matrix 合計は 180/180 scored records。
Method blocks: Line 1 は task-head baselines(Minimal、Neural MLP)。Line 2 は aligned baseline heads(metadata simple/NN、raw-feature simple/NN)、Qwen3-Omni series(Qwen3-Omni v6 LoRA)、Cosmos3 series(Cosmos3-Super Reasoner、Cosmos3-Nano Future Window)に分かれます。Qwen3 v1-v6 は Line 2 内の LoRA/eval lineage で、project evidence lines とは別です。20-task matrix は v6 を使い、v5 は pinned prior release です。Cosmos3-Super Forward-Dynamics LoRA は別の adapter/weights/results artifact として公開され、20-task matrix の method row には含めません。
入口: 2 つの証拠ラインのガイド、2 ラインのデータ、180 件の結果表、2 ライン結果サマリ。
クイックルート
| 目的 | 入口 |
|---|---|
| プロジェクトを素早く理解 | プロジェクト概要, プロジェクト状態 |
| GitHub / Web / HF の入口を選ぶ | 公開エビデンスマップ |
| 20 タスクを見る | 20 タスクガイド, タスク契約データ |
| 結果を比較 | 研究結果まとめ, 180 件の結果表 |
| 1 サンプルを調べる | 1 episode ブラウザ, sample ファイルマップ |
| sample / 128 / HF full dataset を比較 | data explorer analysis, web analysis section |
| 3 つの foundation pipeline を読む | 3 つの foundation pipeline, pipeline 契約データ |
| 再現・監査 | 再現ガイド, 証拠契約 |
構造
- 読み方のルール: metric があるものは 20 tasks、evidence が何を調べるかを説明するものは 4 research directions、training input/output を定義するものは 3 foundation pipelines です。perception、3D memory、language reasoning、action、planning を統合するものは unified embodied model target であり、新しい score axis ではありません。
- データ: 20-frame window が video、audio、depth、pose/SLAM、mocap、IMU、calibration、language annotation を結びます。
- データ探索: 新しい analysis layer は public sample、selected-128 exports、Hugging Face gated full dataset metadata を分けて提示します。
- タスク: 認識、予測、retrieval、reconstruction、order、sync、long-horizon、action-object、sensor bridge など 20 契約。
- 結果: single-episode minimal/NN は 20/20。128-episode 側は metadata、raw feature、Qwen3、Cosmos を証拠タイプ別に分けます。公開 matrix は 180/180 scored records で、174 direct と 6 compact proxy を分離し、proxy targets は明示します。
- 方向: spatial intelligence、human-video world model、vision-language-action に対して、タスク対応と必要証拠を記録しています。長期目標は unified embodied foundation model です。
公開境界
本プロジェクトは派生 artifacts、指標、図、カード、public-safe 要約のみを公開します。Xperience-10M の利用は Ropedia 公式 Hugging Face データカードとアクセス条件に従います。
Published Mirrors
Glossary
Use GLOSSARY.md and docs/data/glossary.json for project terminology:
evidence line, 20-frame window, compact-proxy score, Qwen v1-v6,
Cosmos3-Super, LoRA adapter, HF artifact dataset, and related terms.
Citation
If you use this project, cite the Ropedia Xperience-10M Task Suite by Chaoyue He and also cite the upstream Ropedia Xperience-10M dataset according to its official card.
@misc{he2026ropedia_xperience10m_task_suite,
author = {Chaoyue He},
title = {Ropedia Xperience-10M Task Suite},
year = {2026},
url = {https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite},
note = {Public task and evaluation layer built on the Ropedia Xperience-10M dataset}
}