---
license: mit
pretty_name: Ropedia Xperience-10M Task Suite Artifacts
tags:
  - embodied-ai
  - multimodal
  - xperience-10m
  - evaluation
  - qwen3-omni
configs:
  - config_name: episode_sample
    data_files:
      - split: public_sample
        path: viewer/episode_windows.parquet
  - config_name: selected_128_windows
    data_files:
      - split: selected_128
        path: viewer/selected128_windows.parquet
---

# Ropedia Xperience-10M Task Suite Artifacts

![Ropedia Xperience-10M Task Suite logo](assets/brand/xperience10m-logo-social-card.png)

This Hugging Face dataset repository is the public-safe artifact layer for the
**Ropedia Xperience-10M Task Suite** by **Chaoyue He**. The task suite is built
on the Xperience-10M egocentric dataset released by Ropedia; this repository
publishes derived files, metrics, viewer tables, figures, manifests, result
summaries, and model-diagnostic packages without redistributing gated raw data.

The project is organized into two evidence lines:

- **Line 1: one public sample episode.** This line supports raw-file inspection,
  20-frame windows, 20 task contracts, Minimal baselines, Neural MLP baselines,
  and 40/40 direct single-episode task scores.
- **Line 2: selected 128 episodes.** This line supports same-split comparison
  across metadata/raw baselines, Qwen3-Omni v6 LoRA, Cosmos3-Super Reasoner, and
  Cosmos3-Nano Future Window diagnostics. It contains 140/140 selected-128
  scores: 134 direct scores and 6 compact-proxy scores.

Across both lines, the public matrix contains **180/180 method-task records**.
Direct scores and compact-proxy estimates are intentionally kept separate.

## What This Dataset Repository Contains

This repository is useful when you want the public evidence files rather than
only the website summary. It contains:

- Hugging Face Dataset Viewer tables for the public sample and selected-128
  windows.
- Machine-readable task contracts, result matrices, proxy audits, quality gates,
  glossary records, public mirror maps, and website integrity checks.
- Public-safe metrics, predictions, confusion matrices, charts, figures, and
  reports for Minimal, Neural MLP, metadata/raw baselines, Qwen3-Omni, and
  Cosmos3 diagnostics.
- Source-linked selected-128 feature/index records that point back to official
  gated Xperience-10M episode paths without republishing the raw episodes.
- Website assets used by the GitHub Pages dashboard and Hugging Face mirrors.

## Dataset Viewer Configs

Use the viewer configs separately. They are different evidence lines and should
not be concatenated for scoring.

- `episode_sample/public_sample`: one public Xperience-10M sample episode,
  exposed as `viewer/episode_windows.parquet` and `viewer/episode_windows.jsonl`.
  It contains 1,161 aligned windows with action/subtask labels, object labels,
  time ranges, modality summaries, feature dimensions, and source-file links.
- `selected_128_windows/selected_128`: selected 128-episode public-safe window
  metadata, exposed as `viewer/selected128_windows.parquet` and
  `viewer/selected128_windows.jsonl`. It contains 34,269 exported windows linked
  to official gated episode paths and selected-128 split metadata.

## Start Here

For the shortest path through the artifact repository:

1. Open `PROJECT_STATUS.md` for the current state.
2. Open `TWO_EVIDENCE_LINES.md` and `docs/data/two_evidence_lines.json` to see
   how Line 1 and Line 2 are separated.
3. Open `TASK_SUITE_20.md` and `docs/data/task_suite_20.json` for the 20 task
   contracts.
4. Open `TASK_METHOD_20_RESULT_MATRIX.md` and
   `docs/data/task_method_20_result_matrix.json` for the 180-record matrix.
5. Open `TASK_METHOD_20_GAP_AUDIT.md` and
   `docs/data/task_method_20_gap_audit.json` before interpreting proxy-marked
   cells.
6. Open `XPERIENCE10M_128_EPISODE_FEATURE_INDEX.md` and
   `docs/data/xperience10m_128_episode_feature_index.json` for selected-128
   source and feature provenance.
7. Open `ARTIFACT_GUIDE.md` if you need the full file map.

## Main Artifact Groups

- `viewer/`: Dataset Viewer tables for public sample windows and selected-128
  windows.
- `docs/data/`: canonical JSON records for task contracts, matrices, audits,
  public mirrors, glossary terms, quality gates, website integrity, and model
  comparison summaries.
- `docs/assets/`: charts, branding, task icons, modality thumbnails, pipeline
  figures, and website graphics.
- `results/episode_task_suite/`: public-sample window tables, feature manifests,
  Minimal baselines, Neural MLP outputs, task walkthroughs, predictions, and
  metric summaries.
- `results/omni_finetune/verified_public/`: public-safe Qwen3-Omni and Cosmos3
  diagnostic packages, metrics, predictions, package audits, and run summaries.
- `results/single_episode_diagnostics/`: additional public-sample diagnostic
  labels, object records, and explorer inputs.

## How To Interpret Scores

- The raw metric value is the value to cite or compare.
- The normalized radar value is a plotting value used only for visual
  comparison across metrics.
- Direct scores use the task target directly available in the relevant evidence
  line.
- Compact-proxy scores use public-safe compact features when the direct raw
  target is unavailable in the selected-128 export. These cells remain marked in
  `docs/data/task_method_20_gap_audit.json` and the result matrix.

## Data And Redistribution Boundary

This artifact dataset does **not** redistribute raw Xperience-10M MP4 files,
raw `annotation.hdf5`, `.rrd` visualization files, gated full-dataset contents,
Qwen3 base weights, Cosmos3 base weights, private logs, or full training
checkpoints. Use of Xperience-10M remains governed by the official Ropedia
Xperience-10M dataset card and access terms.

This repository publishes derived public-safe artifacts only: metrics, feature
summaries, predictions, manifests, figures, JSON records, and selected-source
links.

## Related Repositories

- Project website: https://chaoyue0307.github.io/ropedia-xperience-10m-task-suite/
- GitHub source: https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite
- HF Space: https://huggingface.co/spaces/cy0307/ropedia-xperience-10m-task-suite
- Baseline model mirror: https://huggingface.co/cy0307/ropedia-xperience-10m-task-baselines
- Consolidated weights/results: https://huggingface.co/cy0307/ropedia-xperience-10m-weights-results
- Qwen3-Omni LoRA adapter: https://huggingface.co/cy0307/ropedia-qwen3-omni-lora-128ep
- Cosmos3-Super LoRA adapter: https://huggingface.co/cy0307/ropedia-cosmos3-super-forward-dynamics-lora-128ep
- Upstream gated dataset: https://huggingface.co/datasets/ropedia-ai/xperience-10m
- Public sample dataset: https://huggingface.co/datasets/ropedia-ai/xperience-10m-sample

## Citation

If you use these artifacts, cite this artifact dataset, the main Ropedia
Xperience-10M Task Suite, and the upstream Ropedia Xperience-10M dataset
according to its official card.

```bibtex
@misc{he2026ropedia_xperience10m_task_suite_artifacts,
  author       = {Chaoyue He},
  title        = {Ropedia Xperience-10M Task Suite Artifacts},
  year         = {2026},
  publisher    = {Hugging Face},
  url          = {https://huggingface.co/datasets/cy0307/ropedia-xperience-10m-task-suite-artifacts},
  note         = {Public-safe derived artifacts, viewer tables, metrics, and result records for the Ropedia Xperience-10M Task Suite}
}
```

## Contact

Project author: Chaoyue He. Contact and profile:
https://chaoyue0307.github.io/