Ropedia Xperience-10M Task Suite

Evidence first, claims second.

A top-level project should make its proof boundary visible. This ledger separates verified single-episode artifacts from smoke-only Qwen3-Omni work and the pending 32-episode gate.

verified

Aligned Xperience-10M sample windows

5,821 frames become 1,161 synchronized 20-frame windows with an explicit 8,378-d feature contract.

summary_report.json feature_manifest.json

verified

12 minimal heads + 12 neural MLP heads

Every task has a minimal interpretable head and a matching neural MLP run over the same windows, splits, and task contract.

task artifacts neural_mlp/

verified

Four research directions are mapped honestly

The Ropedia directions are labeled as direct, proxy, or diagnostic evidence, plus one coded extension probe per direction.

directions JSON extensions JSON

blocked by data access

Qwen3-Omni is smoke-only until 32 episodes land

The H20/A100 pipeline exists, but current Qwen3-Omni artifacts use one episode and 128 train windows. No 32-episode metric is claimed.

evidence contract machine JSON

verified

Publication bundles pass hygiene checks

The validator checks required assets, raw-data exclusion, Python cache exclusion, heavy archive exclusion, and accidental HF token strings across GitHub and the HF bundles.

publication audit artifact index validator script

Verify the project in 90 seconds.

World-class research packaging should be easy to audit. This path tells a reviewer exactly what to open first, what each artifact proves, and where the current proof boundary stops.

Check the claim boundary

Start with the evidence contract, artifact index, and publication audit. They separate verified single-episode artifacts from smoke-only Qwen3-Omni work.

contract JSON index audit

Inspect one model input

Use the window table and feature manifest to see the exact aligned sample unit, feature blocks, dimensions, and omitted audio feature status.

windows features

Compare minimal vs neural heads

Every task has a small interpretable baseline and a matching neural MLP head over the same feature contract and chronological split.

summary metrics neural heads

Check the scale-up gate

The multi-episode Qwen3-Omni path is prepared, but no real 32-episode result is claimed until the data gate and held-out evaluation pass.

blocker relay review packet

Verified nowOne public episode, 5,821 frames, 1,161 windows, 8,378 current features, 12 minimal heads, 12 neural heads, and 4 direction-extension probes.

Pending scaleA 32-episode held-out Qwen3-Omni LoRA pilot is gated on Xperience-10M access and must pass manifest, training, and evaluation checks.

Not redistributedRaw videos, raw annotations, full Qwen weights, and private gated Xperience-10M data are not included in the public repo or HF bundles.

Ropedia Xperience-10M 12-task suite, first.

The top-level map shows the Xperience-10M sample modalities and the verified 12-task results. Audio is present in the sample MP4 stream, but the current 8,378-d baseline manifest does not featurize it.

ChatGPT-image-backed infographic showing all 12 Ropedia Xperience-10M tasks

From raw episode to checked artifacts.

Every script works from one data contract: aligned multimodal windows, explicit labels, cached feature extraction, and a manifest that makes omitted modalities visible.

Verified Xperience-10M multimodal pipeline diagram

What this project proves

It proves the full engineering loop: reading Xperience-10M sample data, aligning modalities, converting them into model-ready windows, defining meaningful tasks, producing metrics, and packaging every artifact for review.

What it does not prove

It does not claim general embodied intelligence. A single episode cannot support cross-environment generalization; that requires many episodes and held-out episode splits.

Small baselines, no hidden machinery.

Motion-only and current all-feature classifiers use lightweight heads so the comparison stays readable on a laptop and easy to audit. The neural run keeps the same features and splits, then swaps in PyTorch MLP heads.

Motion-only action

0.9688macro-F1, 18 classes

Current all-feature action

0.9791macro-F1, 8,378 features

Motion-only subtask

0.9528macro-F1, 14 classes

Current all-feature subtask

0.9308macro-F1, chronological caveats

Neural MLP heads, same task contracts.

The neural baseline uses small PyTorch MLP classifiers/regressors on the same 8,378-d window features, chronological splits, and leakage filters. This isolates the value of a nonlinear head before moving to heavier Qwen/Omni experiments.

Neural hand forecast

0.1116MPJPE, down from 0.8223 minimal

Neural temporal order

0.8718F1, adjacent-window diagnostic

Neural misalignment

0.7335F1, shifted motion/visual pairs

Neural cross-modal retrieval

0.1530MRR; ridge remains stronger here

Minimal versus neural MLP episode task score chart

The 12 tasks organized into four research directions.

Each task is mapped as direct, proxy, or diagnostic evidence for the Ropedia research tracks. The mapping uses two current baselines: minimal interpretable heads and neural MLP heads over the same feature contract.

partially implemented

A. Human Modeling & Motion Understanding

Direct evidence comes from hand trajectory forecasting and contact prediction; action and object relevance are supporting proxies.

2direct2proxy0diagnostic

proxy tasks only

B. 3D/4D Reconstruction & Neural Rendering

Cross-modal retrieval, modality reconstruction, and misalignment detection check reconstruction prerequisites, not full geometry.

0direct2proxy1diagnostic

strongest implemented

C. Egocentric Vision & Interaction

Action, subtask, transition, next-action, object, caption, order, and alignment tasks directly stress egocentric understanding.

6direct2proxy3diagnostic

early proxy tasks

D. Scene Reconstruction & World Modeling

Current probes cover task state, object relevance, retrieval, reconstruction, temporal order, and alignment but no persistent map yet.

0direct6proxy3diagnostic

Coverage of the 12 Xperience-10M tasks across four research directions

Baseline 1: minimal heads

Softmax, logistic, ridge, and retrieval heads keep every input/output contract readable. They are the first sanity check for whether a task is well-posed.

Baseline 2: neural MLP heads

Small PyTorch MLP classifiers/regressors reuse the same features and splits. They test nonlinear gains before heavier Omni fine-tuning.

Four extra probes make the directions actionable.

These are new data-backed extension tasks computed from the same single-episode feature tensor. They add one concrete input, process, output, and metric for each research direction, while keeping the single-episode limitation explicit.

Four Xperience-10M research-direction extension probes with minimal and neural metrics

A / motion

body_motion_intensity

Case: classify fast reach/pour windows as high motion and steady holding windows as low motion.

Input: non-mocap video, depth, pose, IMU, SLAM, calibration, and language features.

Output: high_motion or low_motion.

0.7827minimal macro-F10.7986neural macro-F1

B / views

multi_view_consistency_retrieval

Case: retrieve the synchronized stereo-left window from a fisheye-camera query.

Input: fisheye_cam0 video features against stereo_left candidate features.

Output: ranked synchronized view candidates.

0.5534minimal MRR0.3469neural MRR

C / phase

action_phase_progress

Case: estimate whether a Pour coffee window is near the start, middle, or end of its action segment.

Input: non-caption multimodal features.

Output: 0-to-1 progress inside the current action.

0.3416minimal MAE0.3038neural MAE

D / world

ego_motion_forecast

Case: predict how the camera translation changes over the next 20 frames.

Input: current sensors excluding camera translation and captions.

Output: future camera-translation delta vector.

0.1989minimal MAE0.0989neural MAE

What changed

The four research directions now have coded extension probes, prediction/rank CSVs, JSON metrics, a Markdown summary, and a website chart generated from real sample-window features.

What still needs scale

A full research result still needs many Xperience-10M episodes, held-out episode splits, stronger encoders, and direction-specific models such as body priors, renderers, or persistent scene graphs.

The 12 tasks share four head families.

The diagram separates the shared episode-window feature pipeline from the task-specific heads, and notes that audio remains dataset context rather than a current baseline feature block.

Verified minimal and neural architecture diagram for all 12 Ropedia Xperience-10M tasks

Junior-friendly task walkthroughs.

Each task is explained as a case study with input, middle process modules, and output. The full generated guide is committed as Markdown and JSON.

timeline_action

Case: a pouring window should map to the current action, such as Pour coffee.

Inputone 20-frame all-modality windowProcesswindow builder, feature assembler, action label builder, classifier, evaluatorOutputcurrent action class

timeline_subtask

Case: a fine action is grouped into a broader drink-preparation stage.

Inputone all-modality windowProcesswindow builder, feature assembler, subtask label builder, classifier, evaluatorOutputcurrent subtask label

transition_detection

Case: detect the moment the demonstrator changes from preparing to pouring.

Inputone window plus boundary labelsProcessboundary builder, feature assembler, binary classifier, boundary matcher, evaluatorOutputboundary or steady

next_action

Case: a preparing-to-pour window should predict the action 20 frames later.

Inputcurrent window at time tProcessfuture label shift, current-only features, classifier, evaluatorOutputaction at t+20 frames

hand_trajectory_forecast

Case: when a hand moves toward a cup, predict the future 3D hand path.

Inputcurrent all-modality windowProcessfuture mocap target builder, regression head, trajectory reshaper, MPJPE evaluatorOutputfuture left/right hand joints

contact_prediction

Case: decide whether the hand/body is touching an object or surface.

Inputnon-contact, non-caption featuresProcessfeature selector, contact target builder, binary classifier, degeneracy checkOutputcontact or no contact

object_relevance

Case: during pouring, infer relevant objects such as milk, cup, or coffee.

Inputnon-caption featuresProcessobject vocabulary, multi-hot targets, sigmoid heads, micro-F1 evaluatorOutputmulti-label object set

caption_grounding

Case: a query like Pour milk into coffee should retrieve the matching moment.

Inputcaption query plus candidate windowsProcessquery builder, candidate builder, projection head, cosine ranker, MRR evaluatorOutputranked matching windows

cross_modal_retrieval

Case: motion and IMU from pouring should retrieve the matching depth/video window.

Inputmotion/IMU/camera query, depth/video candidatesProcessmodality splitter, projection head, candidate index, cosine ranker, top-k evaluatorOutputranked depth/video windows

modality_reconstruction

Case: infer compressed depth/video features from motion, IMU, and camera pose.

Inputmotion/IMU/camera featuresProcesssource-target split, scaling, regression head, inverse scaling, R2 evaluatorOutputdepth/video feature vector

temporal_order

Case: tell whether reaching then pouring has been reversed.

Inputtwo adjacent windowsProcesspair builder, feature combiner, binary classifier, F1 evaluatorOutputcorrect or reversed order

misalignment_detection

Case: detect motion from pouring paired with video/depth shifted later.

Inputmotion side plus visual/depth sideProcessaligned pair builder, shifted pair builder, feature combiner, binary classifierOutputaligned or shifted

Task cards and metrics.

The same 12 tasks are kept filterable here so the supervised, forecast, retrieval, and diagnostic probes can be inspected individually.

timeline_actionsupervised

All featurized modalities to current action label. Chronological split exposes unseen future actions.

macro-F10.0500

timeline_subtasksupervised

All featurized modalities to current subtask label. Useful for segmentation diagnostics.

macro-F10.0495

transition_detectiondiagnostic

Predict steady vs action boundary. Highlights task-transition localization quality.

macro-F10.6552

next_actionsupervised

Current multimodal window to action 20 frames later. Tests short-horizon task flow.

macro-F10.0593

hand_trajectory_forecastforecast

Predict future left/right hand 3D joints. Closer to imitation-learning style signals.

MPJPE0.8223

contact_predictionsupervised

Non-contact modalities to binary contact. Degenerate in this sample because one class dominates.

accuracy1.0000

object_relevancesupervised

Predict relevant object set from non-caption feature blocks.

micro-F10.1839

caption_groundingretrieval

Caption objects/interaction query to matching sensor window.

MRR0.0172

cross_modal_retrievalretrieval

Motion/IMU/camera query retrieves matching depth/video window. Strongest single-episode signal.

top-50.3764

modality_reconstructionforecast

Motion/IMU/camera to depth/video feature vector.

R2-0.0160

temporal_orderdiagnostic

Two adjacent windows to correct vs reversed order.

F10.5487

misalignment_detectiondiagnostic

Motion+visual pair to aligned vs shifted by eight windows.

F10.4866

Every feature block has a source.

The point is not hidden complexity. Every block has a source modality, a dimensional footprint, and a manifest entry.

Diagnostics separate memorization from signal.

The charts make the main lesson visible: within-episode supervised labels are easy under some splits, while retrieval, grounding, forecasting, and alignment remain the useful probes.

Where the evidence lives.

Metrics, predictions, confusion matrices, manifests, lightweight model weights, and derived window artifacts are committed so the repo is reviewable before rerunning anything. Raw Xperience-10M data and Qwen weights are not redistributed.

Artifact index

Selective source-of-truth catalog with existence checks, sizes, and stable-file hashes for reviewer-critical artifacts.

artifact_index.json

Task-suite report

One JSON file with every task metric and split detail.

summary_report.json

Neural MLP task results

Per-task PyTorch MLP metrics, predictions, histories, and checkpoints for the same 12 task contracts.

neural_mlp/

Four-direction taxonomy

Generated JSON, CSV, Markdown, and website data mapping all 12 tasks to the four research tracks with two baselines.

research_directions/

Direction extension probes

Four additional coded probes, one per research direction, each with minimal and neural metrics plus prediction/rank CSVs.

research_direction_extensions/

Task walkthroughs

Beginner-oriented case studies for all 12 tasks, including input, middle process modules, output, metric, and limitation.

task_walkthroughs/

Feature manifest

Start/end index and dimension for every current feature block.

feature_manifest.json

Cross-modal retrieval

The strongest self-supervised signal from the single episode.

metrics.json

Current all-feature action model

Classifier metrics, predictions, confusion matrix, and model weights.

metrics.json

Windows table

Window start/end frames and aligned action/subtask labels.

windows.csv

Reproduction scripts

Three training scripts plus the dashboard generator.

scripts/

Hugging Face Space

The same dashboard packaged as a public static Space.

cy0307/ropedia-xperience-10m-task-suite

Derived HF artifacts

Metrics, predictions, docs, and lightweight derived files without raw Xperience-10M video/data redistribution.

dataset repo

HF baseline models

Minimal NumPy softmax and ridge baseline weights with model card and architecture diagrams.

model repo

A100 HF relay status

HF full-dataset access is pending; an A100 watcher is ready to download a 32-session stratified pilot and transfer it to H20 once approved.

A100_HF_RELAY_STATUS.md

Qwen3-Omni readiness artifacts

Manifests, metadata, metrics, and progress logs from the current smoke/evidence run. No real 32-episode metric is claimed yet.

episode_manifest.json

32-episode data gate

The readiness gate remains the source of truth before any full pilot training claim.

DATA_BLOCKER_REPORT.md

HF collection

Space, artifacts, and model baselines grouped into one public project collection.

collection

Qwen3-Omni pilot is approval-ready.

The full Xperience-10M Hugging Face dataset is gated. While access is pending, the A100 relay has already selected a 32-episode pilot across 32 different session UUIDs and will continue automatically after approval.

Selection

Stratified round-robin over 64 top-level sessions; 680 complete candidates scanned; 32 sessions selected.

Transfer

A100 downloads from Hugging Face, excludes visualization.rrd, validates files, then rsyncs to H20.

Boundary

The current LoRA artifact is a smoke/pilot checkpoint. A real 32-episode result requires the watcher to finish and held-out evaluation to run.

Reproduce the suite.

Raw Xperience-10M data is not redistributed here. Download the public sample separately, then run the same scripts.

git clone https://github.com/Ropedia/HOMIE-toolkit.git
python3.12 -m venv .venv
source .venv/bin/activate
pip install -r HOMIE-toolkit/requirements.txt huggingface_hub hf_xet
pip install torch

hf download ropedia-ai/xperience-10m-sample \
  --repo-type dataset \
  --local-dir data/sample/xperience-10m-sample

git clone https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite.git
cd ropedia-xperience-10m-task-suite
python scripts/episode_task_suite.py --workspace /path/to/workspace --include-neural
python scripts/research_direction_extension_tasks.py
python scripts/generate_visualizations.py