Ropedia Episode Task Suite

Pipeline from raw episode to task suite.

Every script works from the same data contract: aligned multimodal windows, explicit labels, cached feature extraction, and artifact-first evaluation.

Verified Ropedia all-modality pipeline diagram

What this project proves

It proves the full engineering loop: reading Ropedia 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.

Baselines with transparent artifacts.

Motion-only and all-modality classifiers use the same lightweight softmax model so the comparison stays readable on a laptop.

Motion-only action

0.9688macro-F1, 18 classes

All-modality action

0.9791macro-F1, 8,378 features

Motion-only subtask

0.9528macro-F1, 14 classes

All-modality subtask

0.9308macro-F1, chronological caveats

The 12 tasks use three minimal head families.

The diagram separates the shared episode-window feature pipeline from the task-specific heads: linear softmax, ridge projection/regression, and multi-label logistic regression.

Verified minimal architecture diagram for all 12 Ropedia episode tasks

Single-episode tasks, deliberately scoped.

These tasks turn one episode into a lab for temporal structure, modality alignment, retrieval, reconstruction, and action understanding.

ChatGPT-image-backed infographic showing all 12 Ropedia episode tasks

timeline_actionsupervised

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

macro-F10.0500

timeline_subtasksupervised

All 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 modalities.

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

All-modality features are inspectable.

The point is not hidden complexity. Every block has a clear source modality and an explicit dimensional footprint.

Diagnostics separate easy wins from real gaps.

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.

Everything important is checked in.

Metrics, predictions, confusion matrices, feature manifests, model weights, and derived window artifacts are committed so the repo is reviewable without rerunning first.

Task-suite report

One JSON file with every task metric and split detail.

summary_report.json

Feature manifest

Start/end index and dimension for every all-modality feature block.

feature_manifest.json

Cross-modal retrieval

The strongest self-supervised signal from the single episode.

metrics.json

All-modality 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/

Reproduce the suite.

Raw Ropedia 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

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

git clone https://github.com/ChaoYue0307/ropedia-episode-task-suite.git
cd ropedia-episode-task-suite
python scripts/episode_task_suite.py --workspace /path/to/workspace