README.md

---
title: Trajectory Endpoint Labeler
emoji: 🎯
colorFrom: blue
colorTo: purple
sdk: gradio
sdk_version: 6.1.0
python_version: 3.12
app_file: app.py
pinned: false
short_description: Label trajectory success cutoffs for Robometer data
---

# Trajectory Endpoint Labeler

Gradio tool for labeling where robot trajectories reach task success — used to derive per-dataset **success cutoff** percentages for RewardFM / Robometer training.

## Features

- Load trajectories from any HuggingFace dataset
- Filter by success / failure / all
- Frame-precise end-point marking with percent-of-trajectory
- Pattern analysis across labeled trajectories (mean, std, suggested cutoff)
- CSV export (`labels.csv`)

## Usage

1. Enter HF dataset repo (e.g. `jesbu1/epic_rfm`)
2. Set sample counts and quality filter
3. Scrub to the success frame → save label
4. After labeling a batch, run **Analyze Pattern** for suggested cutoff %

## Output

```csv
dataset_repo,config_name,trajectory_id,is_robot,quality_label,task,manual_end_frame,manual_end_percent,notes
```

Suggested cutoffs feed into `dataset_success_cutoff.txt` for training.

---

Part of [Robometer](https://huggingface.co/robometer/spaces) · Built with Gradio + HuggingFace Datasets