Robot Data Audit (RDA)

Quality auditing tool for robot datasets. RDA provides comprehensive metrics for evaluating the integrity, temporal consistency, motion quality, and distribution coverage of robot trajectory datasets.

Features

• Integrity Metrics: Missing frames, NaN values, schema validation
• Temporal Metrics: Timestamp consistency, sensor sync, jitter analysis
• Motion Metrics: Joint limits, velocity profiles, discontinuities, idle detection
• Distribution Metrics: Distribution statistics, coverage analysis
• Three-tier classification: PASS / REVIEW / EXCLUDE
• CLI-first design: Easy to integrate into data pipelines
• Streamlit UI (coming in v0.2.0): Interactive dashboard for exploring results

Installation

From PyPI

Note: PyPI publishing is planned for the v0.2.0 release. Until then, install from source as described below.

# Coming soon — PyPI package name: robot-data-assurance
# pip install robot-data-assurance

From source (development)

git clone <repository-url>
cd robot-data-audit
pip install -e .

With UI support (v0.2.0+)

pip install -e ".[ui]"

Quick Start

1. Audit a dataset

rda audit /path/to/lerobot/dataset

This will run all 12 RDA metrics against every episode in the dataset and print a text summary to the console. A JSON report is automatically saved to <dataset_path>/rda_report.json.

2. Use example datasets

# See examples and usage tips
rda example

3. Customize output

# Save report to a specific path
rda audit /path/to/dataset --output my_report.json

# Output JSON to stdout (for piping)
rda audit /path/to/dataset --format json

# Verbose mode with platform info
rda audit /path/to/dataset --platform so101 -v

4. Preview the UI (coming soon)

rda audit /path/to/dataset --ui

CLI Reference

rda audit

Audit a LeRobot dataset at the given PATH.

rda audit [OPTIONS] PATH

rda example

Show example usage and sample dataset paths.

rda example

Exit Codes

Project Structure

rda/
├── cli/          # Click CLI entry points
├── io/           # Data loading and schema definitions
├── metrics/      # Audit metric implementations (12 metrics total)
├── audit/        # Dataset and episode-level audit orchestration
└── report/       # Report generation and summary

docs/             # API documentation and design specs
examples/         # Example scripts
  ├── basic_audit.py       # Core workflow demo (synthetic data ready)
  └── custom_metrics.py    # How to write custom audit metrics
tests/            # 155 unit tests

Documentation

• API Reference — Full Python API documentation
• MVP Product Spec — Product requirements (v0.2.0)
• Technical Design — Architecture and design decisions
• Project Charter — Mission, goals, and scope
• Roadmap — Release plan and milestones
• Changelog — Version history

Python API Quick Start

from rda.audit.dataset_audit import DatasetAuditor
from rda.io.lerobot_loader import iter_episodes, load_lerobot_dataset
from rda.report import generate_dataset_report

dataset_info = load_lerobot_dataset("/path/to/dataset")
auditor = DatasetAuditor()
result = auditor.audit_dataset(dataset_info, iter_episodes("/path/to/dataset"))

report = generate_dataset_report(result)
print(f"DHI: {report['quality']['dhi']} / 100")

See docs/API.md for the complete API reference, or examples/ for runnable scripts.

Development

Running tests

pytest

Linting

pip install -e ".[dev]"
ruff check rda/

License

MIT