I Audited AgiBot's New RL Dataset with RDA

Description and use

RDA successfully audited the AGIBOT WORLD 2026 dataset, confirming data integrity and aligning spike detection with human‑takeover events, demonstrating readiness for training.

Problem

Validate that a third‑party audit tool can reliably detect data quality issues in a large, fresh real‑robot reinforcement‑learning dataset without manual tweaks.

Outcome

RDA successfully audited the AGIBOT WORLD 2026 dataset, confirming data integrity and aligning spike detection with human‑takeover events, demonstrating its readiness for training pipelines.

Workflow

• 1. Download AGIBOT WORLD 2026 from Hugging Face (skip videos).
• 2. Install RDA: pip install robot-data-audit rda.
• 3. Run audit: rda audit ./your_dataset_dir rda recommend ./your_dataset_dir --offline.
• 4. Review per‑episode JSON output and overall verdicts.
• 5. Spot‑check REVIEW items if desired; no cleaning required.

Evidence

226 spikes (13.2%) fall within ±1 frame of a takeover transition

Source

4,448 integrity checks (missing frames, invalid values, schema, timestamps), 0 failures

Source