I Audited AgiBot's New RL Dataset with RDA

说明与用途

RDA 成功审核了 AGIBOT WORLD 2026 数据集，确认数据完整性并将峰值检测与人工接管事件对齐，证明数据已准备好用于训练。

问题

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.

结果

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.

工作流

• 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.

证据

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

来源

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

来源