Embodied Data Crisis: Efficiency Drops to 10% After Cleaning

创业邦 · wechat · 2026-08-27

The embodied intelligence industry faces a data quality crisis, with one company seeing data efficiency drop to just 10% after cleaning. The article points out that the UMI acquisition route relying solely on first-person visual data has blind spots in physical interaction information. High-quality embodied data must meet three conditions: high fidelity, multimodality, and diversity. Additionally, scaling data faces challenges in device stability, transmission, storage, and governance costs, particularly the difficulty of reducing governance costs. Future value for data companies will shift from selling raw data to providing effective "skill tokens" and complete data infrastructure services, with software-hardware closed-loop innovation being key.

Original post →

More from Embodied

Embodied channel →