# Why embodied AI data volume is not turning into better robot models

> Original title: 具身的数据管线，到底应该怎么做？

- Source: Leiphone
- Published: 2026-10-30T07:11:00.000Z
- Bad Influence: https://badinfluence.news/items/sy0b2ai2oqyz6lao6yx9id6hi
- Original: https://www.leiphone.com/category/robot/espiC5FlGhRbSj1b.html

## Summary

Leiphone argues that embodied AI companies are collecting more robot training data, but usable data and training gains have not kept pace. The article breaks the data pipeline into five stages (filtering, annotation, action conversion, verification, and feedback iteration) and identifies persistent problems in quality standards and cross-embodiment transfer.

## Why it matters

The article breaks down the embodied AI data pipeline stage by stage and names its bottlenecks, which helps readers see why more collected data does not translate into better models.