# Realman provides real-world data for Ant Lingbo's open-source embodied model LingBot-VLA

- Source: RealMan/睿尔曼
- Published: 2026-02-25T11:00:00.000Z
- RoboHot: https://robohot.news/en/items/rhwlmoc8jpp86idh9ei6ain01
- Original: https://mp.weixin.qq.com/s/raw01ut_O7YDlRh1PxUVTA

## Summary

Ant Lingbo has released and open-sourced LingBot-VLA, an embodied intelligence foundation model. Its pretraining used about 20,000 hours of real-world data covering 9 dual-arm robot configurations, and it outperformed Pi0.5 overall in real-robot evaluations across more than 100 tasks. Realman's wholly owned subsidiary Ruiyuan Robotics provides high-precision multimodal real-world data for the RS-01 and RS-02 robots. The technical report shows that model success rates continued to improve without saturating as pretraining data expanded from 3,000 hours to 20,000 hours.

## Why it matters

From a partner's perspective, it explains how real-world data was specifically supplied for LingBot-VLA training and the capability for large-scale data collection, helping readers understand how the data infrastructure actually operates.