Xiang Xiaomi's representative Kong Tao delivers keynote speech on embodied AI models and applications
小米集团具身模型与应用部负责人孔涛发表主旨演讲
On November 4, 2026, at the World Internet Conference Wuzhen Summit, Kong Tao, head of Xiaomi's Embodied Model and Application Department, delivered a keynote speech outlining Xiaomi's progress in embodied intelligence large models.
The text reports on a speech by Xiaomi's representative at the 2026 World Internet Conference in Wuzhen, detailing their VLA model architecture, training data scale (over 100k hours), and factory deployment results; claims regarding specific success rates and benchmark rankings are attributed to internal tests or industry lists mentioned in the speech.
Source: Chinese robotics — Robot learning research · Read original article ↗
English translation unavailable. Choose the original article to read it.
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
Reported numbers
task_scenarios
>1,700
Reported trials
training_hours
≤10 hours
View original evidence
内部自己测试结果仅需要不超过10小时新任务示范数据，可以整体成功率做到75%以上，也就是说拿到一个新任务，来到新场景，只需要采集员采集一天数据，可以任务做到还不错或及格线以上水平。
Open source S21Reported success rate
>75%
View original evidence
内部自己测试结果仅需要不超过10小时新任务示范数据，可以整体成功率做到75%以上，也就是说拿到一个新任务，来到新场景，只需要采集员采集一天数据，可以任务做到还不错或及格线以上水平。
Open source S21
- paper: not_reported
What remains unknown
Not established in the collected evidence: Environment, Control, Data origin.
Reported performance applies to the described task. It does not establish general autonomy or deployment readiness.
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
更进一步有了这样一个训练架构，今年7月份我们做了第一个国内超过十万小时的预训练数据。
Open source S15
基于十万小时训练规模，这个场景超过1700种，横跨家庭的、工业的、商业的真实环境。
Open source S16
内部自己测试结果仅需要不超过10小时新任务示范数据，可以整体成功率做到75%以上，也就是说拿到一个新任务，来到新场景，只需要采集员采集一天数据，可以任务做到还不错或及格线以上水平。
Open source S21
当然有了这么多数据后，我们需要有一个非常高效，高质量的数据标注和处理管线，大概在两个月之前需要两周时间，现在大概几天时间就可以完成。
Open source S17
我们在前期验证，虽然只有十万小时，这个规模下也验证了它的一些初步能力。
Open source S20
Source：Chinese robotics — Robot learning research · news.qq.com