UMI
AssessTechniques
Embodiment-free robot trajectory data used to pre-train manipulation policies.
Why it's here
Placed in Assess: 1 article(s) of evidence from 1 source(s), led by research-stage coverage, with 1 in the last 30 days. Confidence 24%. Low accumulated evidence, so it defaults conservatively pending more signal.
Evidence (1)
- 8Hacker News·7/20/2026researchXiaomi Robotics 1 scales robot policy models with data
Xiaomi-Robotics-1 is a robot policy model trained with large-scale embodiment-free pre-training on 100,000 hours of UMI trajectories, followed by post-training on real-robot data. The authors report that both pre-training and real-world robot success improve steadily as data and model size increase, suggesting predictable scaling behavior for robotics.